POLITECNICO DI MILANO

Scuola di Ingegneria Industriale e dell’Informazione

Master of Science in
Management Engineering

Industrial IoT: a Cost-Benefit Analalyis of

Predictive Maintenance Service

Supervisors: Prof. Angela TUMINO

Co-supervisor: Dr. Giulio SALVADORI

Dr. Elisa VANNINI

Master Thesis by:

Luca FRANCESCHINI Matr. 920497
Alberto MIDALI Matr. 921227

Academic Year 2019 - 2020

Contents
1 Manufacturing ..................................................................................................... 1

1.1 Introduction to Manufacturing ................................................................ 2

1.2 Smart Technologies ................................................................................... 5

Cloud Manufacturing .......................................................................... 5

Industrial Internet of Things ............................................................... 6

Industrial Analytics .............................................................................. 7

Advanced Human-Machine Interface ............................................... 8

Advanced Automation ........................................................................ 9

Additive Manufacturing ...................................................................... 9

1.3 Predictive Maintenance .......................................................................... 10

2 Industrial Internet of Things............................................................................ 13

2.1 Definition .................................................................................................. 14

Architecture ......................................................................................... 17

Connectivity ........................................................................................ 18

2.2 Taxonomy of Revolutions ...................................................................... 20

Industrial Revolution ......................................................................... 21

Internet Revolution ............................................................................ 22

Industrial IoT ...................................................................................... 24

Sizing the Industrial Benefits ............................................................ 30

2.3 Catalysts, Precursors and Risks of the Industrial IoT ......................... 34

Technical Dimension .......................................................................... 34

Economic and Social Dimension ...................................................... 36

Major Challenges and Risks .............................................................. 38

iv Contents

3 Industrial IoT Research .................................................................................... 41

3.1 Survey Analysis ...................................................................................... 42

Knowledge .......................................................................................... 43

Completed/Underway Projects ........................................................ 44

Objectives ............................................................................................ 45

Barriers ................................................................................................ 48

Investment Dimension and Future Orientation ............................. 50

3.2 Case study – Fabio Perini S.p.A. (Körber Tissue) ............................... 51

3.3 Remarks and Checkpoint....................................................................... 53

4 Literature Review ............................................................................................. 55

4.1 Introduction to Literature Review ........................................................ 56

4.2 Material Collection and Selection ......................................................... 57

Articles Collection .............................................................................. 57

Article Selection.................................................................................. 58

4.3 Descriptive Analysis ............................................................................... 59

4.4 Category Selection .................................................................................. 63

4.5 Evaluation and Discussion of the Review ........................................... 64

Authors’ Approach ............................................................................ 65

Benefits ................................................................................................ 69

Barriers and Risks .............................................................................. 71

Servitization ........................................................................................ 73

4.6 Results and Future Direction from the Literature Review ................ 74

5 Methodology ..................................................................................................... 82

5.1 Research Questions and Methodology ................................................ 83

6 Model Design .................................................................................................... 90

Contents v

6.1 Model Introduction ................................................................................. 91

6.2 Hypotheses ............................................................................................... 92

The As-Is and To-Be Overview ......................................................... 92

Type of Intervention and its Characteristics ................................... 95

6.3 Model Principles ...................................................................................... 97

Maintenance Strategy Comparison .................................................. 98

Dynamic Payment Method ............................................................. 101

6.4 Flowchart of Model Functioning ......................................................... 103

6.5 Plant Characteristics.............................................................................. 104

Downtime Costs ............................................................................... 105

Spare Parts ......................................................................................... 110

Penalties ............................................................................................. 111

Additional Costs ............................................................................... 112

6.6 Failure Rate ............................................................................................ 113

6.7 Initial Output ......................................................................................... 117

Corrective Maintenance Charge ..................................................... 118

Preventive Maintenance Charge..................................................... 119

6.8 Industrial IoT Solutions ........................................................................ 122

The Initial Investment ...................................................................... 122

Remote Monitoring .......................................................................... 123

6.9 Final output ............................................................................................ 125

7 Model Application .......................................................................................... 130

7.1 Simplifications Adopted ....................................................................... 131

7.2 Concept Scenario ................................................................................... 133

Layout ................................................................................................ 133

vi Contents

Definition of the Maintenance Strategy ........................................ 135

Benefits Evaluation .......................................................................... 143

7.3 Real Cases Application: ConBio.......................................................... 147

About the Company ........................................................................ 147

Layout................................................................................................ 147

The Flow on the Machine................................................................ 148

Maintenance Strategy – As-Is Situation ........................................ 149

Benefits Evaluation .......................................................................... 152

8 Conclusions and Discussions ........................................................................ 155

Appendix A ............................................................................................................... 163

Appendix B ............................................................................................................... 173

Bibliography.............................................................................................................. 181
List of Figures
Figure 1.1: Trend of academic papers per year .........................................................11
Figure 2.1: The adoption and impact path of the Industrial Internet........................26
Figure 2.2: How likely are the following risks or negative consequences associated
with the Industrial Internet? .....................................................................................40
Figure 3.1: Results of Q1.1 - Have you ever heard of the Internet of Things (IoT)
solutions for Industry 4.0? Sample size LEs, 100. Sample size LEs, 525 ...................43
Figure 3.2: Results of Q2.2 related to Smart Factory ................................................45
Figure 3.3: Results of Q3.1 - What were the main objectives that led the company to
launch IoT projects for Industry 4.0? Up to three options can be entered. Sample
size LEs, 61. Sample SMEs, 47 ...................................................................................47
Figure 3.4: Results of Q4.1 - What are the barriers (internal and external) that in
your opinion, can slow down or prevent the start of IoT projects for Industry 4.0? Up
to three options can be entered. Sample size LEs, 88. Sample SMEs, 525 ...............49
Figure 3.5: Results of Q 5.1 .......................................................................................51
Figure 4.1: Paper distribution per year .....................................................................62
Figure 6.1: Flowchart of model functioning ........................................................... 103
Figure 6.2: Examples of health index time profiles. ............................................... 114
Figure 6.3: Examples of three-failure rate pattern................................................. 116
Figure 6.4: Plot of the function combining variable S to B. .................................... 120
Figure 6.5: Confusion Matrix .................................................................................. 126
Figure 6.6: Perfect versus realistic prediction ........................................................ 127
Figure 7.1: Maintenance explanation schema ....................................................... 131
Figure 7.2: Production cycle in the concept scenario ............................................. 134
Figure 7.3: Layout of ConBio .................................................................................. 148
Figure 7.4: Decanter ............................................................................................... 149
Figure 7.5: An example of the data insert in the model from Cluster 1 ................. 151
Figure 8.1: The number of interventions is shown on the y-axis. The colours blue and
orange represent when the intervention is unscheduled or scheduled respectively.
............................................................................................................................... 157
List of Tables
Table 4.1: Research method summary ..................................................................... 60
Table 4.2: Analytical categories and their definitions .............................................. 64
Table 6.1: Example of how failure rate data for machinery components can be
obtained from field statistics starting by the total number of the registered failure
................................................................................................................................ 117
Table 7.1: Variables of the downtime cost – Cluster 1 ........................................... 135
Table 7.2: Variables of the downtime cost– Cluster 2 ............................................ 140
Table 7.3: Cost of the initial solution and savings passing to a predictive
maintenance solution. ............................................................................................ 152
Table 7.4: The final output of the model for the ConBio example .......................... 152
Abstract

The Industrial Internet of Things (Industrial IoT) is revolutionising the way of

doing business in all industries connecting the industrial machinery with the

surroundings. In this scenario, predictive maintenance seems to be one of the most

promising uses. Indeed, the Industrial IoT allows a continuous stream of a massive

amount of real-time data that are analysed and made available to all users, solving the

problems related to this maintenance practice and improving its effectiveness. Thanks

to this level of technological development, it is possible to deliver the maintenance that

is necessary rather than possibly required. However, the spread of predictive

maintenance is slowed by several factors, including a poor understanding of its

benefits. In this regard, we developed a model to quantify the savings that arise,

switching from a traditional maintenance approach (corrective and preventive) to a

predictive maintenance solution with the help of the Industrial IoT. In particular, this

thesis focus on a precise economic study about the feasibility of investment in

Industrial IoT to implement this type of maintenance with a pay-per-performance

agreement between the vendor, or a general OEM, and the manufacturer.

A sensitivity analysis was performed on the variable measuring the effectiveness

of the prediction. A real case, together with a theoretical case, is performed too. The

result is a model that can help both manufacturers to evaluate the NPV of the

investment, and vendors to understand how to capture part of the benefits they

generate for their customers.

Key words: Industrial IoT, Predictive Maintenance, Cost-Benefit Analysis, Pay-

per-Performance.
Abstract

L’Industrial Internet of Things (Industrial IoT) sta rivoluzionando il modo di fare

business in tutti i settori connettendo i macchinari industriali con l'ambiente

circostante. In questo scenario, la manutenzione predittiva sembra essere uno degli usi

più promettenti. Infatti, l'Industrial IoT permette un flusso continuo di un'enorme

quantità di dati in tempo reale che vengono analizzati e resi disponibili a tutti gli

utenti, risolvendo i problemi legati a questa pratica di manutenzione e migliorandone

l'efficacia. Grazie a questo livello di sviluppo tecnologico, è possibile effettuare la

manutenzione realmente necessaria. Tuttavia, la diffusione della manutenzione

predittiva è rallentata da diversi fattori, tra cui una scarsa comprensione dei suoi

benefici. A questo proposito, abbiamo sviluppato un modello per quantificare i

risparmi che ne derivano, passando da un approccio di manutenzione tradizionale

(correttiva e preventiva) ad una soluzione di manutenzione predittiva con l'aiuto

dell'Industrial IoT. In particolare, questa tesi si concentra su un preciso studio

economico sulla fattibilità di un investimento nell'Industrial IoT per implementare

questo tipo di manutenzione con un accordo pay-per-performance tra il venditore, o

un OEM generale, e il produttore.

È stata effettuata un'analisi di sensibilità sulla variabile che misura l'efficacia

della previsione. È stato anche eseguito un caso reale, insieme ad un caso teorico. Il

risultato è un modello che può aiutare sia i produttori a valutare il VAN

dell'investimento, sia i venditori a capire come catturare parte dei benefici che

generano per i loro clienti.

Parole chiave: Industrial IoT, Manutenzione Predittiva, Analisi Costi-Benefici,

Pay-per-Performance.
Executive Summary

The Internet of Things (IoT) has the potential to transform the way we interact

with our surroundings by making it smarter and more responsive. Thanks to the IoT,

every physical object acquires its counterpart in the digital world, and thus allows

monitoring and managing physical objects electronically. The merger of the two

universes enables taking data-driven decision making to unprecedented levels - to

optimise the performance of systems and processes, save time for people and

companies and improve product quality. The data collected by the sensors in the

factories, as well as the data relating to the status of the goods along the entire supply

chain, can help companies obtain much more from their physical resources, improving

the performance of the machines, extending their life and learning how they could be

redesigned to do even more (McKinsey&Company, 2015).

As the Internet revolution has redefined business-to-consumer (B2C) industries

such as media, retail and financial services, the IoT has the potential to subvert the

manufacturing, energy, transportation and other industrial sectors that together

account for about a third of the Gross Domestic Product (World Bank, 2018). The IoT

applied to the industrial world is dubbed Industrial Internet of Things (Industrial IoT).

It is defined as a system that includes intelligent objects, cyber-physical resources, and

platforms that allow access, collection, analysis, real-time and autonomous exchange

of information within the industrial environment. The collective belief is that this latest

wave of technological change will bring unprecedented opportunities and new risks

to businesses and society. However, although the Industrial IoT is starting to spread

into the global industrial fabric, it is still in its early stage. The surveys conducted in

collaboration with the ‘Osservatorio Internet of Things’ of the Politecnico di Milano

confirm this aspect. Indeed, a study carried out on a global base, reports that most of

the Industrial IoT projects started after 2015. At the Italian level, the Industrial IoT is

even less widespread. A 2019 survey on large enterprises shows that almost the entire

number of projects reported started after 2017 and underlines a limited knowledge of
xvi Executive Summary

the topic by the managers of the companies interviewed. If, on the one hand, it is a

common opinion that the Industrial IoT is truly transformative, on the other, by

studying the Italian and world industrial sector carefully, we realise that many

companies are still struggling to understand the implications of the Industrial IoT on

their businesses and industries. For these organisations, the risks of moving too slowly

are real.

Inside the factories, IoT systems can make production processes more efficient,

guaranteeing hardware-software integration and supporting traditional production

models with the vital support of objects with their own "intelligence". Industrial IoT

also represents one of the six technologies behind the so-called Industry 4.0. According

to this principle, digital technologies - IoT devices, but also sensors, cloud, machine

learning, collaborative robotics, 3D printing - would be able to increase the efficiency

and value of production by stimulating interconnection and cooperation between all

resources, internal and external (Miragliotta G. , 2020).

In the world of Industrial IoT, there are dozens of new possibilities that are

emerging. Predictive maintenance is expected to be one of the most promising uses of

the Industrial IoT (Schallehn, Schorling, Bowen, & Straehle, 2019), and companies

evaluate it as the area of major interest. The surveys carried out at the Italian level

confirm this aspect, showing that 14% of the projects activated in recent years in the

smart factory field regards predictive maintenance and 18% fall under the broader

preventive maintenance hat. It is for this reason that we decided to focus our thesis on

this area of Industrial IoT. It should be specified that predictive maintenance, i.e., the

ability to collect and process data to obtain forecasts regarding the breakdown of parts

of the machinery, is not a topic introduced with the Industrial IoT. However, it should

be stressed that IoT technology makes this maintenance more effective and efficient,

and it is for this reason that it has paid great attention in recent years from both the

scientific and the business world.

Executive Summary xvii

The purpose of the research study

Among the many areas of analysis that can be pursued concerning predictive

maintenance, this thesis focuses on a precise economic study about the feasibility of

investment in Industrial IoT to implement this type of maintenance. Therefore, the

thesis aims to create a model aimed at evaluating the benefits obtainable from this

implementation in the manufacturing sector.

The first step taken to define the research field was the analysis of the existing

literature. We first analysed the papers to study how the IoT improves predictive

maintenance and creates value for the different actors of the value chain. Then we have

deepened the theme to fully understand its characteristics, the IoT structure necessary

for the installation, the technical and economic advantages, the barriers and all the new

business models that could arise from the implementation of this technology. The

study revealed that there are some gaps in the literature and explicit requests of future

researches in the creation of a cost model for the estimation of the economic benefits

of predictive maintenance. Moreover, we found out that among the major barriers to

adoption, there is a poor understanding of the benefits of predictive maintenance and

the consequent economic return on the investment. The articles on the subject report

the benefits only of a qualitative nature and without providing in-depth analysis. They

are generally used as evidence to support the thesis. On the other hand, those who

venture into the creation of a model only estimate the costs rather than provide an

estimation of the total benefits obtainable. For this reason, our thesis work tries to fill

this gap, creating a tool to quantify these benefits and to calculate the return on

investment.

Furthermore, recognising the importance that partnerships have to achieve the

objectives of the industry, we wanted to include this aspect in the model. We asked

ourselves how to incentivise the vendor to invest in IoT technology for predictive

maintenance. The price surplus and the competitive advantage are certainly two

aspects not to be underestimated, but the new business models to which the Industrial

IoT permits to access are the real treasure.

Our primary research question and a sub-question arose from these analyses:
xviii Executive Summary

 Q1: Does an investment in Industrial IoT for predictive maintenance

generate value for the company?

 Sub. Q1: How can the vendor benefit from the value generated to the

manufacturer?

Several approaches have been carried out to answer to the presented research

questions:

 Literature: The literature was extensively revised to find out the existing

gaps and where to concentrate the work, as well as to go deeper into

predictive maintenance theme (57 papers were analysed).

 Interview: To validate the notions collected with secondary sources and

to collect data regarding real situations, we carried out some interviews.

The total meetings with industrialists were three.

 Secondary sources: Secondary sources were also used as a first step

together with the literature to deepen the theme of Industrial IoT.

Similarly, they were used at a later stage to fill some literature

deficiencies, as well as a secondary benefits verification, therefore the

variables to consider in the model.

 Analytical model: An analytical model has been formalised to answer the

research question concretely. The model allows evaluating the economic

feasibility of investment in Industrial IoT for predictive maintenance.

The model, developed in Excel, represents the core of this dissertation

because it enables to concrete answer the research questions.

Model definition

For the evaluation of potential benefits obtainable with predictive maintenance

using the Internet of Things, an innovative approach has been developed. Starting

from input variables that characterise the state of the machinery and the type of

maintenance that is used, an analytical model has been constructed to quantify the

benefits and provides the Net Present Value of the investment.

Executive Summary xix

Hypotheses

Some hypotheses are necessary to make the model work better:

 H0: Since several companies today use corrective or preventive

maintenance (or a combination of the two), we have considered these

two situations as two possible initial stages.

 H1: To incentivise the manufacturer to purchase the predictive analysis

solution, the vendor must be as transparent as possible about the

installation costs. We hypothesised that the machinery supplier would

install the complete solution. By limiting the applicability to a single

machine, it is assumed that different faults can be monitored using the

same architecture, improving the multipurpose and interoperability

characteristics of an industrial IoT solution, no longer closed in a silo

perspective.

 H2: Apart from corrective actions that are unscheduled by nature,

preventive and predictive maintenance has a margin of error. Therefore,

the planned interventions are accompanied by unscheduled

interventions. It follows that the formalisation of costs per intervention

lies in the division into two types, unplanned and planned intervention

costs, estimated on an annual basis.

 H3: To limit uncertainty, we hypothesised that the intervention restores

the initial conditions of the machinery, allowing the system to function

until the project life.

Model Structure

Once the hypotheses to let the model work are set, it is necessary to specify the

type of variable that compose the model. The idea is to quantify the obtainable benefits

passing from a standard maintenance situation to a predictive in a Pay-per-

Performance solution. The benefits are used into the numerator of the NPV formula as
xx Executive Summary

differential cash flow, and they are discounted for the useful life of the machine

analysed.

𝑛
𝑆𝑎𝑣𝑃𝑑𝑀,𝑖 − 𝐷𝑖𝑓𝑓𝑀𝑆𝐶𝑖
𝑁𝑃𝑉 = −𝐼𝑛𝑣0 + ∑
(1 + 𝑊𝐴𝐶𝐶)𝑖
𝑖=1

Once the machine on which perform the analysis is chosen, the first important

step is to define the cost of the as-is situation. According to H0, the two as-is

maintenance can be corrective or preventive. Both solutions are composed by a

variable part, which varies according to the number of interventions, and fixed

annually based components. The structure of the intervention cost is the same in the

case of scheduled intervention and unscheduled, although some values may vary

accordingly. The machine is analysed, and the pieces with similar characteristics (such

as, type of maintenance, mean time to repair) are grouped in clusters.

The variable part, also called intervention cost, is composed of three main values:

𝐼𝐶 = 𝐷𝑇𝐶 + 𝑆𝑝𝐶 + 𝑃𝑛𝑙𝑡

1. Downtime cost (𝐷𝑇𝐶): This cost measures the loss of revenue that would

not have occurred if the machine had worked correctly and all products

had been sold. It is calculated multiplying the Mean Time To Repair of a

specific type of products (cluster) by the sum of the gross profit

contribution of every single piece.

2. Spare parts costs (𝑆𝑝𝐶): This value considers the costs of spare parts

technicians have to change after the breakdown occurs (or when they

perform preventive interventions) to fix the machine. It takes into

account also the change of other pieces that breakdown because of the

main breakage.
Executive Summary xxi

3. Penalties (𝑃𝑛𝑙𝑡): This variable considers a financial outlay caused by a

delay in delivering the goods to the customer because of a machine stop

caused by a breakdown.

Another essential aspect to consider is the failure rate 𝜆. Failures are unavoidable

consequences of using a machine. All the pieces, especially those that have to endure

more significant stress, break. Different patterns have been identified to describe their

degradation (linear, two-, three-, multi-stage) so their reliability, in this way it is

possible to define the statistical distribution of the behaviour of the pieces and

determine the right breaking moment. The proposed model requires a unique value

and could be estimated by summing the failure rate of the specific cluster.

Once the failure rates are set for each cluster, the initial cost of the two as-is

solutions can be computed. The total cost of the corrective intervention is simply the

multiplication of the unscheduled intervention cost by the failure rate since the

intervention is performed every time, and only after, a failure occurs. To this variable

part, a fixed annual charge is added (𝐸𝑋𝐶𝐶𝑟𝑀 ). This value is considered to take into

account the cost of the service and the warehouse costs.

𝐶ℎ𝑎𝑟𝑔𝑒𝐶𝑟𝑀 = 𝐼𝐶𝑢𝑛 ∙ 𝜆 + 𝐸𝑋𝐶𝐶𝑟𝑀

The preventive maintenance cost is a bit more intricate. It is calculated

multiplying the schedule intervention cost by the number of preventive intervention

that has been performed. However, since the preventive maintenance is not perfect

unscheduled interventions may be required. Therefore, the previous value is summed

with the unplanned intervention cost multiplied by an estimation of the error of the

preventive maintenance. Besides, the value obtained is added to a fixed annual part

that considers the cost of the service: internal or external.

𝐶ℎ𝑎𝑟𝑔𝑒𝑃𝑣𝑀 = 𝐼𝐶𝑢𝑛 ∙ 𝐵 + 𝐼𝐶𝑠𝑐 ∙ 𝑆 + 𝐸𝑋𝐶𝑃𝑣𝑀

xxii Executive Summary

In the end, the differential cost between both the as-is solutions and the predictive

maintenance is calculated. The predictive maintenance, as well as the preventive, is

composed of a scheduled intervention cost and an unscheduled one when the

prediction is wrong. Unlike preventive maintenance, the number of planned

interventions is calculated based on the accuracy of the prediction algorithm that is

evaluated using the confusion matrix. The savings formulas that arise are the

following:

𝑆𝑎𝑣𝑃𝑑𝑀−𝐶𝑟𝑀 = 𝐼𝐶𝑢𝑛 ∙ 𝑇𝑃 − 𝐼𝐶𝑠𝑐 ∙ (𝐹𝑃 + 𝑇𝑃) + 𝐸𝑋𝐶𝐶𝑟𝑀

𝑆𝑎𝑣𝑃𝑑𝑀−𝑃𝑣𝑀 = 𝐼𝐶𝑢𝑛 ∙ (𝐵 − 𝐹𝑁) + 𝐼𝐶𝑠𝑐 ∙ (𝑆 − 𝐹𝑃 − 𝑇𝑃) + 𝐸𝑋𝐶𝑃𝑣𝑀

Not always, a shift to predictive maintenance is advantageous. Therefore, the

model only considers the positive savings from the different clusters.

The last step to determine the numerator is to define the value of the subscription

for a pay-per-performance solution. It is calculated multiplying a fee entrusted by the

vendor, by the actual working hours of the machine.

Model Application

In this chapter, applications of the model are presented. In particular, a

theoretical case and a real case are introduced to investigate and understand the

impact of the different variables.

The theoretical model has been introduced to create a complete scenario that uses

all the variables and shows the computations behind the model gradually. It also

presents a sensitivity analysis conducted on the two variables that define the efficiency

of the prognostic algorithm to show the considerable fluctuation that depends on it.
Executive Summary xxiii

ConBio case study

ConBio is a vegan ready meal producer belonging to Granarolo Spa group. The

vast majority of the products are produced starting from vegan protein like tofu and

seitan that are produced internally. The tofu production has been investigated, and the

model applied to a unique and essential machine to which only corrective maintenance

is performed. Since this raw material is used in the production of several different

products, the impact of downtime is considerable. Two clusters have been identified:

one characterised by a long 𝑀𝑇𝑇𝑅 but a low failure rate, the second by a relatively

short 𝑀𝑇𝑇𝑅 and a high failure rate. The model shows positive values for both clusters

generating an overall NPV of 458,008 € with a payback shorter than a year.

Conclusions

The model quantifies in monetary terms the downtime, spare parts, penalties and

other variables associated with the breakdown of a machine by calculating the overall

cost of the as-is maintenance solution and a predictive maintenance solution with a

subscription pay-per-performance. The difference between these costs represents the

cash flows at numerator in the VAN formula, which provides the user with a specific

cost-benefit analysis for a specific machinery. This innovative model allows

manufacturers to make more informed choices; it also represents a useful support tool

for vendors to understand how much their customer's earnings are and to share the

benefits with them.

However, the model has room for future improvements and, among others,

considering other possible types of maintenance interventions is one of the main ones.

The calculation can be extended and adapted to include all maintenance strategies and

create a model that evaluates the optimal solution to be undertaken for each specific

case. Another improvement step could be achieved by understanding more in detail

possible present and future risks that the implementation of an Industrial IoT service

may involve. Finally, it is possible to integrate other considerations in the model which
xxiv Executive Summary

we have not taken into account, such as increasing the useful life of the machinery

which is declared as one of the benefits of predictive maintenance.

The model, the analysis and the validations performed, allow answering the

formulated research questions.

1. Q1: Does an investment in Industrial IoT for predictive maintenance

generate value for the company?

Savings achievable using an Industrial IoT system for predictive

maintenance depend on the characteristics of the as-is situation. In general,

they have a highly positive NPV in the case of corrective maintenance in the

as-is, with a high return on the investment on the whole useful life and a

payback lower than two years. In the case of preventive maintenance as the

starting situation, the positivity of the NPV depends much more on the

prediction algorithm effectiveness. Therefore, it is not possible to generalise

the conclusion, but it is necessary to adapt the model to each specific case.

2. Sub. Q1: How can the vendor benefit from the value generated to the

manufacturer?

Selling smart machines able to interact with the surrounding and stream

a massive amount of valuable data provide the vendor with a considerable

competitive advantage. Moreover, these data can be used directly from the

vendor to provide the predictive maintenance service in a pay-per-

performance agreement that generates additional revenues and better

customer relationships.

The results obtained from this thesis and the future developments that could

derive from it to overcome current limitations and explore new areas can lead to very

interesting and more accurate results. All stakeholders would greatly benefit from this,

and the research for more efficient sensors and more precise algorithms will be boosted

by the increased interest and investments on the subject.

Introduction
Fifty years ago, we delivered a message from a computer in the University
College of Los Angeles 3,000 miles away to a computer in Boston. It was the first time

two computers could communicate using a system different from the telephone lines.

In the following years, many other networks came online, but only ten years later, we

start the creation of one common network, a "network of network". It was the origin of

what we call today the Internet (Computer History Museum). The same years the

concept of the smart device was first discussed. Coca-Cola created a connected

vending machine able to report its inventory and whether newly loaded drinks were

cold or not and peoples around the world started to craft bizarre smart devices (Gupta

& Simmons, 2010)

In less than fifty years we evolved from connecting two computers to connecting

everything surround us, from the connected refrigerators that can purchase more milk

and eggs online before they run out, to machines that tell the operator how to optimise

productivity or detect a failure before it occurs: it is the time of the Internet of Things

Revolution.

The term Internet of Things (IoT) is nowadays gaining more and more

popularity. People are familiarising with it dealing with the purchase of a new

refrigerator, reading about new solutions to save our planet or discovering new ideas

aimed to solve the problem of traffic congestion or to create more reliable public

transportation. For decades sci-fi movies presented futuristic scenarios dominated by

technologies able to interact with humans, enabling revolutionary skills and creating

unimaginable tools that made the viewer dreams about the future. Today, technology

makes some of these things possible.

Internet of Things is a much wider theme than what we think or watch on the

screen, and it ranges from the futuristic technologies we dream, to much more concrete

and less astonishing solutions. The term 'things' is specifically used to refer to

everything around us, from the smallest particle to the biggest machine, from what we
Introduction xxvii

have to what we cannot even imagine. The things we refer to are all the smart objects

able to interact with the environment surrounding them. The IoT challenge seeks to

make smart more things possible, bridging the gap between the physical and the

virtual world enabling advanced services and other applications 1.

Much of the hype of these recent years centres on consumer applications, such as

digital assistants, who leads the scene between 2017 and 2018 in the smart fashion

industry (Reply), connected cars and consumer wearables like smart glasses.

However, the real revolution stands on the industry side of the market. According to

a forecast of Oxford Economics, the IoT's industrial applications, also called Industrial

Internet, will transform many industries, including manufacturing, oil and gas,

agriculture, mining, transportation and healthcare that account nearly two-third of the

world economy (Oxford Economics). The Industrial Internet may ultimately dwarf the

consumer side in potential business and socioeconomic impact. The integrated digital-

human workforce will redefine the job market, creating new working positions and

reshaping the very nature of work (World Economic Forum, 2015).

Given the great importance the Industrial Internet covers, this work focuses

exclusively on it. In particular, the aim of Chapter 1 is to introduce the importance of

manufacturing and the new trends. Chapter 2 is focused on the definitions of

Industrial Internet of Things, its boundaries and the future dimension of the

phenomenon. The industrial IoT theme is then analysed through surveys at the Italian

and world level in Chapter 3 to figure out the current level of diffusion. Chapter 4

analyses the literature presenting the most studied areas and discovering the least,

concluding with recommendations for further studies. Chapter 5 presents the

methodology followed in conducting the thesis and the research questions formulated

from the existing gaps in the literature. To answer these questions, a specific model

has been developed, and it is presented in chapter 6. In Chapter 7, the model is applied

to a theoretical and a real case to show its functioning and validate the results.

Conclusions are reported in Chapter 8.

1 See Appendix A for an in-depth analysis of IoT.

 Chapter 1

1 Manufacturing

In this first chapter, the changes in manufacturing will be deepened, presenting

all the smart technologies that are transforming this sector. These innovative

technologies available in the market are revolutionising the way of doing business and

the related operational processes around the world. A final focus on maintenance will

highlight the gaining of its strategic relevance in many sectors.

 2 Chapter 1. Manufacturing

1.1 Introduction to Manufacturing

Manufacturing is the cornerstone of the overall economy. In every developed and

developing country, we witnessed political proclamations regarding its relevance

(Trump, 2018). In Italy2, Carlo Calenda, Minister for Economic Development from 2016

to 2018, referred to manufacturing as following: «Our manufacturing companies are

the engine of economic growth and development with their ability to produce

prosperity and employment, feed the supply chain and service activities, contribute to

financial, economic and social stability.»3 (Calenda, 2017). Three points of view can be

considered to provide a general overview of manufacturing:

 Economic: the economic impact of manufacturing in developed-developing

countries is vast. In Italy represents approximately 15.1% of the GDP (World

Bank, 2018), not considering the part of the service-related that multiplicate

the former value of 2/3 times. Another interesting data regarding the Italian

landscape is about the export of this value: almost 40%; manufacturing

represents a key and strategic point in the competitiveness of Italy in the

world.

 Social: employment from manufacturing is crucial in developing countries as

in developed ones. In Italy, we counted 3.9mln directly employed workers,

and 6.5mln workers related to manufacturing business service (comparing to

an Italian population of 60mln). Every new job in manufacturing generates

two jobs in services, referred to as a "pull" effect (Semeraro, 2012).

2 Although we adopted a worldwide viewpoint, this paragraph reports some

more specific references on Italy. This choice was taken to better introduce the surveys
presented in the third chapter focused on that country.

3 «Le nostre imprese manifatturiere rappresentano il motore della crescita e dello

sviluppo economico con la loro capacità di produrre ricchezza e occupazione,
alimentare l’indotto e le attività dei servizi, contribuire alla stabilità finanziaria,
economica e sociale.» (Calenda, 2017)

.
 1.1 Introduction to Manufacturing 3

 Environmental: in Italy, energy consumption regarding industry accounts for

19.5% of the total primary source. Manufacturing has a strong lever to reduce

its carbon footprint, packaging waste, water usage and overall effect on the

environment. A strategy characterised by the new trend in sustainability (e.g.

the circular economy) will be key in the next years.

Manufacturing is part of a complex and highly competitive global economic

system that has witnessed different evolving business needs with their connected tech

requirements. Globalisation, decentralisation and fragmentation require more and

more visibility, business integration, decision support considering multiple products

and sites. The competition has different dimensions: competing on time requires the

need for fast decision making, simulations to promise reliable data; competing on

product variety with shorter lifetime involves the ability to manage uncertainty and

complexity; competing on service ask for connected products and extracting

knowledge from data.

Industrial IoT promises to deal with these business needs representing the new

source of competitive advantage. Governments play an essential role in contributing

to the development of this new field. Germany activated "Industrie 4.0", the German

industrial plan in 2011. Italy needed five more years to activate "Piano Nazionale

Industria 4.0" (2016)4, following UK (2011), Netherland (2014), Sweden (2014), France

(2015).

4 The Italian industrial plan is divided into a first part related to contribute for
Innovation and a second one about contribute for competitiveness. Regarding the
former, the instruments that the government has introduced are:
 4 Chapter 1. Manufacturing

The government incentives combine with the organic growth saw a considerable

evolution of the Italian Industrial IoT market: the investments related grew by 40%

between 2018 and 2019, from 1,35mld€ to 1,9mld€ respectively (Osservatorio Industria

4.0 - Politecnico di Milano, 2019).

Apart from the individual national industrial plans of European countries, it is

vital to report the concepts inside "2030 Vision for Industrie 4.0". The vision, jointly

developed by experts from Platform Industrie 4.0 (Germany), underlines as

international cooperation is indispensable for many industry 4.0 issues. It states that

"skills for the digital age or IT security are fields that are better addressed together.

Standardisation and regulatory challenges require cooperation with other countries or

supranational institutions" (2030 Vision for Industrie 4.0, 2019). The success of the

Industrial IoT system is connected with the success of European cooperation, and as a

result, Europe union will be more connected with Industrial IoT innovation.

 Iper e super ammortamento: the value of depreciation for investment in new

tangible assets, purchased or leased devices and technologies (enabling the
transformation into a 4.0 perspective) has been increased to 250% of the
original one. Regarding investments in intangible capital goods (software
and IT systems), it is possible to have access to the amortization of 140% of
investments.

 Nuova Sabatini: Contribution to partially cover the interest paid by the

company on bank loans.

 Credito d’imposta R&S: 50% of incremental R&D expenses will be considered

as a tax credit.

 Patent box: Optional taxation regime facilitated on derived income from the
use of intangible assets: industrial patents, trademarks registered, designs
and models industrialists, know-how.
 1.2 Smart Technologies 5

1.2 Smart Technologies

The Industrial IoT is the enabling and core element of industry 4.0 (Osservatorio

Industria 4.0 - Politecnico di Milano, 2019). The connection between the physical and

the digital world is central, and it can likewise be seen in the convergence of IT/OT

Smart Technologies. We presented the six smart technologies identified by the digital

innovation observatories of Politecnico di Milano, due to the importance that together

they will have in the manufacturing and the inter-correlation between them. It is

significant to underline two aspects: (1) a strong correlation between the ability to

adopt "traditional" technologies successfully (e.g. MES systems, PLM, traditional

automation) and the probability of success to approach Industry 4.0; (2) the Smart

Technologies represent the foundation of Industry 4.0, and not an endpoint. This

transformation will have to redesign processes and organisational models in the

problematic balance between operational management, continuous improvement and

radical innovation.

Cloud Manufacturing

For understanding naturally what cloud manufacturing is, we have to move

through cloud computing, the prior enabling technology. Through the internet, cloud

computing enables widespread, easy and on-demand access to a virtualised, shared

and configurable set of resources to support production processes and supply chain

management. Resources can range from infrastructure layer, IaaS (e.g. virtual

machines, storage), through the platform layer, PaaS (e.g. offering environments

already equipped with development applications, database management system, web

server), till the application layer, SaaS, where applications and data are also hosted

online on virtualised resources. Cloud computing has well-known benefits from high

reliability to scalability and availability in a distributed environment. Three significant

results obtained:
 6 Chapter 1. Manufacturing

1. The democratisation of engineering analysis, this result was reached thanks

to the possibility to request flexible-high computational power; this lets small

enterprises have access to practice like computational fluid dynamics (CFD),

virtual manufacturing and integrated planned that require considerable

computational power.

2. Industrial IoT platforms, they will be analysed with the next sub-chapter due

to the strict connection with Industrial IoT technologies.

3. Value chain collaboration, the visibility gained thanks to data shared in cloud

and managed collectively; this result represents the state-of-the-art of

Information System (Miragliotta, 2019)

In this area, Cloud Manufacturing paradigm is emerging: what does it happen if

the dynamically scalable and virtualised resource is not a hardware infrastructure, a

platform or a software application, but manufacturing? All the benefits related to

cloud computing will remain, scalability, high performance, real-time quoting, pay-

per-use, with the possibility to add services ranging from product design, testing,

management, and all other stages of a product life cycle. In cloud manufacturing,

distributed resources are encapsulated into cloud services and managed in a

centralised way (Xu X. , 2012). The type of manufacturing that seems more suitable to

move to the cloud is represented by additive manufacturing (Wu. D., Rosen D. W.,

Wang L., Schaefer D., 2014).

Industrial Internet of Things

The fundamentals of Industrial IoT are the smart objects (i.e., capable of

identification, location, status diagnosis, data acquisition, processing, implementation

and communication) and the smart networks (independent, standard and

multifunctional). It represents the connection between the physical and the digital

world in an industrial environment5. Industrial IoT platform can be defined as cloud

5 In the second chapter, a structure definition of the Industrial IoT will be defined.
 1.2 Smart Technologies 7

or edge environments for device management, data management, data analytics,

automation & data security functionality. The platforms also represent the point of

separation between the hardware state (sensors and actuation) and the software layer

(applications), making the progressive shift towards the software component evident.

The latter is often represented by Industrial Business Apps, i.e. vertical applications

dedicated to decision support in specific areas (such as optimising the life of cutting

tools, or maintenance). The big players in the Industrial IoT platform world also differ

in the strategies they use to develop this application layer. Three different approaches

are reported:

1. Offer applications natively available on the platform, developed directly or

through strongly supported communities;

2. Show a greater openness towards spin-offs, public and private accelerators,

as a source from which to attract developers;

3. Work on the accessibility of data, the availability of versatile and well

documented development tools, more distant from the application

development layer. In this way, they focus on application self-development

by the manufacturing company itself, through its own IT Function or

consultants or connected SW houses (Miragliotta, Macchi, & Terzi, 2019).

The availability of data in a platform environment opens up the connection to

environments like digital control room applications (allowing 360° monitoring of

production resources) and real-time scheduling applications (one of the historical

limitations of which was the lack of real-time feedback with the status of the shop

floor).

Industrial Analytics

With Industrial analytics, we indicated methodologies and tools for the treatment

and processing of big data coming from Industrial IoT systems or from the exchange

of data between IT systems to support the planning and synchronisation of production

 8 Chapter 1. Manufacturing

and logistic flows. There are included the applications of new techniques and tools for

business intelligence, visualisation, simulation and forecasting, data analytics.

The challenge of big data management is representing by the expansion of three

dimensions: volume refers to the amount of data, variety refers to the number of types

of data, and velocity refers to the speed of data processing (Laney, 2001).

It is useful to present the distinction between the types of data analytics. For this

purpose, an everyday explanation has been used: the question answered by the

descriptive analysis is: 'what happened?' – the diagnostic analysis: 'why did it

happen?' – the predictive analysis: 'what future? – the prescriptive analysis: 'how to

react to events?' – the pre-emptive analysis: 'how to avoid events?'

Advanced Human-Machine Interface

We referred to recent technological developments in the field of wearable devices

and new human/machine interfaces for the acquisition or transmission of information

in voice, visual and tactile formats. These devices include established systems, such as

touch displays or 3D scanners for the acquisition of gestural motion. At the same time,

more innovative and bidirectional solutions are being developed, such as augmented

reality viewers (both superimposed and peripheral vision) to support operational

activities and operator training.

Advanced Human-Machine Interfaces are enabling news ways to acquire or

deliver information to the operator, using voice, vision, touch and vibration, and to

empower human-centred operations. They range from consolidated solutions (e.g.,

touch displays) to more innovative solutions (e.g., visors, wearables). The salient fields

are represented by visual management, remote maintenance, training, and warehouse

operations (Osservatorio Industria 4.0 - Politecnico di Milano, 2019).

 1.2 Smart Technologies 9

Advanced Automation

Reference is made to the latest developments in automated production systems,

which are enriched with solutions characterised by high cognitive capacity, context

adaptation, self-learning and reconfigurability. The distinctive features of advanced

automation are different: the ability to interact with the environment, self-learning and

automatic driving; the use of vision and pattern recognition techniques (handling

systems, quality control); and finally the ability to interact with operators, thanks to

robots designed to operate in and alongside operators, rather than rigidly separated

from them. The development of these skills has taken place progressively over time in

the research and development laboratories of universities and manufacturers, and

now advanced automation becomes a concrete option in the design of a production

system (Osservatorio Industria 4.0 - Politecnico di Milano, 2019).

Additive Manufacturing

It is defined as a game-changing technology with exceptional capabilities and

able to substitute and reinvented historical processes. More commonly known as 3D

printing, this technology it leads to the creation of an object through "printing" layer

by layer, and it represents a revolution respects to the historically available production

processes (removal or plastic deformation of material) (Osservatorio Industria 4.0 -

Politecnico di Milano, 2019). The beginnings of this technology date back to the first

half of the '80s. In the last years, it has had an overwhelming development, expanding

the number of basic technological processes (selective laser sintering, electron beam

melting, fused deposition modelling and stereolithography) and the number of

treatable materials (both plastics and metals) with good finishing performance and

mechanical resistance. From the application point of view, additive manufacturing

finds application in 4 specific areas: 1- Rapid Prototyping; 2- Rapid Manufacturing; 3-

Rapid Maintenance & Repair; 4- Rapid tooling.

 10 Chapter 1. Manufacturing

1.3 Predictive Maintenance

In a manufacturing plant, maintenance has always been perceived as an

expensive activity and an obstacle to the pursuit of the economic goal. Only recently,

this widespread idea begun to be eradicated from the collective ideal and an unheard-

of awareness of its centrality has begun to spread. Smart Technologies, in particular

Industrial IoT, is playing a central role in this process, re-evaluating underutilised

maintenance techniques that suffered from large investments and poor returns in the

past: it is the case of predictive maintenance.

The concept of predictive maintenance was first introduced by the American

company Rio Grande Western Railroads back in 1940 and spread to other industries

throughout the 1950s, 1960s, and early 1970s. “Automotive, aerospace, military, and

manufacturing are the main industries where predictive maintenance has been

embraced and have shown several benefits in both efficiencies and cost savings”.

(Prajapati, Bechtel, & Ganesan, 2012). Although the idea of detecting early signs of

fault or failure to initiate maintenance procedures at the right time appeared in the

second part of the 20th century, “its rudimentary version, dates further back” (Selcuk,

2015). From its definition onwards, better ways of collecting, transmitting and

processing data have been sought. In the beginning, the prediction was based on the

skills of an expert maintenance person who used his senses of seeing, hearing, smelling

and touching to detect a sign of a problem. Over time, sensors have been created for

these tasks. The maintenance technician's experience has changed from the use of the

senses to detect these signs to the use of intellect to interpret them. Today, Industrial

IoT permits to collect unprecedented amounts of data and transmit them in real-time

to all users who have the access. The system is able to collect and evaluate data, decide

the right time to intervene, activate and conduct the necessary maintenance

procedures such as the mobilisation of personnel and the order of spare parts (Selcuk,

2015). Thanks to the current level of technological development, it is possible to deliver

the maintenance that is necessary rather than possibly required.

Given these premises, we can define predictive maintenance as “measurements

that detect the onset of system degradation (lower functional state), thereby allowing
 1.3 Predictive Maintenance 11

causal stressors to be eliminated or controlled prior to any significant deterioration in

the component physical state. Results indicate current and future functional

capability.” (U.S Department of Energy, 2010).

As we have highlighted, although visual inspection and human senses still

provide valuable information relating to the state of the system to be maintained,

predictive maintenance relies heavily on sensors for collection, on Industrial IoT for

transmission and processing. Consequently, the reduction in the cost of sensors typical

of recent years and the possibilities created by the Industrial IoT allowed this

technology to acquire credit in the industrial world. Therefore, it can be said that

predictive maintenance, as we know today, was born only in recent years following

the introduction of the Industrial IoT. This growing interest is also confirmed by the

relevant literature. Searching on Scopus "Predictive maintenance" as a key word it

turns out that the first papers published on the subject date back to 1970 but only after

2000 the number begins to rise above 100 to jump in the five years between 2015 and

today.

Figure 1.1: Trend of academic papers per year

 12 Chapter 1. Manufacturing

According to the energy department of the United States of America (2010), the

benefits obtainable by implementing predictive maintenance are manifold. In

particular:

 Return on investment: 10 times

 Reduction in maintenance costs: 25% to 30%

 Elimination of breakdowns: 70% to 75%

 Reduction in downtime: 35% to 45%

 Increase in production: 20% to 25%.

However, even if its benefits are known for ten years, there is not a clear

understanding of what they mean for a company in economic terms since they can

widely vary from case to case. This uncertainty, together with the difficulties to extract

valuable insights from data, explain why predictive maintenance has failed to take off

as broadly as expected (Schallehn, Schorling, Bowen, & Straehle, 2019). Although it is

still far behind its potential level of adoption, the long-term predictions remain

positive, and there is the certainty that takes hold in the following years and vastly

improves efficiency.
 Chapter 2

2 Industrial Internet of Things

An operational definition of Industrial IoT is provided together with the

architecture and an overview of the different connectivity typologies. The Industrial

IoT is compared to the three main revolutions on history, and a taxonomy of all these

revolutions has been performed better to understand this phenomenon, its

implications and its size. Finally, catalyst, precursors and risks are deeply analysed too

to figure out the reasons and the barriers behind the affirmation of this new

technology.
 14 Chapter 2. Industrial Internet of Things

2.1 Definition
It is tough to find a unique definition of "Industrial Internet of Things". Indeed,

there is not unanimity in literature, neither there is common consensus between firms,

international conferences and governments. Each of them has its view of this

revolution, with different definitions and different boundaries.

Some scholars define the Industrial Internet of Things (Industrial IoT) as one of

the "Smart Technologies" enabling the digital transformation in manufacturing, the so-

called Industry 4.0. According to this view, the Industry 4.0 lays its foundations in two

macro-categories: the first one, closed to the Information Technology (IT) area embed

Industrial Internet of Things, Industrial Analytics and Cloud Manufacturing; the

second one, closer to the Operational Technology (OT), include Advanced

Automation, Advanced Human-Machine Interface and Additive Manufacturing

(Miragliotta, Macchi, & Terzi, 2019). This classification presents the environment the

Industrial IoT belongs and the engines that propel the Industry 4.0 transformation.

However, it is a bit too vague and does not provide insights on the technology itself.

In order to create a precise definition of Industrial IoT that fit in the boundaries

presented above, we need something that embeds: (a) the kinds of technologies that

are used in an Industrial IoT settings and (b) the distinctive implication that it allows.

Therefore, we can compose the following. The Industrial IoT is a technology that

enables the connection between the physical and the digital world in an industrial

environment through a three-level architecture. Sensors and nodes compose the first

level, gateways and repeaters that allow this connection the second, platforms where

data are stored and managed the third (Miragliotta G. ). Although this definition limits

the Industrial IoT to the generation, transmission and storage of a large volume of data,

there is still an evolution compared to the traditional solutions. Indeed, the industrial

IoT empower flexibility (each machine that before communicate only within its

system, can now talk to every-‘thing' connected, generating a large amount of data)

and accessibility (the platform: i. Allows these data to be stored; ii. Acts as a general

skeleton where all the other smart technologies can access and operate; iii. Is a

universal link for all type of business applications).

 2.1 Definition 15

So far, we created a definition that perfectly fits into the industry 4.0 framework

and satisfies (a) and (b), however, it is too narrow and technology centred for our

purpose. To compose a satisfying definition, we must abandon this rigid and

theoretical framework and pay our attention to the real-world applications of the

Smart Technologies. From this perspective, emerge a deep integration among them

that thins their boundaries and reveals intersections. This suggests looking to the

Industrial IoT in a much broader acceptation, including in its definition also the

distinctive aims and purposes to which those technologies are put.

A different approach is presented in the report of the World Economic Forum in

2015. To define the Industrial IoT, they start from its roots, the Internet of Things. They

define the IoT as "a network of physical objects that contain embedded technology to

communicate and sense or interact with their internal states or the external

environment". Based on this definition, they simply say that the Industrial IoT is "a

short-hand for the industrial applications of IoT, also known as the Industrial Internet

of Things, or IIoT" (World Economic Forum, 2015). Similarly, Thames and Schaefer in

their paper regarding Cloud Manufacturing define the IoT as "a collection of physical

artefacts that contain embedded systems of electrical, mechanical, computing, and

communication mechanisms that enable Internet-based communication and data

exchange. The Industrial IoT follows the same core definition of the IoT, but the things

and goals of the Industrial IoT are usually different" (Thames & Schaefer, 2016).

Even if this conception provides a template for a definition of the Industrial IoT

and a primary criterion to distinguish IoT devices from Industrial IoT devices, it is too

simple and not enough detail for our purpose.

In the attempt to formulate an improved conception of Industrial IoT, we

searched further in the contemporary academic and industry-driven literature looking

for more informative definitions than those already mentioned. We found a few that

improved the basic one presented above. "Industrial Internet or Industrial Internet of

Things (IIoT) is built for bigger 'things' than smartphones and wireless devices. It aims

at connecting industrial assets, like engines, power grids and sensor to cloud over a

network" (Helmiö, 2017). A second definition states that "The Industrial Internet of

Things (Industrial IoT) is made up of a multitude of devices connected by

 16 Chapter 2. Industrial Internet of Things

communications software. The resulting systems, and even the individual devices that

comprise it, can monitor, collect, exchange, analyse, and instantly act on information

to intelligently change their behaviour or their environment – all without human

intervention" (Real Time Innovations Inc, 2015). The first definition introduces the

presence of a connection and tells us a little about the nature of this: that the relevant

assets are connected to a cloud, beyond a network. The second one omits the specifics

about the connection. However, it makes it clear what the functions of the Industrial

IoT devices are: to monitor, collect, analyse information to take useful insight to

manage the behaviour of the assets without human intervention. Two words of this

latter are significant for our purpose: to analyse and to act. Indeed, the Industrial IoT

is the bridge between the physical and digital world that make the Industry 4.0

revolution possible. It is the system where the other Smart Technologies exist and the

place where they can operate. For this reason, as several definitions do, we eliminated

the rigid boundaries of the industry 4.0 framework, generalised the concept and

embraced in the definition of Industrial IoT also some other Smart Technologies.

Thus far, we presented a general framework for Industry 4.0 and scrutinised how

Industrial IoT is part of it. We gave this Smart Technology a restricted definition, and

we put forward reasons why it is reasonable to consider the Industrial IoT in a broader

acceptation. Finally, we presented several different definitions found in the literature

that gradually add details to a first more simplistic. Now we can provide a complete

definition.

"Industrial Internet of Things: A system comprising networked smart objects, cyber-

physical assets, associated generic information technologies and optional cloud or edge

computing platforms, which enable real-time, intelligent, and autonomous access, collection,

analysis, communications, and exchange of process, product and/or service information, within

the industrial environment, so as to optimise overall production value. This value may include;

improving product or service delivery, boosting productivity, reducing labour costs, reducing

energy consumption, and reducing the build-to-order cycle" (Boyes, Hallaq, Cunningham,

& Watson, 2018).

As well as the first definition we gave of Industrial IoT related to the Industry 4.0

framework, this includes (a) the kinds of technologies that are used in an Industrial
 2.1 Definition 17

IoT settings and (b) the distinctive implications that it allow. However, as requested

by our purpose, this definition expands the boundaries previously set. It states that the

Industrial IoT system comprises "generic information technologies", the same Smart

Technologies that we consider as separate entities in the first acceptation. Indeed, the

implication is not only the mere connection between the physical and virtual world as

before, but also the functionalities it enables: real-time, intelligent, and autonomous

access, collection, analysis, communications, and exchange of process, product and

service information. Moreover, it also emphasises the clear distinction between the

consumers and the industrial applications specifying that the system operates in the

industrial environment. From now on, this will be our working definition of Industrial

IoT, and we will refer to all these functions as typical of this system.

Architecture

A reference architecture is a description of the Industrial IoT system at a higher

level of abstraction that helps identify problems and challenges for different

application scenarios. The design of an Industrial IoT architecture must highlight

extensibility, scalability, modularity and interoperability between heterogeneous

devices that use different technologies (Sisinni, Saifullah, Han, Jennehag, & Gidlund,

2018). In this part, we will present a quick overview of the several reference

frameworks of industrial IoT architectures originated in the past. Then we will define

and scrutinise what we intend to take as a reference (Weyrich & Ebert, 2016).

Generally, the Industrial IoT architecture models have a multilayer structure

with a specific function in each level depending on the business needs and technical

requirements. For example, the International Telecommunication Union has designed

an architecture consisting of five levels: detection, access, network, middleware and

application levels. Some academics have suggested instead, an architecture on three

levels: perception level (or detection), network level and service level (or application)

(Jia, Feng, Fan, & Lei, 2012); (Atzori, Iera, & Morabito, 2010). Others else have proposed

a four-tier architecture, which includes the detection layer, the network layer, the

service layer and the interface layer (Xu, He, & Li, 2014). The Reference Architectural
 18 Chapter 2. Industrial Internet of Things

Model Industrie 4.0 (RAMI) is a three-dimensional map showing all the crucial aspects

of Industrie 4.0 in a structured manner. Its axes identify the hierarchy levels that

represent the different functionalities within factories or facilities; the life cycle & value

stream of facilities and products; and the layers (Hankel & Rexroth, 2015).

The Industrial Internet Consortium published in 2019 an updated version of the

"Reference Architecture" document to continue the work of finishing and advancing

the 'Industrial Internet Reference Architecture' (IIRA). In particular, among those

presented we will adopt in this paper the widely accepted three-level model which

includes the edge, the platform and the enterprise levels.

1. The edge tier collects data from the edge nodes, using the proximity
network.

2. The platform tier receives, processes and forwards data from the edge level
to the enterprise level and the control commands in the opposite direction.

This level has multiple functions: it consolidates processes and analyses data

flows from different levels; provides management functions for devices and

resources; and offers non-domain specific services such as data query and

analytics.

3. The enterprise tier implements domain-specific applications, decision

support systems and provides interfaces to end-users (Industrial Internet

Consortium, 2019).

Connectivity

In this subchapter, we will make a brief excursus of the connection technologies

of the Internet of Things paradigm in general and not only applied to the industrial

world. In the analysis of the 'Osservatorio Internet of Things' of the Politecnico di

Milano, they have created eight clusters of the main IoT connection technologies

available internationally based on their architectural, functional and operational

characteristics (Osservatorio IoT Politecnico di Milano, 2018).

2.1 Definition 19

 Passive RFID (Radio Frequency Identification): it is the most straightforward

technology with which an object can be integrated into the Internet of Things.

It groups all the radio frequency automatic identification standards that do

not require the presence of a battery onboard the object;

 Active RFID (Radio Frequency Identification): provides additional

functionality compared to passive RFI thanks to the use of a battery, which

improves communication performance (reading distance) and enables

autonomous operation. They have included in this cluster only the protocols

with simple functionalities, mainly oriented to the point-to-point

communication of the object identification code, while the most advanced

protocols are included in the cluster of the Low Power Mesh Networks;

 Personal Communication: it group's standards for communication in short-

range networks (PAN - Personal Area Network) designed for consumer

applications and characterised by very narrow communication bands (such

as Bluetooth low-energy, ANT, NFC). The diffusion of these technologies has

received a significant boost thanks to their integration with most of the latest

generation mobile devices (smartphones, tablets);

 Wireless Bus: these standards (which include for example Wireless M-Bus,

KNX, X10) are a "wireless" alternative to wired solutions that have already

been used in the industrial world for some time. The Wireless M-Bus, a

protocol that supports, among others, the 169 MHz frequency band,

represents the most widespread technology. The technologies belonging to

this cluster do not allow sophisticated communication architectures;

 WiFi: These protocols allow wireless access to local broadband networks.

Developed for multimedia applications, which require the transmission of a

large amount of data, they have high energy consumption, which entails

severe limitations of applicability in the IoT field;

 Mesh Low-Power Networks (RMLP): networks formed by low-power nodes

and characterised by complex, self-configuring network architectures,

capable of supporting dynamic data routing and optimised for low energy

consumption (such as ZigBee, WHart). There is currently a great ferment on

 20 Chapter 2. Industrial Internet of Things

these technologies, considered one of the cornerstones of the development of

the IoT, and much work is being done on the standardisation of protocols;

 Cellular networks: these are the usual cellular communication technologies,

i.e. GPRS, GSM (2G), HSPA (3G), LTE (4G). Due to the high energy

consumption, they are mainly applied in the cases in which they can be

powered, as well as in combination with RMLP and Wireless Bus for

communication between second-level devices and control centres;

 PLC (Power Line Communication): the transmission of information takes

place by modulating the electrical signal used for the power supply. There

are both protocols designed for the residential world and for the medium and

high voltage network: the main difference concerns the maximum

communication distance and the supported data rate.

2.2 Taxonomy of Revolutions

Humankind is facing the edge of a new wave of innovation, a wave of the same

stormy sea that shaped the world we observe today. Only the companies capable of

riding this wave will survive and prosper, becoming the leaders of tomorrow. To fully

understand this disruptive phenomenon and his implications we must explore the

main events that reveal how we got here and how past innovations formed this wave

that we are calling "Industrial IoT" (Evans & Annunziata, 2012).

In this historical journey, we will analyse the main events that defined today's

businesses, following the path that led us from pre-industrial manufacturing facilities

to today's large multinationals. Taking the industrial perspective, we will investigate

the main technological innovations and the productivity growth, to depict the whole

picture of opportunities and benefits that the Industrial IoT can create to the

businesses.
 2.2 Taxonomy of Revolutions 21

Industrial Revolution

In the middle of the eighteenth century, the world experienced an unprecedented

change. For the very first time, the humankind was able to generate a continuous and

powerful flow of inventions and innovations in a few decades (Romagnoli, 2008). The

magnitude of this change was so powerful that it had a profound impact on the society,

the economy and the culture of the world. According to the GE Industrial Internet

paper, we dubbed "Industrial Revolution" the period between 1750 and 19006. During

these 150 years, innovations in technology applied to manufacturing, energy

production, transportation and agriculture started a sustained period of economic

growth and transformation.

Started in 1750 in Northern Europe, the most advanced economy at that time, the

First Industrial Revolution spread later in the United States, where railways played a

crucial role in accelerating the economic development (Evans & Annunziata, 2012).

This radical change hinges on the steam engine that began the irreversible transition

process from the exploitation to the creation of energy with the adoption of a new

energy paradigm (Romagnoli, 2008). The Second Industrial Revolution came later in

1870. While the First Industrial Revolution led to the growth of industries, such as coal,

iron, railways and textiles, the Second Industrial Revolution witnessed the mass

production, the emergence of electricity and synthetic chemistry, giving rise to an even

more powerful boost. Many other features characterised these 150 years (Freeman &

Louçã, 2010). The rise of large industrial enterprises created significant economies of

scale that led to a reduction in costs and prices resulting in an increase of volume

traded. The investments in dedicated plants and equipment have grown dramatically.

Innovation began to be thought of systematically; science became central to

technological development. Enterprises started to invest in research and development

(R&D) and work to exploit new inventions to create and profit from new markets.

6 As specified in the introduction, the aim of our paper is not to provide a detailed
taxonomy of the revolution rather present the main aspects useful to understand better
the change we are living today with the Industrial IoT. For further analysis of this
period look (Romagnoli, 2008); (Van Zanden, 2009); (Freeman & Louçã, 2010).
 22 Chapter 2. Industrial Internet of Things

Despite the considerable gains in the economy and society, the industrial revolution

also had downsides. Two aspects in particular need to be highlighted: the high

exploitation of natural resources, with the consequent impact on the external

environment and the poor working conditions. It is also important to mention for our

analysis that much of the research and incremental innovations after the Industrial

Revolution has been focused on improving efficiency, reducing waste and improving

the working environment (Evans & Annunziata, 2012).

Internet Revolution

The invention of the transistor in 1950 changed the world yet again (Evans &

Annunziata, 2012). It allowed all types of devices (electric motors, calculating

machines and operators in general) not only to run multiple programs simultaneously

at a very high operating speed but also to decrease its size. These characteristics have

given impetus to process innovations in all industries (changes in the plants and the

organisation of production) that resulted in huge gains of efficiency. The timeframe of

the Internet Revolution is much shorter than the 150 years of the Industrial Revolution.

It starts in the 1950s with the creation of the first large mainframe computer and the

invention of the concept of "packet switching"7 in the mid-1960s'. After the first

message sent in 1969, the technology continued to grow for all the '70s. Hundreds of

protocols were developed. Only in 1983, researchers began to assemble all these

protocols to create the "network of network" that became the modern internet

(Andrews, 2019). However, the invention the helped popularise the internet among

the public was the creation of the World Wide Web in 1990 (World Wide Web

Foundation); (CERN).

In a much faster and more connected world, and thanks to this powerful engine

based on openness and flexibility, the network diffusion was tremendous and the

speed of adoption breathtaking. In 1981, less than 300 computers were connected to

7 “Packet switching” is a method for effectively transmitting electronic data that

would later become one of the major building blocks of the internet (Andrews, 2019).
 2.2 Taxonomy of Revolutions 23

the Internet. Fifteen years later the number had increased up to 19 million. Today the

number is in billions. As well as the number of connections, the speed and volume of

information transmitted has grown dramatically. The speed of the best modems in

1985 was of 9.6 kilobits per second (Kbps) (Evans & Annunziata, 2012). Today the

fastest router transmits information 1 million times faster at a speed of 7.2 gigabits per

second (Gbps). Besides, the new 5G standard will guarantee a speed up to 10Gbps

enabling faster wireless internet that has the potential to benefit everything from

entertainment and gaming to education and public safety (T-Mobile). New platforms

for commerce and social exchange arose, scaling down the costs of commercial

transactions and social interactions. Companies started to appear on the network

creating their website and new efficient markets for exchanges.

The Internet Revolution resulted in several different transformations. It changed

the way of think about the production systems permitting deeper integration and more

flexible operations. It enabled concurrent innovation, transforming the orderly linear

approach to research and development and, thanks to the rapid exchange of

information and decentralise decision-making, it has spawned more collaborative

work environments unconstrained by geography. As a result, old models of

centralised internal innovation have given way to start-ups and more open innovation

models that take advantage of a more abundant knowledge environment. In

conclusion, we can affirm that the Internet Revolution was very different from the

Industrial one. While the Industrial Revolution was resource-intensive, the Internet

Revolution has been information and knowledge-intensive. It has highlighted the

value of networks and the creation of platforms. It has created new means to cut back

environmental footprints and produce more eco-friendly goods and services (Evans &

Annunziata, 2012).
 24 Chapter 2. Industrial Internet of Things

Industrial IoT8

Today, in the twenty-first century, the rise of the Industrial Internet of Things

paradigm promises to change the world yet again. The Industrial Revolution brought

the tools and the machinery into the factories; the Internet Revolution linked them

eliminating the geographical constraints, now, the Industrial IoT enables to open up

new frontiers to accelerate productivity, reduce inefficiency and waste and improve

the experience of human work (Evans & Annunziata, 2012).

Back in 2012, General Electric, in his paper on industrial IoT, stated, "the

Industrial Internet Revolution is already underway. […] However, we currently stand

far below the possibility frontier: the full potential of Internet-based digital technology

has yet to be fully realised across the global industry system." (Evans & Annunziata,

8 Today's rate of innovation is so high that it is not anymore possible to make a

clear distinction between the revolutions as it was in the past. At least a hundred of
years passed between the invention of the steam engine that defines the beginning of
the First Industrial Revolution and the electricity, that we can declare one of the main
innovations of the Second. With such a large period of time between the two, it is easy
to say what belongs to one and what to the other. Today there are so many innovations
every year that it is even difficult to be updated. There are several debates regarding
what is the Fourth Industrial Revolution, what are its boundaries and also if there are
the conditions for defining these innovations the Fourth or considering them an
extension of the third. We do not want to go into the discussion that generalizes too
much our topic. We believe that, from an industrial point of view, the potential benefits
that the Industrial IoT can generate are comparable to the one of the Internet or
Industrial Revolution. Therefore, the Industrial IoT for us is itself a Revolution.
 2.2 Taxonomy of Revolutions 25

2012). From that moment on, the Industrial IoT projects have taken off. Our survey 9

underlines that more than 80% of the projects undertaken worldwide have started after

2012. This data are confirmed by another survey of 2019 based on an Italian sample,

which shows that these last years have been the most thriving in terms of Industrial

IoT investments. Before 2017, only 16% of the Italian firms started an Industrial IoT

project while in the following three years, more than 60% invested in IoT. Moreover,

the number of Industrial IoT platforms has increased dramatically from 2015. Many

companies supplying software for logistics or production management have included

IoT solutions in their offer, which has turned from a differentiating element to a must-

have. Indeed, 74% of firms of our worldwide sample actively collaborate with these

companies to implement IoT solutions. However, the GE consideration regarding the

potentiality of the Industrial IoT is still valid: "we currently stand far below the

possibility frontier" (Evans & Annunziata, 2012). Our Italian survey underlines that

only 66% of the companies are aware of the Industrial IoT 10 and the average

knowledge of the theme is just sufficient (6.4/10). In particular, we must underline the

huge gap between large and small and medium-sized enterprises (SMEs). According

to a survey from 'Osservatorio Internet of Things' of the Politecnico di Milano on an

Italian sample, in large companies, 97% of the respondents heard about Industrial IoT.

In comparison, the value in the SMEs drops to 39% (Tumino, 2020).

9 Together with the ‘Osservatorio IoT' of the Politecnico di Milano we have

conducted two surveys and analysis. All the details regarding the methodologies and
further analysis are disclosed in Appendix B.
10 To assess whether people are aware on the Industrial IoT theme or not, we
assumed that only those respondents who marked a more than sufficient value (≥6/10)
to the question related to the knowledge of the Industrial IoT are enough aware of the
phenomenon and have enough knowledge to consider an investment.
 26 Chapter 2. Industrial Internet of Things

A World Economic Forum paper from a conference held in 2015 divided the

future of the Industrial IoT world into two macro periods: near- and long-term. They

also stated that near-term transformations would likely accelerate over the following

two years, while 79% of the respondents believe that the long-term ones would occur

within five years. Today, five years after the publication of that paper, we have the

opportunity to look back in time to evaluate what has been achieved and what is still

to be done.

Figure 2.1: The adoption and impact path of the Industrial Internet

Near-term

Near-term opportunities refer to gain in efficiency aimed at increasing revenues and

reduce costs. In a 2014 survey by the World Economic Forum, respondents indicate

that companies are adopting IoT solution in the near-term either to drive down

operational costs, 81% indicate that it is a "very to extremely important" driver for

adoption, or to optimise asset utilisation, 74%. Accordingly,(World

the Economic
most mentioned
Forum, 2015)

application of Industrial IoT in the survey is predictive maintenance and remote

management of resources, which can reduce equipment failure or unexpected

downtime by analysing operational data now available (World Economic Forum,

2015). According to our global sample, Jeep, TetraPak, Caterpillar, Rio Tinto and ABB

are between the early adopters of the Industrial IoT technologies. In particular, Rio

Tinto, the world's second-largest metals and mining corporations, implemented

 2.2 Taxonomy of Revolutions 27

sensors, analytics and real-time data to help identify existing maintenance issues

before an actual malfunction or accident happens. Collecting data from the company's

fleet continuously, they can save $2 million a day each time a breakdown is avoided

(McGagh, 2014). Efficiency is not only cost saving, but it also is waste reduction.

Sustainability continues to play a central role for companies; 67 % of participants to

the survey mentioned above consider the improvement in sustainability as an

important driver of adoption of IoT solutions. Companies have been investing in IoT

in all these years, and today we can affirm that knowledge of technology and on its

potential in terms of increasing efficiency has reached a good diffusion. In light of this

awareness, companies are continuing their digitalisation process.

Long-term

In the long-term, the World Economic Forum paper predicted that the Industrial

IoT would generate structural changes, creating the opportunity for new business

models based on services to disrupt industries. Indeed, the wave of change is so

powerful that new digital entrants will increasingly subvert incumbents by bringing

the power of software, the speed and scale of the Internet and nimble business models.

Few companies, even the world's largest ones, have the necessary instruments to

manage this wave. The only way to compete effectively for established companies is

to start cooperating in an unprecedented way and start thinking in terms of ecosystems

(ICT4Executive, 2014). In this joined scenario, any company must find his position in

the value chain, but it is also important to consider that platform providers are those

who capture the biggest pie and therefore are the best positioned to lead the formation

of the partnership (Agrawal, 2016)

Traditionally, the reputation of product companies has been built around a solid

pillar: providing high-quality products at competitive prices. The companies tried to

meet customers' needs most efficiently and effectively possible by creating: aircraft
 28 Chapter 2. Industrial Internet of Things

with higher capacity but burn less fuel; tractors that plant faster; and light bulbs that

last longer but consume less energy. In recent years, thanks to IoT systems, producers

have the opportunity to satisfy their customers by providing them only what they need

to achieve specific results: the optimisation of the transport of people over long

distances, the increase in crop yield and the lighting only when necessary. This

attention to the resolution of the why behind the buy is a key factor in the ongoing

evolution from products to services. The increasing availability of smart products will

accelerate this process. In the World Economic Forum paper, this new solution is called

"outcome economy", and it is defined as the economy "where businesses compete on

their ability to deliver quantifiable results that matter to their customers in a specific

place and time." (World Economic Forum, 2015). The two main insurmountable

obstacles toward achieving the outcome economy are the deep understanding of

customer needs and the context in which products and services will be used, and the

possibility of quantifying the results in real-time (World Economic Forum, 2015). The

IoT solutions applied to the industrial world have solved both of these problems.

Smart objects generate a continuous stream of a tremendous amount of real-time data

that can be analysed to get unprecedented insights into the use that consumers make

of it.

The World Economic Forum survey also underlines that 74% of the respondent

indicated that the possibility to create new revenue streams through new products and

services is significant in driving the adoption of Industrial IoT solutions. Indeed, as

anticipated in the paper, companies have started to move toward this new service

economy. According to our global sample, 55% of the companies have understood the

potential of services related to IoT solutions and have introduced them in their offer.

For example, Rolls Royce's TotalCare option removes the burden of engine

maintenance from the customer and transfers the management of associated risks to

the company itself. They provide a suite of predictive maintenance and repair services

for their jet engines, including status monitoring and direct maintenance to increase

reliability and durability. They completely aligned their business model to the one of

the customers, making the uptime the common good (Rolls Royce).
 2.2 Taxonomy of Revolutions 29

Another example is Taleris America LLC, an American based company resulting

from the partnership between Accenture and GE Aviation System. While Rolls Royce

is an excellent example of the incumbent that evolve its business model, Taleris is a

perfect example of a business outcome of the new service-based economy. Unlike Rolls

Royce's TotalCare service, which focuses on jet engine uptime, Taleris addresses the

broader issue of airline delays and cancellations caused by equipment failure focusing

on the optimization of the airline's entire fleet. In this way, Taleris has control over

general maintenance programs. This systemic approach leads to several benefits such

as fewer disruptions, lower costs, better spare parts inventory management and more

satisfied travellers (World Economic Forum, 2015).

A second long-term goal is to move towards an integrated digital and human

workforce. Much has changed since the Industrial Revolution period where employers

exploited their workers up to their limit. Nowadays, the workforce became a core

resource for the company, their safety and expertise a valuable asset. The Industrial

IoT will lead the world towards a mixed workforce, where humans with machines will

work together to deliver outcomes that neither could produce alone (Brynjolfsson &

McAfee, 2014). According to the World Economic Forum survey, 94% of respondents

believe that the Industrial IoT will fundamentally transform what skills and jobs are

required in the future. If we design and apply technology to empower rather than

replace people, this "human-centred automation" or "augmentation" can redefine

existing jobs and create new ones.

An example of this is Amazon with its recent acquisition of Kiva Systems. An

AGV system receives the input of where to pick the required shelf, and it brings it to

the human just in time to have a reasonable queue in front of the operator

(AmazonRobotics). This system of humans and robots work side-by-side allows

fulfilling orders 70% faster than a non-automated warehouse (World Economic Forum,

2015). Today, workers must be physically on the shop floor to operate on a machine.

Thanks to the Industrial IoT, a manufacturing engineering can potentially receive

notifications on his tablet or smartwatch when a machine is malfunctioning. An

example of this application is present in our global database. Alfa Romeo, an Italian

car manufacturer part of the FCA group, has collaborated with Samsung to create a
 30 Chapter 2. Industrial Internet of Things

digital system to visualize, plan and manage the production process. They placed

tablets in strategic factory locations and equipped each operator with a smartwatch. In

the event of anomalous situations, alerts are displayed directly on the operator's

smartwatch that guides him step by step in the operations to follow. After sending the

repair confirmation with the smartwatch, the activity is recorded as completed, and

the production line can continue to work (Digital4, 2017).

Sizing the Industrial Benefits

It is not easy to define univocally the scale of the opportunity of the Industrial

IoT. As we stated at the beginning of this chapter, there is much confusion around the

terminology and the boundaries of the term Industrial IoT. There are journalists, firms,

international experts and academics that consider the Industry 4.0 to be synonymous

with the Industrial IoT. Others to whom Industry 4.0 describe the Fourth Industrial

Revolution and the Industrial IoT a technology enclosed in it and some others to whom

the Industry 4.0 is only the erroneous translation into English of the German

government plan "Industrie 4.0". At the same time, the correct English equivalent is

Industrial Internet. Accordingly, there are some to whom the Industrial IoT embed

only what is related to the production plant and others that make a clear distinction in

the IoT world considering Industrial IoT everything that does not concern the

consumers' world. As we have presented in our introduction, in this chapter, we

adopted the latter full acceptation that helps us in understand better the potential

benefits of this digital revolution. However, it is essential to specify that the focus of

this paper will be on the manufacturing production plant, and we will shrink the

perimeter only to this later on.

In the attempt to create the clearest picture possible of the opportunities of the

IoT, it is useful to understand first the size of the system in which it is integrated.

However, there is no single simple measure. Therefore, according to the GE Industrial

Internet paper, we suggest three different perspectives: economic share, energy

requirements, and physical assets. Although not exhaustive, taking the three measures
 2.2 Taxonomy of Revolutions 31

together gives a useful perspective on the vast potential scale and scope of the

industrial IoT.

Economic Perspective

In traditional economic definitions, the industry includes manufacturing, natural

resources extraction, construction, and utilities sectors (United Nations, 2008). The

global industry, based on these categories, represents in 2018 about 25% of the global

GDP, around $22 trillion of the $86 trillion dollars of the world economy. The

manufacturing represents the vast majority of the industrial part, 16% of the world

GDP, while the other industries together account for 10%. These numbers hide some

differences at the regional level that depend on the composition of the economy. In

developed countries, the industry counts 23% of the total GDP while in developing

one-third of the overall economy (The World Bank, 2018)11.

Even though 25% of the global economy is an immense value, it does not capture

the full extent of the Industrial IoT potential. The Industrial IoT promises to enhance

and disrupt a much broader array of sectors: transportation, health-care and

agriculture will take part in this tremendous revolution. In recent years, this digital

technology created disruptions also in the services sector. Dozens of start-ups

providing platform-based solutions to industrial manufacturers receive millionaire

funding each year (Columbus, 2019).

Energy Consumption Perspective

11 Industry corresponds to ISIC divisions 10-45 and includes manufacturing (ISIC

divisions 15-37). It includes the value added in mining, manufacturing, construction,
electricity, water, and gas sectors. The value-added is the net output of a sector after
adding all the outputs and subtracting the intermediate inputs. It is calculated without
deductions for the depreciation of manufactured goods or the depletion and
degradation of natural resources. The International Standard Industrial Classification
(ISIC), revision 3 or 4, determines the origin of the value-added.
 32 Chapter 2. Industrial Internet of Things

As we have seen before, the adoption of IoT technologies to create energy-saving

efficiencies and cost reduction is one of the main reasons that has driven investments

in this area. Furthermore, the Industrial IoT can even be seen as a solution to the

problems related to the scarcity of resources and environmental sustainability that are

creating increasing pressures on the energy system. Therefore, the energy footprint

associated with the global industrial system represents another interesting perspective

to analyse in order to understand the potential of Industrial IoT. The industrial sector

uses enormous quantities of energy. If we add energy production and conversion to it,

the scale of the benefits of the industrial IoT includes more than half of the world's

energy consumption (Evans & Annunziata, 2012).

In the industrial sector, the steel and metal industries and the petrochemical one

are the main energy consumers representing around 50% of the energy consumed.

Some studies have shown that if the best technological practices were implemented,

heavy industry energy consumption could be reduced by 15 to 20% (IEA, 2009).

Transport is another sector that consumes a great deal of energy. This sector occupies

27% of global demand (mainly oil), of which approximately 50% is used in the

industrial sector for road transport, air and ship connections. Assuming that most

large and part of light vehicle fleets can benefit from the use of information technology

and networked devices and systems, around 14% of the global demand for

transportation fuel can be affected by Industrial IoT technologies. (Evans &

Annunziata, 2012).

Physical Asset Perspective

A third perspective that helps us understand the potential of industrial IoT is that

of the physical resources involved in various parts of the industrial system. The

industrial system includes millions of machines worldwide, ranging from simple

electric motors to highly advanced computerized cosmography used in healthcare

delivery. All these devices generate a huge amount of information (temperature,

pressure, vibrations and other key indicators). They are useful for understanding the

performance of the unit itself and concerning to other machines and systems. The
 2.2 Taxonomy of Revolutions 33

Industrial IoT allows to improve or create systems to monitor, modelled, and

manipulate all these crucial metrics remotely, to provide safety, enhance productivity,

and operational savings (Evans & Annunziata, 2012).

Until now, we have seen the areas where the Industrial IoT can have an impact,

and now we will examine what it can offer to the industrial world. The industrial IoT

opens the door to several benefits that are the cause and the consequence of the other.

On the one hand, the new sensor technologies allow detecting an increasing number

of data more precisely; on the other hand, the increasing computational power allows

analysing these data more quickly and more in detail. Together, these two technologies

create a combination that transforms machines from automatic to autonomous.

Besides, this intelligent instrumentation significantly improves the machine's

performance, increases its efficiency by reducing costs and increases reliability. An

intelligent machine is a machine connected to a system of machines that allows a

previously impracticable level of integration and collaboration. By collecting and

analysing data from all these machines, it is possible to obtain valuable insight into the

business's operations. With this information, it is possible to optimize the decision-

making process that further increases the productivity of the machines and maximize

enterprise's performances (Industrial Internet Consortium, 2019). Furthermore,

continuous learning allows the better design of new products and services, leading to

a virtuous cycle of increasingly better products and services resulting in greater

efficiency and lower costs (Evans & Annunziata, 2012).

An interesting point regarding the industrial Internet is what is called the power

of 1%. What it relates to is how small improvements, even a small 1%, can produce

substantial system-wide savings if scaled up across the sector. For example, in the

aviation sector, fuel savings of 1% per year equate to savings of $ 30 billion considering

the whole sector. Likewise, the 1% fuel savings for gas-powered generators in a power

plant enable operational savings of $ 66 billion. Finally, in the oil and gas industry, a

1% reduction per year in capital expenditure on equipment would generate savings of

approximately $ 90 billion. The same goes for the agricultural, transportation and

healthcare industries. Therefore, this broad view shows us that in most sectors, if the
 34 Chapter 2. Industrial Internet of Things

industrial IoT allows making only a modest 1% improvement, there would still be a

huge saving and a return on the investment (Evans & Annunziata, 2012).

2.3 Catalysts, Precursors and Risks

of the Industrial IoT

We thought it necessary to end this first chapter by making a brief excursus on

the aspects that had and are making possible the digital transition in the industrial

world. Today, the Industrial IoT development process has made strides and is

emerging as a revolutionary technology globally. However, its success is not yet a

foregone conclusion. To achieve its goal of combining the physical world of machines

with the digital world of data and analysis to reach its full potential, both continuous

progress in software and hardware and continuous government incentives to

encourage its adoption and dissemination are essential. In this paragraph, we will first

scrutinize the technical enablers of the Industrial IoT. Then, we will take a look at the

economic and social dimension, concluding with the main challenges and risks.

Technical Dimension

The technical state of the art is of fundamental importance when we want to

investigate the causes that have favoured the development and affirmation of new

technologies. The Industrial IoT is the result of innovations already underway, which

include technological innovations, innovations of systems, networks and processes,

which together represent a set of catalysts and vital activators.

 2.3 Catalysts, Precursors and Risks of the Industrial IoT 35

Listed below are some categories of high-level technologies that have been

instrumental in the development of industrial IoT12:

Sensors

Sensors play the role of "translators" of real quantities into virtual information.

The continuous evolution of the sensors with the progressive miniaturization of the

devices, the reduction of costs and energy consumption, is one of the fundamental

aspects that has stimulated the rapid diffusion of these devices, making smart an

increasing number of devices. Furthermore, technological innovation in this field has

allowed a significant increase in data quality. By comparing data from the dozens of

heterogeneous sensors installed on the machines, it is possible to have very high

reliability of the data (Capone, Pitic, Tumino, & Salvadori, 2018).

Networks

Industrial networking is very different from consumer networking. The

requirements from applications on the supporting networks are very different, as well

as the technical requirements and the application scenarios, the physical conditions

encountered in mining extraction, for example, differ significantly from those of

agriculture. The industrial network infrastructures and the technologies that support

them are constantly evolving. The introduction of new technologies increases the

possibilities of the networks (increasing reliability and coverage, reducing the latency

and energy needed to operate). It allows an increasing number of industrial resources

to connect (Höller, 2018).

12 The list complies with that reported in the slides of Giovanni Miragliotta
professor of Supply Chain Management at the Politecnico di Milano (Miragliotta G. ).
 36 Chapter 2. Industrial Internet of Things

Advanced Analytics

The continuous miniaturization of the processors has led to a continuous increase

in computing power, a reduction in the cost and the energy required. This trend led to

the creation of increasingly powerful calculators and made the advent of advanced

analytics possible. Thanks to advanced analytics, it is possible to examine a large

amount of data or content autonomously or semi-autonomously using sophisticated

techniques and tools (Gartner). In this way, the Industrial IoT acquires its own brain

and is able to discover insights, make predictions or generate recommendations.

Platforms

In order to get value from the Internet of Things, it helps to have a platform on

which to create and manage applications, to run analytics, and to store and secure

sensible data (Lamarre & May, 2017). The transition from specialized platforms, with

the consequent manage of data in a silos way, to general platforms, embedding the

entire features from device management to security, represented a fundamental step.

Moreover, the addition of an ecosystem of APIs that allows data moving smoothly

between platforms and the rise of modular services, together with the possibility of

providing these services directly in the cloud and the presence of virtualized

interchangeable hardware, has made the development of these solutions cheaper and

faster (Miragliotta G. ). The world of the IoT platforms is booming, we went from 260

of 2015 up to 620 in 2019, and dozens of platform-based start-ups arise every year

(Lueth, 2019).

Economic and Social Dimension

Technical innovations are not the only dimensions that drive the Industrial IoT

adoption, and a technical assessment of the reediness level of the technology is not

enough for it and the ICT technologies in general. The European Connect Advisory

Forum is considering new alternatives to substitute the Technology Readiness Levels

 2.3 Catalysts, Precursors and Risks of the Industrial IoT 37

(TRLs) developed in the 1980s by NASA and particularly suited to that context.

Indeed, they believe that ICT applications require a development that takes into

account adequate business models, user involvement and social aspects and, therefore,

they require new models as guidelines. One of the proposed models is the "Market

Adoption Readiness Level" (MARL) which, in addition to the technology readiness

parameter, requires the evaluation of three values: users, data, and the level of risk

(Connect Advisory Forum, European Commission, 2014). We will not go into detail

with the parameters of this model that are too general; instead, we will examine some

economic and social aspects that we believe fundamental drivers of the Industrial IoT.

Government Programs

According to a World Bank report of 2017, the government have a central role in

catalysing the space and contributing as partners/leaders in the long term (The World

Bank, 2017). Since the first time Germany activated the 'Industrie 4.0' plan in 2011,

several other countries understood the potential of the IoT technology and the key role

that the government played as a facilitator and initialised their own. Thus, in the

following years, the 'Advanced Manufacturing Partnership' plan was born in the

United States in 2012, the 'Made in China 2025' plan raised in 2015 in China and the

'Piano Nazionale Industria 4.0' in Italy in 201613. Only by creating the right

infrastructures and implementing the right practices to encourage investments,

companies will have the chance to seize the immense opportunity that is the Industrial

IoT.

Knowledge and Awareness of the Potential

It is impossible even to imagine investing in IoT unless having a thorough

knowledge of the subject. As we have seen before in this chapter, the Industrial IoT is

a very confusing and complex issue. Although knowledge at the corporate level has

13 A detailed analysis of these industrial plans will be conduct later in Chapter 2.

 38 Chapter 2. Industrial Internet of Things

increased significantly in recent years, it is still relatively scarce. It is for this reason

that universities with their observatory, researchers, and international consortia, play

a fundamental role in the dissemination of knowledge and best practices on the topic.

Furthermore, this technology will subvert the current job market, increasingly

requiring a skilled workforce. Companies and workers must update if they want to

fully understand the potential of industrial IoT in their business and make the correct

and necessary investments to keep up with the times.

Commitment to Innovation

A company that adopts Industrial IoT must commit to innovation, as well as

adopt a long-term perspective for the return on the investment of the project. Funds

will be needed for sensors, to update machines and systems. Investment and patience

will be required as the data acquisition process, the configuration of the analysis

parameters and algorithms may not give immediate results; success and returns may

take some time but the investment will be more than pay off.

Major Challenges and Risks

Despite the great promise and new opportunities of the Industrial IoT, many

factors could hinder future growth. As shown by the previously cited survey done by

the World Economic Forum, almost two-thirds of respondents agree with the

widespread opinion that security and interoperability are the two biggest obstacles.

Other significant barriers mentioned include the lack of a clearly defined return on

investment (ROI) (53%), legacy equipment (38%) and technological immaturity (24%)

(World Economic Forum, 2015).

Regarding risks, 76% of the respondents identify cyber-attack vulnerabilities as

their most important concern. A related but slightly different risk is the violation of

the privacy of personal data, which is also classified at the top (68%). Both are justified

when considering the impact that a hacker attack could have on a connected plant,
 2.3 Catalysts, Precursors and Risks of the Industrial IoT 39

which, for instance, could block the entire production of a manufacturing company or

deny the flow of energy from a power plant to an entire city (World Economic Forum,

2015). A World Economic Forum report by 2019 underlines that no industry is

untouched by cybercrime. They estimated that from 2019 to 2023, approximately $5.2

trillion in global value would be at risk (Ghosh, 2019)14. Investments in

countermeasures are increasing as a result.

Since the industrial IoT has the potential to introduce countless innovations in

existing business models and disrupt industries, for a large majority of the incumbent

surveyed (88%), this represents another significant risk. As this technology reduces the

internal barriers to each industry, it opens the doors to new players, even purely digital

ones such as platform providers. Companies will have to adapt to this new competitive

environment and take advantage of the flexibility offered by this technology by

collaborating with many organizations across the ecosystem. Indeed, this

collaboration will be essential if companies want to meet the growing expectations of

their customers, i.e., it is difficult to see how a company can master the entire digital

value chain (World Economic Forum, 2015).

Finally, the last risk that has been most highlighted by respondents is the

potential shift of jobs that will occur in some industries due to increased automation.

The technological evolution over the years had already shown something similar, such

as what happened in the communications sector when technological solutions

replaced the switchboard jobs. As smart machines and IoT technology become more

and more widespread, more and more jobs will be affected, even those considered

purely human. For example, a McKinsey analysis shows that, with the current rate of

technology improvement, half of today's work activities could be automated by 2055

(Manyika, et al., 2017). However, it should be noted that, just as previous technologies

that have eliminated certain types of jobs have created others, so Industrial IoT will

do, creating new jobs that require unique human attributes, such as creativity, critical

thinking and collaboration. Industry leaders and governments must note that

14 For further information regarding the value of the cybersecurity issue see
(Desjardins, 2017).
 40 Chapter 2. Industrial Internet of Things

technology is constantly increasing the bar for low-skilled jobs and requires

continuous updating of skills. This consideration, again, highlights what has been said

previously regarding universities. Together with all kind of educational centre, they

will play a primary role in this game. Actions are urgently needed to refocus attention

on education, adapting current educational systems and approaches to better prepare

the new generations for the upcoming digital work environment (World Economic

Forum, 2015).

(World Economic Forum Industrial Internet Survey, 2014)

Figure 2.2: How likely are the following risks or negative consequences associated with
the Industrial Internet?
 Chapter 3

3 Industrial IoT Research

Two surveys investigating the theme of Industrial IoT will be presented, and

their results discussed. They regard both the Italian manufacturing sector, one looking

to the Small-Medium Enterprises (SMEs) and the other to Large Enterprises (LEs).

Then, the case of Fabio Perini S.p.A. is reported as an example of a company dealing

successfully with the business changes brought by the Industrial IoT, offering services

from predictive maintenance to “productivity” consultancy. At the end of this chapter,

the predictive maintenance is highlighted as the most interesting Industrial IoT

application, considered as the starting point for the literature review.

 42 Chapter 3. Industrial IoT Research

3.1 Survey Analysis

The Industrial IoT will experiment a great period of diffusion, faster in some

areas than others, but with the potentialities to impact the majority of manufacturing

companies. For more details on its diffusion and possible applications, we presented

two surveys regarding the Industrial IoT with the focus on Italy. They were conducted

inside the ‘Osservatorio Internet of Things’ of the Politecnico di Milano15. A survey is

related to Large Enterprises (LEs)16, the other regards Small and Medium Enterprises

(SMEs)17. Furthermore, about ten direct interviews from the sample of LEs were

scheduled and conducted to deepen some of the more interesting survey answers. The

15 The ‘Osservatorio Internet of Things’ is part of the Digital Innovation

Observatories of the School of Management of the Politecnico di Milano, that represent
the reference point in Italy on digital innovation. The research activities are carried out
by a team of almost 100 Professors, Researchers and Analysts working on 40 different
Observatories that address the key issues of Digital Innovation in Business (including
SMEs) and Public Administration, divided into three areas:
 Digital Transformation, which includes Observatories that analyse in a
transversal way the digital innovation processes that are deeply transforming
our country;
 Digital Solutions, which brings together the Observatories that study in depth
specific application and infrastructure areas related to new digital
technologies;
 Verticals, which includes the Observatories that analyse digital innovation in
specific sectors or processes (Osservatori Digital Innovation della School of
Management del Politecnico di Milano).

16 In October 2019, 714 questionnaires were sent to companies operating in the

Italian industrial context to understand their point of view on the Industrial IoT theme.
The research was carried out using the CAWI methodology (Computer Assisted Web
Interview). All the details regarding the analysis in Appendix B.

17 In December 2019, a sample of 525 small and medium enterprises,

representative of the Italian scenario by sector, geographical area, number of
employees and turnover, was carried out. The research was conducted using the CATI
methodology (Computer Assisted Telephone Interview). All the details regarding the
analysis in Appendix B.
 3.1 Survey Analysis 43

aim of both the researches was to collect data about: (1) level of knowledge regarding

IoT field; (2) the area of completed/underway projects, and related services added; (3)

objectives, benefits, and use of data; (4) barriers and how to manage them; (5) economic

dimension of the investments done in Industrial IoT, and the area of interest in future

projects. The sub-chapters will follow the structure of the survey to deepen the

application of Industrial IoT in manufacturing.

Knowledge

The Industrial IoT is bringing opportunities to accelerate productivity and reduce

inefficiency in LEs as in SMEs. How aware are these companies? In a first instance, a

huge disparity is witnessed between the two groups: only 39% of SMEs respondents

respect to 97% of the LE ones indicate to have at least heard about Industrial IoT (see

Figure 3.1). Furthermore, inside the percentages, it was detailed the level of knowledge

with answers between 1 to 10. The medium result between 6 and 6,5 was observed for

both the two samples. These results show the need for a significantly increased in

awareness (especially for SMEs), necessary to fully understand the potentials and

make the correct and necessary investments. As reported previously and now

underscored by the surveys, the Industrial IoT is still at an early stage; in this phase,

the dissemination of knowledge represents a critical aspect, and universities are called

upon to play a key role in overcoming these first steps.

Figure 3.1: Results of Q1.1 - Sample size LEs, 100. Sample size LEs, 525 - Have you
ever heard of the Internet of Things (IoT) solutions for Industry 4.0?
 44 Chapter 3. Industrial IoT Research

Completed/Underway Projects

The majority of LEs, 66%, has in their actual present/past an Industrial IoT

project, respect to 13% of SMEs. The observatories of the Politecnico di Milano

proposed a categorisation of the Industrial IoT application in three main fields: Smart

Factory, Smart Supply Chain and Smart Lifecycle. For each category other sub-classes

were individuated. In the Smart Factory, the maintenance (considering preventive and

predictive grouped) represents the class with most of the applications (39% in LEs and

24% in SMEs), ensuring equipment's functionalities and status monitoring. The

supervision of the equipment is completed with application in energy management, a

relevant 18% of the applications reported in the SMEs survey. Production optimisation

ensures that existing resources are optimally exploited, taking into account their

limitations and constraints. Material handling and quality control are further classes

of investment that contribute to maximize the level of service to customers in terms of

speed, flexibility and compliance with delivery dates. Digital Twin and Work safety

complete the panel of applications in the factory boarders. In a broader perspective,

improvement in Smart Supply Chain are achieved including asset tracking and

parameter monitoring, logistics asset management, warehouse asset tracking and fleet

management. To conclude, in the Smart Lifecycle, the following areas were identified:

optimization of the new product development process, end-of-life management and

supplier management in the lifecycle management process. In Figure 3.2, we showed

the results of the LEs’ survey broken down for in the difference sub-classes of the Smart

Factory field.

Indicate, for each IoT project for Industry 4.0 launched by your company, the

state of progress (pilot project and executive project). Sample size LEs, 90 (a total of

212 application).
 3.1 Survey Analysis 45

Figure 3.2: Results of Q2.2 related to Smart Factory - Sample size LEs, 90. Sample SMEs,
31 – Indicate the category (and sub-classes) for each IoT project for Industry 4.0 launched by
your company (pilot project and executive project). The graph indicates how the applications
inside the Smart Factory category are distributed between the sub-classes.

In the IoT application presented above, additional service could be embedded,

varying from information services, based on sending real-time notifications in case of

predefined events, to pay-per-use logics, services that require payment for related

items (i.e. machinery, logistics assets) based on actual use and not at the time of

purchase. Energy management services have a large presence as well, and they include

the receipt of reports/alerts with the analysis of energy consumption trends and

personalized advice on how to save money. A service related to preventive or

predictive maintenance was indicated by one in two respondents, representing with

information services the category more reported. At the opposite, insight the LEs

sample, no demand has been found regarding insurance coverage based on actual use

of related objects.

Objectives

Both the two surveys testify efficiency as the key driver that motivate the

adoption of Industrial IoT solutions. This result is in line with the near-term macro

period illustrated in chapter 1, in which efficiency has a leading role. Effectiveness is

 46 Chapter 3. Industrial IoT Research

the next most indicated driver for LEs, while the results of SMEs show the company

image improvement, nominated 40% of the times by the respondents. Other relevant

drivers, with a percentage higher than 15% present in one of the two surveys, are (a)

the willingness to experiment with new solutions, (b) the adaptation to legal or

regulatory obligations, and (c) the creation of a database rich with data made available

by the connected objects (Figure 3.3). Regarding benefits, they are rarely evaluated in

SMEs (in 86% of the companies); this percentage grows to 95% adding who evaluated

benefits only qualitative. In the LEs survey, results are more desirable, but, as a

counterpart, the several following calls testified the missing of a comprehensive

analysis of return on investment also in this group. A comparison of expected costs

and benefits is confirmed to be a challenging task.

Furthermore, it was asked to indicate the use of the data collected. The answers

vary from a profitable use (the data collected, both raw and reprocessed, have been

used profitably by the company) to a data collection ending (data are collected without

extracting real value, the company has not used them). The analysis of the results

inside LEs, shows that 45% of the companies have not used /little used the data

collected. The definition of appropriate strategies to enhance the value of collected

data is a significant issue for companies. Regarding this point, five data valorisation

schemes could be formalised (Osservatorio IoT - Politecnico di Milano, 2017): (1)

process optimisation, (2) new generation of product/service, (3) Product/service

customization, (4) direct data monetization and (5) Advertising & Commerce.18 Aptiv,

18 The five data valorisation schemes are reported in more detail:

1. Process optimization: the data collected are used to improve the internal
processes of the companies themselves, with positive effects in terms of
increasing efficiency (reducing time and costs) and / or effectiveness
(customer service);
 3.1 Survey Analysis 47

a multinational company in the automotive sector for highly technological solutions

with a manufacturing plant in Italy, belong to the first group. It controls the

consumption of all the machines in the plant, obtaining the visibility to tackle waste.

In their agenda, a predictive maintenance application to reduce the time to repair and

the downtime is marked for the next years.

Figure 3.3: Results of Q3.1 - Sample size LEs, 61. Sample SMEs, 47- What were the main
objectives that led the company to launch IoT projects for Industry 4.0? Up to three options can
be entered.

2. New generation of product / service: data on the use of connected objects by

customers can be exploited in the process of developing improved versions,
to reduce the most common defects and improve usability;
3. Product / service customization: a company can decide to customize its offer
based on the data collected, in order to be able to better meet the specific
needs of customers;
4. Direct data monetization: a company can decide to sell the collected data to
interested third parties, thus generating a new source of revenues;
5. Advertising & Commerce: the possibility of proposing targeted advertising
is extended to the world of connected objects, in which therefore the source
of information is the habits that occur in the real world.
 48 Chapter 3. Industrial IoT Research

Barriers

Lack of knowledge of the Industrial IoT application/ lack of internal skills is the

major recognized barriers both in the SMEs that LEs survey, indicated by more than

half of respondents. Concerning to this barrier, it was asked what companies have

done to cope with the gap in knowledge: more than half of the respondents have not

taken any specific action to try to reduce it (even though the lack was recognized).

Who is tackling this issue proceeds to train internal staff, acquire new dedicated

professional skills, and work with external consultants with specific expertise. At the

same time, hardly anyone experienced the miss of new professionals on the market or

external consultants. Besides, survey respondents cite several other potential

challenges and concern:

 Lack of understanding of the value of solutions;

 Difficulties in integrating new and old hardware and software;

 Poor availability of economic resources;

 Internal resistances;

 Lack of suitable products/ IoT technologies not yet mature;

 Cybersecurity issues;

 Privacy issues;

 Difficulty in accessing government incentives;

 Lack of suppliers.

In Figure 3.4, the results of the surveys regarding this topic are represented.

In line with the barriers cited 'Technoform Bautec Italia' (an Italian medium

company, specialist for standard and tailored plastic solution) reports the difficulty of

collecting data, managing them in real-time and evaluating a return of the investment.
 3.1 Survey Analysis 49

Figure 3.4: Results of Q4.1 - Sample size LEs, 88. Sample SMEs, 525 - What are the
barriers (internal and external) that in your opinion, can slow down or prevent the start of IoT
projects for Industry 4.0? Up to three options can be entered.
 50 Chapter 3. Industrial IoT Research

Investment Dimension and Future

Orientation

How much are companies investing in Industrial IoT solutions? Among those

who activated a project, one in four allocates less than 5% of the company's turnover.

Subsequently, an optional question asked more specific details of the economic

dimension: results report that investments included in a range from 20k€ to 2mln€. In

this scenario, the ability to exploit the government incentives needs to be understood.

In the narrow circle of the SMEs that develop at least one application, the importance

of exploiting the incentives can be considered underestimated: only 24% of the

respondents declare that being able to access incentives is a substantial driver. To

conclude the as-is situation, the most used government instruments highlighted by the

survey are the increased value (+ 250%) of the depreciation for the assets and the tax

credit for incremental R&D expenses.

The attention for Industrial IoT application will increase in future. As shown by

the surveys, approximately 70% of the companies will confirm investments related to

the specific field where they already developed a project. The total number of

applications will increase due to new adopters. An even more significant growth will

arise from the adoption of services related and reported in Figure 3.5 applications and

services about preventive and predictive maintenance.

Regarding the maintenance services, a significant boost of the predictive part can

be observed. The question 5.1(Figure 3.5) investigated the service activated in the past

project and the one willing to be used. The answers are scaled for the number of

respondents to the one regarding the future. Sample base regarding the past project is

59 companies. The sample base regarding future project is 83. This trend of shift

towards services was analysed with a case study, Fabio Perini S.p.A., a company

present in the LEs survey and then interviewed (Antonio Mosca – Head of Digital). In

the survey, this company reports a level of knowledge of 10 over 10, active Industrial

IoT application (internal processes efficiency) and a profitable use of data. The

prominent relevance of this company was not considering it as part of the Demand as
 3.2 Case study – Fabio Perini S.p.A. (Körber Tissue) 51

solution adopter but of the Supply, as a successful example of service delivery for

external clients.

Figure 3.5: Results of Q 5.1 Sample base regarding past project, 59. Sample base
regarding future project, 83 - (the answers are scaled for the answers received regarding the
future) – (future) What additional services enabled by the Internet of Things technologies is the
company interested in activating inside the factory? (present) Up to three options can be
entered. (Have the IoT applications for Industry 4.0 that you have launched included additional
services?).

3.2 Case study – Fabio Perini S.p.A.

(Körber Tissue)
Established in Italy, in 1966 by the inventor/entrepreneur Fabio Perini, Fabio

Perini S.p.A. is a mechanical engineering company. The company is specialized in the

manufacturing and design of industrial machinery for the papermaking industry and

the tissue converting industry. The company has eleven overseas divisions located in
 52 Chapter 3. Industrial IoT Research

three continents (Europe, America and Asia). Thanks to many original design patents,

in 20 years the company Fabio Perini transformed itself from a one-man business to a

multinational enterprise, covering up the 75% of the worldwide market of the tissue

converting machinery. Since 1994 Fabio Perini S.p.A. is part of the German technology

group Körber and belongs to the division Körber Process Solutions. The company

employs 616 workers and has a turnover of 163.5mln€ (AIDA, 2019).

Fabio Perini S.p.A. was the first mover in its specific sector. Its primary service

solution consists in to optimise the OEE of the client's whole plant. This result is

obtained by managing data in the cloud and elaborating them in its internal

performance centre, associated machine experts within advanced technology (data

collection, measuring, analyse, reporting, machine learning). At first, Fabio Perini

promoted this service with SMEs, promising insights and pillars for OEE

improvement, and in a year since October 2018, it has been possible to connect a

thousand machines around the globe. This outcome was achievable by having

investments already in place: a relevant part of the sensors (measuring vibration,

temperature, etc.) and a useful front-end monitor, previously inserted as after-sale

service and then proved to be very useful for Industrial IoT application. With SMEs,

no barriers were reported, and in case the client is worried about the cloud solution, it

is possible to switch on the premise. In a second phase, also LEs, initially reluctant

already having their internal team at their disposal, begin to buy Fabio Perini's services

to leverage on its Tissue Performance Centre.

The company is proceeding in its digital transformation, aiming to become the

leading provider of complete solutions for the tissue ecosystem, to optimize the

Overall Equipment Effectiveness of the entire lines. A set of smart software are

proposed, such as a digital solution for predictive maintenance, to identify deviation

in process functionality and plan maintenance operations, or virtual production

supervisors, to measure and keep under control, in real-time, the most critical

properties of tissue and finished product. Technology devices are proposed as well:

wearable glasses provide technicians with augmented reality that allows for remote,

interconnected sharing and viewing to rapid troubleshoot problems, therefore the

possibility to resolved issue remotely with no need to send a technician on site.

 3.3 Remarks and Checkpoint 53

The last significant point to talk concerns servitization, a new business model for

this specific industry by which customers are guaranteed help from Fabio Perini in

raising their efficiencies and expanding their production capacity without adding

capital expenditures. Pay-per-Performance and Pay-per-User are the new payment

methods that Fabio Perini is experimenting (Körber Tissue, 2019) with related

characteristics. The former lets the client pay by sharing the economic value, agreeing

with Fabio Perini to raise and maintain the operational efficiency to a pre-defined level.

The latter let to pay for what they use with related benefits from no upfront capital

cost, flexibility in payments and viable long-term business opportunity in place.

3.3 Remarks and Checkpoint

Recalling the definition proposed in the second chapter, the Industrial IoT “[…]

enables real-time, intelligent, and autonomous access, collection, analysis,

communications, and exchange of process, product and/or service information in the

industrial environment to optimise overall production value” (Boyes, Hallaq,

Cunningham, & Watson, 2018). The answer to the question 'Why now?' was given by

Porter & Heppelman (2014): "An array of innovations across technology landscape

have converged to make smart, connected products technically and economically

feasible". As we have seen in the surveys, much has already been achieved. Looking

to the next years, the field of maintenance results to be highly attractive, as highlighted

by the number of direct applications and by the relevant part of services. Maintenance

that in its noblest form is expressed in predictive maintenance acquires more

exquisitely the features of the Industrial IoT when the vendors manage the

maintenance process in a centralized way, making predictions much more accurate.

Every client will manage this information to gain benefits also from the coordination

and optimisation of production and logistics. Predictive maintenance has always

existed "from the moment a man puts his ear to the machine, and he foresees that it

will soon break down", and now it can be entrusted entirely to the vendor of the
 54 Chapter 3. Industrial IoT Research

machine. Vendor has the structural knowledge base, and with Industrial IoT, he can

exploit the data arriving in real-time from the various interconnected machines of the

clients. Outsourcing of maintenance can be assessed. The sharing of operating data

externally to the company could represent a barrier, and a change in mentality may be

necessary, improving the existing supplier relationship. For the demand side, the

benefits lie in optimising production with cost savings and increased productivity. On

the supply side, a competitive advantage can be unlocked considering the scalability

potential of predictive maintenance, as well as being able to trigger mechanisms for

improving the machines by knowing the problems in their actual usage (in the field of

smart lifecycle application). The investment in IoT has to be seen from the perspective

of a path that will lead from time to time in adding solutions to obtain ever-greater

overall advantages.
 Chapter 4

4 Literature Review

The literature review is a crucial point for any academic project. In this chapter,

a literature review to understand the benefits of predictive maintenance is performed

to deepen the theme and discover possible gaps on which perform further analysis.

First, the chapter presents the steps and the analysis conducted to obtain the final

sample of papers. Then it shows a descriptive analysis to provide the reader with

essential information about it. Categories have been selected based on documents

content and our goal to get insight from the papers. A specific analysis is conducted

on them to describe the information included. Finally, gaps are presented and

discussed.
 56 Chapter 4. Literature Review

4.1 Introduction to Literature

Review
The first essential step to carry out any academic project on a particular topic is

to understand the level of knowledge achieved in that area completely. Therefore, the

literature review assumes a crucial role in this context (Webster & Watson, 2002).

Indeed, it aims to identify what has been written on a topic, reveals any interpretable

trends or patterns, identifies issues or questions that require further investigation to

generate new frameworks and theories where there are gaps in the literature (Paré &

Kitsiou, 2017).

In this chapter, we analyse the current literature in the area of predictive

maintenance that represents the focal point in the Industrial IoT era (Kaur, Selway et

al., 2018). For our analysis, we decided to adopt the content analysis approach as

proposed by Seuring & Gold (2012). We chose this approach because the content

analysis allows exploring the manifest content of texts and documents and, at the same

time, makes it possible to excavate the latent one underlying the meaning of terms and

arguments. “It is a specific strength of content analysis that this method can combine

qualitative approaches retaining rich meaning with powerful quantitative analyses,

where it seems meaningful for the analyst” (Seuring & Gold, 2012).

The chapter is organized as follows. The first section presents the methodology

for the research and the selection of the papers. The second presents the descriptive

analysis (i.e. the year of publication and research methods), the third exhibit the

selection of categories for the analysis of the theme deriving from the review and the

following description of these categories. In the final section, we report the conclusions

and highlight the potential directions for future research in this field.
 4.2 Material Collection and Selection 57

4.2 Material Collection and

Selection
The first step of the literature review analysis is the collection and selection of the

material (Seuring & Gold, 2012). Here we will present the identification process and

the subsequent steps that led us to a delimitated papers selection, as well as the

formalized approach that we applied to validate the sample we obtained to eliminate

possible subjective bias and enhance reliability.

Articles Collection

According to Seuring & Gold (2012), the first step is to define the unit of analysis.

We decided to include in our literature sample English-speaking peer-reviewed paper

and conference paper on Industrial IoT in manufacturing. Given the novelty of the

topic, we decided not to limit the years of publication; rather it was considered an

important issue to understand the dawn of the literature and the beginning of the

ferment. For compiling the paper sample, we fulfilled a literature search in three of the

primary source of online literature databases: Scopus, ISI Web of Science and Google

Scholar. It is important to underline that, to have good coverage of the research field,

many other references have been investigated with the same keywords as well. The

search was applied to the title of the articles, the abstracts and the keywords in order

to conduct the most comprehensive research possible without including papers not

centred on the topic.

The search was carried out based on different keywords, such as Industrial IoT

or Industrial Internet. However, the sample size obtained with these keywords and

the extent of the topic is so wide that a more narrow analysis was required (more than

six thousand paper has been found only on Scopus). Moreover, because of the lack of

clarity of the terms as we have identified in the previous chapters, we have also

included the word Industry 4.0. In order to refine the search of the published papers,
 58 Chapter 4. Literature Review

we decided to delimit the topic to the maintenance of manufacturing assets. Therefore,

we combined the words Industrial IoT, Industrial Internet and Industry 4.0, with

manufacturing, maintenance, predictive, and benefits also complementing the sample

by cross-referencing them (Seuring & Gold, 2012).

Article Selection

From the database investigation, 655 results have been found, downloaded and

organized in excel (Polanin, Pigott, Espelage, & Grotpeter, 2019). The second step was

to eliminate all the duplicated papers. Since the research words are a bit redundant

(for example, the word Industrial IoT appears in all the searches), the number of

duplicates was around 20%, and only 508 papers were kept. Then, in step three,

authors screened the title of all these papers deciding whether to include or exclude

them from the selection. Both the reviewer conducted the screening independently in

order to foster objectivity (Paré & Kitsiou, 2017). For this step, we established some

exclusion criteria: (E1) studies not explicitly focused on IoT; (E2) studies that do not

address the manufacturing domain; (E3) studies that have not been published in books

and book chapters or well-known international conferences. After this screen, 155

papers remained.

For the last step of the screening phase, we decided to read the abstracts and give

a quick look at the text of these studies in order to see if the focus of the selected papers

was in line with ours. According to Polanin et al. (2019), we conducted this screening

separately using an abstract screening tool with questions. Given the advanced state

of the procedure, we preferred to limit our assessment (yes/no/unsure) to two

questions: “Is the focus of the paper on predictive maintenance?” and “Is the focus of

the paper not on a specific analytic aspect of predictive maintenance?”. At the end of

this process, the disagreement was around 20%, and a reconciliation occurred

(Polanin, Pigott, Espelage, & Grotpeter, 2019).

According to the screening criteria mentioned, from the initial sample of 508

articles, 57 studies remained, ready for an in-depth review. The final number of articles
 4.3 Descriptive Analysis 59

has not been established a priori, but is the result of a careful process of combined

research and screening. Eventually, the material was categorized following two steps

in accordance with Seuring & Gold (2012). A descriptive analysis was conducted first

(i.e., distribution over the time, nationality) and then categories based on the content

following an iterative process that will be explained later in the chapter.

4.3 Descriptive Analysis

The reviewed articles are categorised according to the research methods used by

the authors following the typologies used by Meixell and Norbis (2008) and Perego,

Perotti, and Mangiaracina (2011). A quick overview of the definition of the different

methods follow:

 Literature Review: the paper collects and analyses studies and papers

previously published on a given topic, presenting them in the most

objective way possible.

 Conceptual Framework: it models a phenomenon through the use of causal

maps, matrices, diagrams;

 Analytical Method: a scientific analysis methodology that allows to solve

a problem through a well-defined mathematical calculation procedure;

 Simulation: a model of reality that allows to evaluate and predict the

dynamic unfolding of events or processes following the constraints

imposed by the analyst or user;

 Case study: empirical analysis that investigates a contemporary

phenomenon in its real context. It is often aimed at verifying a theoretical

model and illustrating its applications;

 Survey: a statistical survey aimed at a predetermined sample of people

and at collecting a set of opinions, preferences and behaviours in relation

to the purposes it is conducted. It can be structured, semi-structured and

 60 Chapter 4. Literature Review

unstructured, depending on the margin of discretion granted to the

interviewee;

 Action research: it is essentially a research through action. It requires the

direct involvement of the researcher in the action process. The purpose

is to analyse a field-related practice, to solve a particular problem and to

introduce guidelines for best practice.

Overall, about half of the works are a conceptual framework, whereas the others

are based on empirical research (e.g. case studies, surveys). Only a few of them (7 out

of 57) are analytical or simulation models. The distribution of the analysed papers

shown in Table 4.1. It is important to mention that some works present more than one

research method. For example is the case of Farooq, Bao et al. (2020), they present a

predictive maintenance model to improve the reliability of the system and then

demonstrate the feasibility of the proposed model in a case study.

Conceptual frameworks account for 49% of the total papers (or 28 out of 57)

(Terrissa, Meraghni et al., 2016; Sahal, Breslin et al., 2020; Garcia, Costa et al., 2020;

Bengtsson & Lundström, 2018). For example, Chehri & Jeon (2016) propose an efficient

framework for the Industrial IoT, providing a state-of-the-art approach for industrial

applications. Another example is given by Kaur, Selway et al. (2018), who present a

framework to extend previous open standards to achieve predictive maintenance. The

article first presents international standards for condition-based maintenance, then

presents an industrial interoperability ecosystem (OIIE) that allows devices and

Table 4.1: Research method summary

 4.3 Descriptive Analysis 61

systems to communicate effectively in both inter and intra-enterprise contexts using a

variety of standards, models of data and exchange protocols, thus improving

predictive maintenance.

Empirical studies are based on case studies, surveys and action research and

represent 19%, 5% and 4% respectively). While the surveys mainly aim to understand

the level of knowledge of the Industrial IoT technology in predictive maintenance,

trying to understand the opinions regarding the possible future implications, the case

studies are mostly empirical demonstrations to validate different types of models for

predictive maintenance. Roda, Macchi & Fumagalli (2018) are representative of the

first type. They interviewed nine Italian companies to investigate the vision of the

future of maintenance in the Industrial IoT era and “to show empirical evidence on

how manufacturing companies are approaching the digital transformation process of

maintenance.by interviewing.” For the case studies, it is pertinent the paper of Ruiz-

Sarmiento, Monroy et al. (2020), “which describes a predictive model based on a

Bayesian filter […] to estimate and predict the gradual degradation” of the machineries

involved in the production of steel sheets and allow operators to make informed

decisions regarding maintenance operations.

Quantitative model (analytical model or simulation count 11% and 2%

respectively) are relatively limited, although progressively increasing, and are

generally associated with economic evaluations of the potential benefits that would be

obtained from the application of the Industrial IoT to predictive maintenance. For

example, He, Han et al. (2018) propose a cost-benefit analysis model based on the state

of reliability of the machine concerning the task for which it is assigned. More recently,

Huang, Chen et al. (2020) published a study on a revenue-sharing model to

demonstrate the convenience of purchasing the maintenance service directly from the

vendor, service that requires the Industrial IoT for data transmission.

Finally, literature reviews represent 11% of the overall sample and have no

specific relation one to the other. For example, Del Ser, Galar & Sierra (2019) provide

a comprehensive review in the field of data fusion and machine learning for industrial

prognosis. While Bousdekis, Lepenioti et al. (2019) made a review of the literature on
 62 Chapter 4. Literature Review

the field of decision making in predictive maintenance in the context of intelligent

manufacturing.

Regarding the year of publication, we decided not to limit the period of the

research, knowing about the novelty of the theme, to figure out when scholars started

to work on it. We did not find any significant studies related to our research topic prior

to 2015. The distribution of the reviewed papers, which were published between 2015

and 2020, is shown in Figure 4.1. The number of papers published remained stable for

in the biennium 2015 – 2016. From 2017 an incredible growth path started, publications

doubled each year, reaching their peak in 2019, revealing the greater attention that to

the theme. The majority of the papers concentrate in the two years 2018, 2019: 35 papers

out of 57. A final consideration worth to be done in this concern. Since this review was

performed in early 2020, the publication number for this year is lower. However,

considering that we are only in the first quarter of the year, we expect the previous

growth rate to be confirmed again for this year19.

Focusing on the country of origin of the published papers, the most significant

contribution is given by Italy and China, which together represent 22% of the sample.

In it also interesting to notice that European countries together provide the large

majority of the papers, which provide together more than 68% of the papers.

Figure 4.1: Paper distribution per year

19 There is also to consider that publications in 2020 may have been slowed down
by the recent problems caused by the sanitary emergency.
 4.4 Category Selection 63

4.4 Category Selection

In their guidelines on how to conduct a rigorous content analysis, Seuring & Gold

(2012) particularly highlight the central role of defining categories to classify the

reviewed material. Following their suggestion, we decided to use a two-step-process

(deductive and inductive). From the first screen of the selected literature, we

deductively defined some categories based on some authors’ classification to refine

and introduce new ones in a subsequent inductive approach. However, the novelty of

the themes covered required much effort to adapt and create ad-hoc categories, since

the existing ones were too general or too specific in a given area.

In our analysis, we were interested in understanding the role of Industrial IoT in

predictive maintenance, the improvements and transformations it can produce from

an economic and productivity perspective, and the problems that hinder its adoption.

Therefore, we grouped the analysis around four key themes (see Table 4.2):

1. The approach with which authors deal with the theme of industrial IoT

(Authors’ approach): It is important to know on which aspect of the Industrial

IoT in predictive maintenance the authors focus on, what the purpose of the

article is, in what terms the topic is analysed. Is the author focusing on future

trends of predictive maintenance, on presenting a structure of the Industrial

IoT framework, or on business organisational change taking care of the

economic aspects?

2. Benefits achieved after adoption (Benefits): This variable measure how much

each paper consider the benefits that predictive maintenance can generate on

production with the implementation of Industrial IoT solutions;

3. Criticalities, barriers to adoption and risks (Barriers & Risks): Even with the

support of Industrial IoT, predictive maintenance still presents barriers and

risks that limit its adoption and diffusion. This variable aims to assess how

much each paper deals with this problem and whether the criticalities come

from a technical or a more business perspective;

4. Business model transformations (Servitization): The Industrial IoT in the field

of predictive maintenance also presents great opportunities to shift towards

 64 Chapter 4. Literature Review

new business models based on services. This variable aims to assess how

much and in which terms the literature considers this topic.

Table 4.2: Analytical categories and their definitions

4.5 Evaluation and Discussion of the

Review
The sample of literature review papers on the adoption of industrial IoT in the

context of predictive maintenance was analysed based on the categories presented in

the previous chapter. The results are presented and discussed to understand where

the literature is focused on with an objective perspective. Gaps are eventually

presented.

Before going into detail and explore all the variables carefully, it is necessary to

generally analyse how the literature refers to the theme of predictive maintenance.

Looking on the web libraries for the key words previously mentioned (Industrial IoT,

Predictive maintenance, Industry 4.0…), thousands of results are generated, and all of

them were subsequently refined using the screening process previously described.
 4.5 Evaluation and Discussion of the Review 65

However, it is interesting to notice that some papers that still satisfy all the screening

criteria and have their focus on predictive maintenance, present the Industrial IoT in

manufacturing with a much more general focus. These papers represent only 21% of

the sample. For instance, Khan, Ateeq et al. (2017) provide a holistic view of the

different stages of IoT projects to see how it generates business value for an

organization focusing then on predictive maintenance as a use case. Similarly, Kans &

Ingwald (2016) trace the evolutionary path of services up to Service Management 4.0

whose purpose is to create value for the customer. In this context, predictive

maintenance is reported as one of the main services that create this value.

Authors’ Approach

The first proposed categorization divided the literature according to the different

themes. It was decided a strict division that identifies the paper according to its main

characteristic, bearing in mind possible interdependencies. Industrial IoT framework

represents the most tackled argument, 27 papers over the 57 selected (47%) reporting

the architecture and guideline details for the Industrial IoT application enabling

predictive maintenance. The variable Business decisions is represented by 18 papers

(32%) that point out the focus on the economic aspects and organizational impacts

beyond the Predictive Maintenance technical application. Finally, in the variable

Factory field 12 papers (21%) have been categorized, gathering all those papers brings

survey, perspective and trends of the maintenance sector. The main topics covered in

these three classes will be detailed below.

 66 Chapter 4. Literature Review

Industrial IoT Framework

Different architecture models are presented20, a multilayer structure

characterizes each model, and despite the different name of the individual level of

technical design, the described architecture is similar. As reported in Strauß, Schmitz

et al. (2019) at the lowest level, measured variables and the sensors suitable for

monitoring must first be determined. The selection of the interfaces for connecting the

sensors is addressed in the gateway or communication level, necessary in order to

reduce the data volume, to pass only the selected and extracted characteristics and not

all measured data to a central cloud. Concepts of using edge computing reduce data

volume, save computing power in the cloud and allow real-time data analysis (De

Leon, Alcazar et al., 2017). In the central cloud, data from various sources is stored,

aggregated and abstracted to enable data analysis. Finally, the top layers of the

reference architectures describe an integration into the business processes, i.e., the

integration of machine learning models within the process organization of the

maintenance division (Strauß, Schmitz et al., 2019). Due to the absence of largely

adopted open standards, the Industrial IoT architecture must be an interoperability

solution that enables devices and systems to communicate effectively in both inter-

and intra- enterprise contexts using a variety of standards, data models, and exchange

protocols (Kaur, Selway et al., 2018). The set of features necessarily moves from Real-

time capability to Modularity and Security (Wan, Tang et al., 2017). In a broader

perspective, Industrial Iot-based manufacturing aspires to be an ecosystem able to

manage the 5Vs of big data (volume, velocity, variety, value, and veracity) concerning

data ingestion, management, analytic, and visualization layers (Yu, Dillon et al., 2020).

Lastly, papers focusing more on application case regarding Predictive maintenance are

classified in this variable, i.e., proof of concept, case study and simulation (Bergonzi,

Colombo et al., 2017; O’Donovan, Leahy et al., 2015; Ayad, Terrissa et al., 2017; Chacada,

Barbosa et al., 2019).

20 In the first chapter, we discussed the several reference frameworks and among
those presenting we adopted the widely accepted three-level model which includes
the edge, the platform and the enterprise levels.
 4.5 Evaluation and Discussion of the Review 67

Business Decision

A possible evaluation of the cost of maintenance is approached (Osako,

Matsubayashi et al., 2019; Drewniak & Gabriś et al., 2019; Tsao, Lee et al., 2019) and

different case studies with benefit considerations are reported. Since failure of critical

assets has been rated as the most significant risk to operational performance,

manufacturers increasingly see maintenance no more as a necessary evil but as a

strategic business function. The majority of manufacturing companies lack in

reliability engineers, and even more in data scientists. Investment in data analytics

roles or the introduction of new roles that bring complementary skills to those inherent

to maintenance processes is essential. The success depends on skills and knowledge,

and the promotion of cross-functional teams is suggested (Bousdekis, Apostolou et al.,

2020). From a broader perspective, the diffusion of a culture of continuous

improvement inside the organization needs to be considered: the digital

transformation is seen as a journey towards Industry 4.0 and not as a one-shot

investment. Change management and a corporate governance strategy would be

highly relevant. In this journey, the maintenance function is transforming its role in

order to better support value creation (Jasiulewicz-Kaczmarek & Gola, 2020). The

industrial-scale deployment of Predictive Maintenance involves many other aspects

and impacts various sectors of the workplace, like logistics, occupational health, safety

& environment, design and top management. Above all, the objective of supporting

decision making to act on the physical systems optimally is pursed (Compare, Baraldi

et al., 2019). Strictly related to this variable, two different categorizations will be treated

subsequently: Benefits and Servitization.

Factory Field

New trends and techniques in the field of predictive maintenance arose,

representing alternatives to traditional management policies relying on visual

inspection and early identification of signs of abnormalities during operation.

Predictive Maintenance, frequently linked in papers to PHM (Prognostic & Health

Management) (Baur, Albertelli et al., 2020), is the core of this new change and it heavily
 68 Chapter 4. Literature Review

relies on sensor technologies21. Nowadays, based on Predictive Maintenance, the

cognitive maintenance is getting more and more attention: a concept of maintenance

completely integrated in industrial plant management that will gradually force an

interaction between maintenance and management of production processes solutions

(Drewniak & Gabryś, 2019). The adjective cognitive stays for advanced technology at

the intersection of big data, machine learning, and artificial intelligence analytics.

Strictly related to Predictive Maintenance, it is reported e-Maintanance, a management

concept whereby assets are monitored and managed over the Internet. Similar field to

Remote Maintenance and Management Systems (RMMS), which makes possible to

execute Predictive Maintenance works especially in isolated or hazardous locations.

Data visualization, digital twins, and augmented reality are further new technologies

and concepts that provide companies with highly advanced efficient systems in

maintenance (Noureddine, Solvang et al., 2020), new capabilities for remote and

ubiquitous maintenance. Bokrantz J., Skoogh et al. (2017) enclosed these concepts

inside the definition of Smart Maintenance:

“An organizational design for managing maintenance of manufacturing plants

in environments with pervasive digital technologies […] a multidimensional concept

constituting of the four dimensions of data-driven decision-making, human capital

resource, internal integration, and external integration. [...] it aims to achieve effective

and efficient decision-making and responsiveness to internal and external

components.”

Lastly, Bengtsson & Lundström (2018) highlighted the risk in the future of

leaving basic maintenance concepts and management underdeveloped or even

21 Sensor devices constitute a fundamental layer, directly interact with the

environment by measuring variables of interest and by processing acquired data to
perform onboard analysis tasks. Sensor techniques can be broadly categorized into (a)
process parameter measurements, (b) vibration analysis, (c) oil analysis, (d)
thermographic analysis, (e) acoustic analysis and (f) other Predictive Maintenance
techniques (Sule, 2015). Sensor technologies have evolved concurrently to sensor
fusion techniques (Turner, Emmanouilidis et al., 2019).
 4.5 Evaluation and Discussion of the Review 69

unimplemented, putting too much trust in technological advancement. In order to

achieve a “total maintenance effectiveness”, the combination of the “new” with the

“old” is requested.

Benefits

The papers addressing the benefit topic, reporting discussion and in-depth

analysis, count for 26% of the considered literature (15 of 57). Overall, the literature

focus has been mainly on the technical deployment of Predictive Maintenance, and

further contributions to define economic benefits and IoT investment justification are

still missing (Drewniak & Gabryś, 2019). Regarding the benefits analysed, two main

classes have been highlighted: operational efficiency and cost reduction. Other aspects

(i.e. safety) will be discussed after the focus on these first two main perspectives.

Operational Efficiency

The possibility to rely on Residual Useful Life computation (predicting failure of

the equipment inferred by degradation pattern) maximizes the equipment availability

and reliability (Sezer, Romero et al., 2018). A better scheduling of maintenance could

be pursed, achieving an improvement in key sector indicators, i.e. Mean Time Between

Failure (MTBF) and Mean Time To Repair (MTBR). The historical trade-off regarding

Preventive Maintenance (PvM) of trying to maximize the useful life (lifetime

extensions) of the production equipment while avoiding unplanned downtime and

minimizing planned downtime, is overcome with Predictive Maintenance (Poosapati,

Katnevi et al., 2019). Furthermore, a significant part of the value of Predictive

Maintenance would come from improvements in production quality and a better

orchestration of maintenance-related processes, i.e. a better coordination in the

logistics for maintenance operations.

 70 Chapter 4. Literature Review

Cost Reduction

In the optic of just-in-time maintenance equipment, unnecessary work orders are

eliminated with a gain in time efficiency and a reduction of costs per parts and labour.

The warehouse management could reduce the stored spares setting a just-in-time

logistics. As often quoted phrase is the following: “The right part in the right place, at

the right time”.

Other reported benefits regard the safety of people, which could represent a key

point for specific equipment and manufacturing workspace; few papers explicit the

environmental benefits that could be ensued. Poosapati, Katnevi et al. (2019)

distinguished different factors: prevention of environmental damage, emissions

reduction and land conservation, energy consumption reduction and energy savings

or efficiency.

Percentages of improvement after adopting predictive maintenance could be

considered as a useful benchmark (Poosapati, Katnevi et al., 2019; Sahal, Breslin & Ali,

2020; Bokrantz, Skoogh et al., 2020):

 Savings in maintenance costs up to 30%;

 Up to 75% less failures/breakdowns;

 Decrease of the total machine downtime from 30% to 50%;

 Extension of the operation life from 20% to 40%.

These percentages testify the possible benefits of the development of IoT-enabled

Predictive Maintenance for Industry 4.0, but they did not mention the as-is situation

or other possible maintenance strategies. Even if CM and PvM presented structural

limits (i.e. PvM could not manage hysteresis of the processes), a comparison with these

most-used maintenance strategies needs to be performed. Maintenance cost models

must be developed to evaluate the economic benefits of predictive maintenance; an

example of this attempt is the dynamic cost-oriented predictive model (Tobias &

Yanpei, 2019). As reported in Drewniak & Gabryś (2019), the identification of the

components that could be eligible for Predictive Maintenance is the first step to be

taken to address the issue of selecting the best maintenance strategy in a system.
 4.5 Evaluation and Discussion of the Review 71

Barriers and Risks

The literature reports several papers that address the adoption issues

(O’Donovan, Leahy, & Bruton, 2015; Yu, Dillon et al., 2020; Müller & Däschle, 2018;

Compare, Baraldi, & Zio, 2019; Nikolic, Ignjatic et al., 2017). Overall, the behaviour of

the companies seems to be influenced by both the characteristics of the company itself

and by the reliability of the technology. On the one hand, some main socio-economic

barriers have been identified, such as the ability to demonstrate to the customer the

greater value and the ability to it consequently (Müller & Däschle, 2018). On the other

hand, some technical barriers, which discourage many players from adopting such

solutions are reported. We focus on this distinction to examine more in depth the main

obstacles that prevent companies from adopting Industrial IoT solutions for predictive

maintenance.

Socio-Economic Barriers

According to the examined literature, the most recurrent economic barrier to

predictive maintenance adoption with an Industrial IoT architecture is the cost of the

investment. Adu-Amankwa, Attia et al. pointed out that the initial investment into the

technological infrastructure can be significant when retrofitting to the factory’s

original design and Strauß, Schmitz et al. (2018) reported that “75% of the industrial

companies would not be willing to invest more than 500 EUR to digitize their

machine”. Companies are so reluctant to make such investments because of the

intrinsic difficulty in establishing the payback in the digital transformation of

maintenance (Roda, Macchi & Fumagalli, 2018; Compare, Baraldi & Zio, 2019).

Other authors address the same monetary issue from another point of view. They

set their perspective in the supply of industrial machines and figured out that

suppliers may face some challenges as well in integrating digital solutions for

predictive maintenance in their offer. This aspect is caused by the difficulty of the

customer to recognize its value and the subsequent impossibility of the vendor to
 72 Chapter 4. Literature Review

capture it with the traditional business models (Müller & Däschle, 2018; Omar,

Minoufekr & Plapper, 2019).

Other socio-economic barriers that authors reports are the lack of organization

culture (Omar, Minoufekr & Plapper, 2019; Roda, Macchi & Fumagalli, 2018) and lack

of skilled workforce with the subsequent cost of training the personnel to interpret

data to schedule maintenance accordingly (Adu-Amankwa, Attia & Janardhanan,

2019; Baur, Albertelli & Monno, 2020; Roda, Macchi & Fumagalli, 2018).

Technical Barriers

Technical barriers deal with the reliability and security of the technology. The

Industrial IoT “enabled factories to rapidly scan many millions of unstructured data

items in different formats from a multitude of diverse sources” (Yu, Dillon et al., 2020).

This capability creates the opportunity to reduce uncertainty in the prediction and

allow a better schedule on the one hand (Diez-Olivan, Del Ser et al., 2019), but a greater

vulnerability to cyberattacks. Roda, Macchi, & Fumagalli (2018), figured out in their

survey that one of the barriers perceived as critical is the potential lack of guarantees

from current Cyber Security technologies on the complete effectiveness in data

protection. Adu-Amankwa, Attia et al. (2019) identify maintenance and security data

infrastructure as a key challenge of adoption.

Finally, it is relevant to notice that even if several papers deal with the data

collection and management, only one mentions the lack of a widespread standard

solution for new technology as a barrier perceived by companies (Roda, Macchi, &

Fumagalli, 2018).
 4.5 Evaluation and Discussion of the Review 73

Servitization

The industrial IoT not only promises to transform the factories’ plant, but it also

has the capability to transform the intrinsic way of doing business (Noureddine,

Solvang et al., 2020). Bokrantz, Skoogh et al. (2020) in their survey, demonstrate that

companies recognize the importance of transmitting data and are ready to share

relevant information with suppliers and partners. The Industrial IoT technology links

all these players together, paving the way for new practices and techniques to be used

in maintenance operations (Noureddine, Solvang et al., 2020). On the one hand, the

supplier has access to valuable information to improve product development and

develop services based on extended knowledge bases. On the other hand, the customer

receives back this value in the form of more relevant knowledge, more reliable

products and better services (Bokrantz, Skoogh et al., 2020). While some authors only

recognized the potential of the data streaming as a possibility for new revenue streams

(Noureddine, Solvang et al., 2020), others propose a variety of possible business model

features that can be offered to the customer (Müller & Däschle, 2018). The Industrial

IoT is driving the “transition of traditional pure products business models to

Product/Service-Systems (PSS)” enabling “the creation of value through services in

different parts of the network.” (Osako, Matsubayashi, Takey et al., 2019). On the same

line are Kans & Ingwald (2016) that present a four-level framework representing the

logical development of the service business model from a narrow technical perspective

to a holistic product-service perspective. Huang et al. goes a step further modelling the

value generate both for the supplier and the customer in the Product/Service-Systems.

They theoretically and practically demonstrated that a PSS reduces the downtime of

the equipment and brings many other monetary benefits to customers, manufacturers

and local repair companies included in the transaction as third parties.

 74 Chapter 4. Literature Review

4.6 Results and Future Direction

from the Literature Review

Since the first appearance of the IoT theme in the industrial field, academic

papers have had an exponential growth. Industrial IoT is revolutionizing all sectors

and the traditional way of doing business. Among the others, the theme of the

Industrial IoT in predictive maintenance is about the major focus point of this

transformation (Kaur, Selway et al., 2018).

We directed a literature review most systematically and transparently possible,

following the guideline of Seuring & Gold (2012). We examined 57 research

contributions to the application of the Industrial IoT technology to improve predictive

maintenance in manufacturing. The period of publication of the paper cover six years,

from 2015 to 2020. Furthermore, the vast majority of them, is concentrated between

2019 and 2020 (33 out of 57), highlighting the novelty of the theme and the growing

interest that researchers have in this topic. As for the methodology, our review

revealed that many of the articles examined are conceptual articles. A fair number of

empirical studies (i.e. mainly based on surveys, or case studies or interviews) were

also analysed. We found very few contributions based on simulation and

mathematical modelling.

We developed this structured review that provides a guide to previous research

in predictive maintenance in the 4.0 era. In general, we have found several interesting

topics in the literature. Firstly, we have analysed on which aspects of the Industrial IoT

in the field of predictive maintenance the authors focus the most: Industrial IoT

framework, Business decisions, and Factory field. Secondly, we have examined the

improvements and transformations it can produce from an economic and productivity

point of view: benefits and servitization. Finally, we have observed the problems that

hinder its adoption.

Although the novelty of the theme, the potential of the Industrial IoT in the

various areas of manufacturing, and in particular in predictive maintenance, has

 4.6 Results and Future Direction from the Literature Review 75

aroused great interest from scholars all over the world. However, some themes are still

underrepresented or missing, as emerged from the revised literature. In order to

provide some insights that lead to further research, we list the main shortcomings.

1. The benefits of adopting predictive maintenance with the aid of the Industrial

IoT technology are little investigated

Costs – benefits analysis to the adoption of the Industrial IoT technology

for predictive maintenance are rare, even if the importance of their role in the

adoption process is acknowledged by various authors (Roda, Macchi, &

Fumagalli, 2018; Compare, Baraldi, & Zio, 2019). Furthermore, it would be

interesting to understand the level of knowledge of these technologies among

companies to figure out how deeply companies perceive the benefits. For

example, Strauß, Schmitz et al. (2018) report that companies are reluctant to

invest too much in improving their machines to the smart level, a signal of a

poor understanding of the potential reachable benefits and, therefore, the

return on investment.

2. There is no unanimous sharing of the direction of predictive maintenance

using Industrial IoT solutions

Even if authors recognize the potential that Industrial IoT has to evolve

maintenance, there is no such unanimity in which direction this evolution

will be. The vast majority of them believe that IoT technology has the

potential to improve the accuracy of predictive maintenance and focus their

studies mainly in this direction. Others believe that this technology takes

maintenance to the next level and consider predictive maintenance as a key

but not a unique element. The problem that derives from this is a continuous

proliferation of names (e.g., Smart Maintenance, e-Maintenance,

Maintenance 4.0), which can be traced back to a single universal matrix: the

Industrial IoT technology. In this regard, a holistic view of the different

nomenclatures is rare in literature and would be necessary to enhance clarity

and direct efforts in a common direction.

3. Studies on the evolution of traditional business models to adapt to the future

of the maintenance are rare

 76 Chapter 4. Literature Review

If on the one hand the transition toward the digital world is deeply

analysed by all the authors, on the other, only a few of them investigate the

possibilities for the companies to transform the business model with the use

of the Industrial IoT and the subsequent disruption that can cause to the

industries. The magnitude of the change requires much more effort and

attention in the attempt to smooth the transition toward the servitization of

the world and the maintenance in particular.

To conclude, it is necessary to highlight that this study has a significant limitation

that must be considered. Despite the efforts to include all relevant papers in the

research, some articles may have been inadvertently omitted. However, we believe

that the sample accurately represents the literature on the implementation of the IoT

strategy to predictive maintenance and therefore, we believe that confidence may be

placed on the resulting assessment.

Literature Review Tables 77
78 Chapter 4. Literature Review
Literature Review Tables 79
80 Chapter 4. Literature Review
Literature Review Tables 81
 Chapter 5

5 Methodology

The chapter presents all the path that has been followed to define the research

questions and the way to answer them. Moreover, it presents and explains all the

approaches used to obtain information.

 5.1 Research Questions and Methodology 83

5.1 Research Questions and

Methodology
As we have seen in previous chapters, the manufacturing sector represents a

pillar for the economy of most countries in the world. In recent years, developed and

developing countries are relying more and more on the growth of this sector for the

expansion of the whole economy, and digital transformation is playing a fundamental

role. Among the most relevant technologies of the digital transformation, the Internet

of Things has the most significant potential. According to a McKinsey paper, the

Industrial Internet of Things, Industrial IoT, can generate an economic impact of $1.2

trillion to $3.7 trillion per year. This considerable value "would arise chiefly from

productivity improvements, including 10 to 20 per cent energy savings and a 10 to 25

per cent potential improvement in labour efficiency. Improvements in equipment

maintenance, inventory optimization, and worker health and safety are also sources

of value in factories." (McKinsey&Company, 2015). However, even if several countries

put in place much economic measure to help companies in the digital transition, the

Industrial IoT technology is still spreading slowly.

Given these premises, we first set out to analyse this technology in the

manufacturing sector to understand its potential and investigate its benefits. We

realized that the potential of this technology is innumerable and there are several areas

in which is possible to start a project that requires specific investments (predictive

maintenance, material handling, rather than monitoring parameters along the cold

chain). Given the size of the field, we decided to focus on the application that seems to

benefit the most and have the most significant impact: predictive maintenance.

Subsequently, we analysed the literature to investigate first how the IoT

improves this technology and creates value for the different players in the value chain.

Secondly, we have deepened the theme to fully understand its characteristics, the IoT

structure necessary to set up, the technical and economic benefits, the barriers and any

new business models that may derive from it. From this analysis, we found out that
 84 Chapter 5. Methodology

among the major barriers to adoption, there is a poor understanding of the benefits

and the consequent economic return of the investment. The papers on the subject

report the benefits only of a qualitative nature and without providing a deep analysis.

They are generally used as evidence to substantiate their thesis. On the other side,

those who venture into the creation of a model limit themselves to roughly estimate

the costs rather than giving an estimate of the benefits obtainable. For this reason, our

thesis work tries to fill this gap, to create a tool to quantify these benefits and able to

calculate the return on investment.

Besides, we believe IoT technology gives many vendors the ability to incorporate

new services into their business model that result from greater remote asset control. In

an attempt to create a model that looks to the future, we decided to incorporate the

predictive maintenance service, the amount of which is calculated based on the hours

of actual machine activity. The vendor is able, thanks to the sensors installed on the

machine, to have all the data regarding its use. Thus he can align its business model

with that of its customer without risking moral hazard or opportunistic behaviour by

the manufacturer. Given the potential benefits that predictive maintenance can

generate for the customer, it seemed appropriate that these be shared in part with the

vendor. It is in this light that we thought that our model could have a dual-use. If on

the one hand, it allows the user of the machine to understand what the return on his

investment will be, on the other hand, he allows the person who sells the machine to

understand the benefits of which his customer benefits and consequently to define his

margin.

To summarize, we can say that we first asked ourselves to understand what the

benefits of IoT were in the manufacturing sector. After identifying predictive

maintenance as one of the main ones, we questioned ourselves why it was so

important, and we sought an answer in the literature on this by tightening the focus

and creating a tailor-made research question. From this arose our research question:

Q1: Does an investment in Industrial IoT for predictive maintenance generate

value for the company?
 5.1 Research Questions and Methodology 85

As stated earlier, the literature makes very few contributions to this topic. To try

to answer as precise as possible, we decided to create an analytical tool that allows the

user to evaluate the value of the investment following the insertion of variables. In this

way, not only we can answer our research question but also we can adapt our model

to different situations.

Primarily we asked ourselves what the tool that an entrepreneur or manager of

a company uses to evaluate whether or not to invest was. In this way, we aligned as

much as possible the output of our model with what is required in the strategic

decision-making sphere. We have identified two main tools useful for our decision:

the NPV and the ROI index. To decide between the two tools, we asked ourselves

about the pros and cons, comparing them with the goal we had in mind. By needing a

tool capable of evaluating the return on investment in an absolute manner, also

considering the time value of money, we agreed that the VAN tool was the most

appropriate for our purpose. Once we defined that the model output would be a cost-

benefit analysis using the traditional NPV approach, we started looking for literature

papers to follow a systematic approach. Although applied to a different area, we

decided to follow the guidelines by Hanley and Spash (1993). They defined seven main

steps to conduct the cost-benefit analysis (CBA):

1. According to Hanley and Spash (1993), the first step corresponds with the

definition of the project and the identification of the resources to be used. In

this first step, we also gather as many models as possible we found in the

literature in order to have a framework to base on. Unfortunately, as stated

before the topic is not much addressed; only a few contributions were found.

One, in particular, was very useful for laying the foundations of our model.

The paper in question dates back to 2006, when the IoT still had not made its

appearance in an industrial way, and predictive maintenance was still based

on rudimentary methods of data collection. However, the paper presents a

simple but effective model for estimating the benefits based on the average

repair time in the case of scheduled and unscheduled operations. This value

multiplied by the hourly cost of the work allows us to calculate the benefit

that is obtained by scheduling the interventions each an intervention is

86 Chapter 5. Methodology

necessary. This value then multiplied by the number of times that the

predictive maintenance is successful, allows to obtain the total benefit

(Hecht, 2006). However, this model has numerous limitations such as the

fact that it exclusively considers a repair time differential as a benefit, does

not consider the value of money over time and considers a rough estimate

of the efficiency of the prediction that does not allow it to compare with any

other type of maintenance if not corrective.

2. The second step corresponds to the identification of all possible impacts of

the project (i.e. benefits and disadvantages). To gather as many ideas as

possible and to make sure that we take into consideration all the aspects

influenced by investment in predictive maintenance, we used different

sources. We have collected all the benefits listed in the previously analysed

literature and used the citations to investigate thoroughly. We carried out

interviews in the industrial field, to have practical feedback on the main

problems that emerge during a break. Finally, we used secondary sources to

have the widest possible spectrum.

3. In the third step, all the possible impacts previously identified are examined,

and their relevance in economic terms is assessed. In this step, many of the

benefits and all the disadvantages that we previously identified were

eliminated because they were considered irrelevant in economic terms and

would have only added uncertainty to the model.

4. The Hanley and Spash model (1993), is structured to provide the guidelines

for an application of a CBA analysis, considers in the fourth phase, the

physical quantification of the economically relevant impacts. In our case, this

step translates into the development of physical variables that will be

quantified in economic terms in the next step.

5. The fifth step involves the transformation of the previously identified

variables into monetary variables.

6. The sixth phase of the analysis applies the net present value test (NPV),

which assesses whether the sum of discounted benefits exceeds the sum of

discounted costs. While the model part stops at the previous step, these sixth

and seventh steps are presented in the model application phase.

 5.1 Research Questions and Methodology 87

7. The seventh and final step of the CBA analysis is the sensitivity analysis to

evaluate which variable has the most significant effect on the NPV. A

scenario will be presented precisely for this purpose.

This approach presented will not be structurally present in the subsequent

chapters where, for reasons of clarity, we preferred to follow a different logical course.

However, it is essential to underline that these steps have been carefully followed

during the model definition phase. It should also be noted that during all these steps

we have tried to keep the model as generic as possible while making some

assumptions and imposing some limitations that will be presented at the beginning of

the next chapter. However, the idea was to create a tool as faithful as possible to reality,

for this reason, the interviews carried out were useful for understanding the most used

maintenance strategies (e.g. preventive and corrective maintenance).

Finally, in order to ensure that IoT technology spreads rapidly and widely also

in the industrial sector, all the players involved must make a profit from investments

in digital technology. What could be the incentive of a vendor to invest in making his

machinery intelligent, if all the resulting benefit is captured by his manufacturer? The

price surplus and the competitive advantage are certainly two aspects that should not

be underestimated, but the new business models to which the IoT allows access is the

real treasure. In an attempt to include this aspect in the search, we have created a

research sub-question:

Sub. Q1: How can the vendor benefit from the value generated to the manufacturer?

We decided to include this aspect in the model in the form of a predictive

maintenance service offered to the manufacturer to replace previous subscriptions. To

bring this into the Net Present Value formula, we interpreted all the costs of the

interventions previously carried out as benefits and the cost of the new subscription

as a cost and therefore subtracted from the total value.

 88 Chapter 5. Methodology

To sum up, we can affirm that the methodology we used comprehends different

approaches. In particular, we have used:

 Literature: It represents the first step we have taken to deepen the theme

of Industrial IoT, to fully understand the context and define the

boundaries of this work. The analysis of the literature was fundamental

to understand all the introductory notions that we then needed for our

thesis work. Furthermore, once the primary thesis topic was defined, the

literature was extensively revised (57 papers were analysed) to go deeper

into the specific context and find the existing gaps and where to

concentrate the work.

 Interview: To validate and find confirmation of the notions collected with

secondary sources and to collect data regarding real situations, we

carried out some interviews. In particular, they were essential to recover

the data that allowed us to test the model, both to give a real dimension

in the two imaginary scenarios, and to create the presented case study.

The total interviews with industrialists are three.

 Secondary sources: Secondary sources were also used as a first step

together with the literature to deepen the theme of Industrial IoT. In the

same way, they were used at a later stage to fill in some deficiencies in

the literature, as well as a secondary verification of the benefits, and

therefore of the variables to be taken into consideration in the model.

(papers from companies or major consultancy companies). Among

others, the main sources used were the articles found on the websites of

major companies, large consultancies, white papers and databases

recognized as valid globally.

 Analytical model: To answer the research question concretely, an

analytical model has been formalized, which allows evaluating the

economic feasibility of investment in Industrial IoT for predictive

maintenance.
 Chapter 6

6 Model Design

This model is a new approach to the analysis of the benefits achievable from an

implementation of a predictive maintenance service, introducing the vendor actor

offering this service in a pay-per-performance agreement. The financial tool is the

NPV. After some initial hypotheses, the formulas are described, explained the

variables contained and presented its functioning. The manufacturing field appears

complex, and it would be an ambitious claim stating to be able to manage it

completely; therefore, we need to present a flexible structure that remains valid

beyond the various estimates that will be presented, and that allows to insert

new/other cost items if necessary

 6.1 Model Introduction 91

6.1 Model Introduction

Many variables in a production plant affect the total cost of the maintenance

activity, from the downtime to the cost of the replaced spare parts. According to

Komonen 2002, the percentage of maintenance costs compared to the total capitalized

assets varies between 1% and 25% (Komonen, 2002). Such an incidence rate shows how

fundamental it is to design the maintenance strategy better, choosing the right balance

between the various maintenance policies, which maximizes both efficiency and

effectiveness. The choice of maintenance policies must follow precise logics, deriving

from in-depth knowledge of the plants, from failure analysis, from economic

evaluations on the cost of the life cycle of company assets. The model we developed

presents a detailed way of analysis of ongoing maintenance costs and helps to make

informed decisions on the implementation of a new predictive strategy.

More in detail, based on input variables that characterize the state of the initial

system taken into consideration, the model provides an estimation of the benefits that

can be obtained by transforming the traditional maintenance paradigm into an

intelligent one that implements predictive maintenance. In addition to this benefits

deriving from purely technical variables, the model also takes into consideration the

transformation of the business model allowed by the IoT system. The quantity and

quality of data that the sensors can collect, together with the technology of transmitting

information at very high speed, permit the vendors to have the opportunity to

complete their business model by offering the manufactures (i.e. the companies that

uses the machine) services more aligned to their needs. The model gives the possibility

to consider the benefits deriving from abandoning the traditional logic of a fixed

annual subscription or payments per service call (generally in urgency and at very

high costs) to switch to a payment-per-performance logic. The total benefit obtained

from the sum of these two is then used as a differential cash flow within a classic NPV

formula for cost-benefit analysis, which allows understanding the feasibility and the

goodness of the investment.

This model to estimate the economic value creation has been implemented in

Excel and tested in different scenarios. Thanks to a division into two parts between
 92 Chapter 6. Model Design

costs deriving from technical variables and those deriving from the implementation of

the new subscription, the model can evaluate the benefits deriving from the two and

suggests whether to invest in Industrial IoT for a predictive maintenance solution or

just a new type of subscription.

This structure of the model allows the results to be interpreted from two points

of view: the manufacturer and the vendor himself. The manufacturer can objectively

evaluate the convenience of an investment in Industrial IoT solutions for predictive

maintenance on each of his machines; the vendor, on the other hand, can take

advantage of the model to understand the benefit generated for the user of his

machinery, leveraging on this value as a competitive advantage over competitors.

In the next sections, we will go in deep with the functioning of the model. We

will first set the hypotheses necessary to let the model works, and we will explore the

formulas that compose the model to have an entire overview. Once the general and

secondly go in deep in understanding the functioning step-by-step.

6.2 Hypotheses
The model is a simplification of reality intended to promote understanding and

comprehension. Hypotheses are essential to achieve these objectives, and to arrive at

their definitions, the base terminology of the different maintenance approaches is

initially presented.

The As-Is and To-Be Overview

According to Zio & Compare (2012), there are two macro areas of possible

approaches in maintenance: corrective and preventive. In the first group, the

components are managed until failure, and then repairs or renovations are
 6.2 Hypotheses 93

conducted22. Preventive maintenance, on the other hand, includes all the actions

performed in the attempt to keep an item in specified conditions, providing systematic

inspection, detection and prevention of incipient failures. Accordingly, preventive

approaches can be further divided into three subgroups: scheduled maintenance, if the

actions are performed based on a predetermined condition 23; condition-based

maintenance, which uses monitoring to identify problems at an early stage and

performs actions when the level of degradation reaches a threshold; and predictive

maintenance, the degradation of component is predicted in the future, and its

Remaining Useful Life (RUL) is estimated. Even if different approaches of the same

macro area, the last two can be considered as one the evolution of the other. Indeed,

the predictive maintenance approach takes its roots in the condition-based type and

evolves it. Starting from this macro classification, the proposed model considered two

possible as-is situations: the corrective maintenance and the scheduled maintenance

(specifically the time-based one, based on a calendar schedule). In this way, we can

have a more realistic starting point of the model since several companies adopt both

these strategies and a more representative value of the return on the investment. It

would be an error considering predictive maintenance as only a substitute for the

corrective one (Compare, Baraldi, & Zio, 2019). Based on this consideration, we could

set a hypothesis zero to highlight the starting point of the model:

H0: As-is is represented by corrective and time-based maintenance.

22 The corrective maintenance is the oldest approach and is nowadays still

adopted, especially for equipment which is not crucial for the production performance
of the plant or considers safety-critical, and whose spare parts are easily available and
not expensive (Zio & Compare, 2012).

23 The key issue in the Preventive Maintenance is to determine the optimal

intervention intervals (i.e., the time spans between two successive maintenance
actions), which balance the costs of actions with the production and safety benefits
obtained from repairing/substituting a component before it experiences a failure.
 94 Chapter 6. Model Design

Some considerations regarding the solution proposed are also necessary. The

Predictive Maintenance is based on different procedures/methods: knowledge-based,

analytical and data-driven (Alzghoul, 2014). The experience needed to perform the

first method and the difficulty in modelling the machine/system failure with the

second (van Rijn, 2007), leave space for growth to the third option, empowered by data

accessibility enabled by Industrial IoT. Literature concentrates much effort into

explaining the framework to implement data-driven predictive maintenance, from

data collection and preparation to algorithm evaluation (examples, supervised or

unsupervised). The architecture to be adopted is widely present as well, evaluating

from edge to fog and cloud solutions, structure capable of managing real-time and big

data. In the model, this part is assumed to be already performed by the vendor of the

machinery/system, which provides the manufacturer with a complete solution for

monitoring and predictive analysis. The to-be considered is represented by a

predictive maintenance service bought from the manufacturer, furthermore, in a pay-

per-performance agreement (a virtuous example of this offer is Fabio Perini spa, the

case study presented in the section 3.2). It is supposed that different failures can be

monitored using the same architecture, enhancing the multi-use and interoperability

characteristics of an Industrial IoT solution, no more closed in a silos perspective. The

model is designed for a cost-benefit analysis of a single machinery at a time.

Application does not exclude old machinery (where no data acquisition and

connectivity are available in the as-is, the so-called brownfield problem24), as long as

a retrofitting with sensors allows the creation of the cyber-physical system for

predictive maintenance.

H1a: For a specific machinery, the manufacturer buys a predictive maintenance

service offered by the vendor in a Pay-per-Performance agreement.

24 The so-called brownfield problem is the challenge of embedding predictive

maintenance solutions into existing framework conditions (Strauß, Schmitz,
Wöstmann, & Deuse, 2018).
 6.2 Hypotheses 95

H1b: The initial investment includes the “acquisition” of the Industrial IoT

architecture, enabling a complete monitoring and predictive analysis.

Type of Intervention and its

Characteristics

A further hypothesis regards the cost per intervention. Following the definition

given on the type of maintenance (with a first division between corrective and

preventive), the maintenance interventions could be generally divided into two

groups as well, depending on whether they are performed before (and so we assumed

a scheduled intervention) or after the component failure (unscheduled intervention)

(Zio & Compare, 2012). Intervention costs could be expanded, including all the effects

of the stop on the overall system (for example, penalties). Unscheduled intervention

cost can be supposed more volatile than scheduled one, due to the uncertainty of the

timing of the breakdown event. The time when a component fails is crucial on the

general performance (Hecht, 2006), influenced even just from different availability of

maintenance capability. In the model, an estimate of the necessary variables will be

formalized, giving them an average annual value.

H2: The formalization of costs per intervention lies in the division into two types,

unscheduled and scheduled intervention cost, estimated on an annual basis.

The next argument regards the types of failure that could affect the

system/machinery. The non-accidental failures are characterized by an unavoidable

natural phenomenon: the degradation (Baur, Albertelli, & Monno, 2020). Traditional

analysis tools are adopted to investigate the causes and effects of failure (for example,

root cause analysis, fault tree analysis, and Failure Mode and Effect Analysis). In a
 96 Chapter 6. Model Design

conceptual phase, the asset behaviour/conditions depending on time can be

represented by a health time curve. What is usually reported is the P-F curve, that

generally it represents the same concept but specifically highlighting two essential

ideas of predictive maintenance: the potential fail (detectable state) and the functional

fail (failed state). The interval between the two represents the time windows in which

predictive maintenance should be executed to limit the consequences. Only

components emitting signals of deterioration can be successfully managed with

predictive maintenance. In the model, we assumed that once the intervention is

performed, the initial conditions are restored. The system is supposed to run until its

design life, regardless of the maintenance approach used. With this assumption, the

potential benefits from moving from a corrective to predictive maintenance could be

underestimated, considering preventive interventions being able to extend the useful

life of the machinery (McKinsey&Company, 2015).

H3: The intervention will restore the initial condition of the machinery, letting

the system runs until the design life.

In this section, we discussed the starting point and the hypothesis of the model,

from maintenance approach definitions to the key concepts that will be relevant in the

next chapters. Other smaller assumptions will be taken during the model’s

explanation, coherent with the objective of this thesis to represents the possible

benefits and costs of an Industrial IoT application from the manufacturer’s but also

from the vendor’s point of view.

 6.3 Model Principles 97

6.3 Model Principles

The final output of the model is represented by the Net Present Value of an

Industrial IoT solution enabling a predictive maintenance service in a pay-per-

performance agreement with the vendor. The NPV allows keeping different cash flows

per year. The cash flows are represented by the two major themes of the model: the

savings deriving from the transition from as-is maintenance to predictive

maintenance, and the cost difference between the dynamic subscription Pay-per-

Performance, and previous maintenance cost.

The investment in sensors can be declined in various applications, allowing a

predictive analysis as well as a more diffuse check of the operating status of the

machinery. This aspect will be taken into consideration in order to evaluate a transition

to a pay-per-performance payment agreement even if the predictive maintenance shift

results disadvantageous on some failure type clusters. The NPV structure taken as a

reference is the following.

𝑛
𝑆𝑎𝑣𝑃𝑑𝑀,𝑖 − 𝐷𝑖𝑓𝑓𝑀𝑆𝐶𝑖
𝑁𝑃𝑉 = −𝐼𝑛𝑣0 + ∑ (1)
(1 + 𝑊𝐴𝐶𝐶)𝑖
𝑖=1

 𝑁𝑃𝑉 [€] (Net Present Value): It represents the final object of this model, taking

into account the different value of the cash flow over the useful life of the

asset;

 𝑆𝑎𝑣𝑃𝑑𝑀,𝑖 [€] (Savings Predictive Maintenance per year): The global Saving that

derives from the shifts to a Predictive Maintenance approach;

 𝐷𝑖𝑓𝑓𝑀𝑆𝐶𝑖 [€] (Difference in Maintenance Service Cost): The differential between

the cost of the Pay-per-Performance subscription and the previous

maintenance cost;

 𝐼𝑛𝑣0 [€] (Initial Investment): The investment is presented as the undividable

cost to digitalize the machinery (Hypothesis H1a);

 98 Chapter 6. Model Design

 𝑊𝐴𝐶𝐶 [%] (Weighted Average Cost of Capital): This variable represents the

Weighted Average Cost of Capital of the firm. It is supposed to be fixed over

the year during the lifetime of the machinery;

 𝑛 [𝑦𝑒𝑎𝑟] (Useful Life): Year of useful life (𝑖 is the variable used to move from

year 1 to 𝑛).

In the next two sub-chapters the 𝑆𝑎𝑣𝑃𝑑𝑀 and 𝐷𝑖𝑓𝑓𝑀𝑆𝐶 variables are described

more in detail.

Maintenance Strategy Comparison

To perform the correct assessment of the investment and to consider the different

types of failure of the single machinery, it is necessary to identify different clusters that

can be formed according to the following variables:

 The maintenance strategy adopted;

 The economic impact of the intervention;

 The performance of the forecasting algorithm for the specific failure type;

 Other failure characteristics;

The change from as-is to predictive maintenance needs to be analysed cluster by

cluster, and this is required to take into account the wide diversity of machine failures

that leads to different saving evaluation. Predictive maintenance can be

disadvantageous in some circumstances, and in those cases, the algorithm will suggest

maintaining the as-is maintenance strategy. Only the positive savings of the specific

clusters will be taken into account for the calculation of the 𝑆𝑎𝑣𝑃𝑑𝑀 to be included in

the NPV formula.

𝑆𝑎𝑣𝑃𝑑𝑀 = ∑ 𝑆𝑎𝑣𝑃𝑑𝑀,j 𝑤𝑖𝑡ℎ 𝑆𝑎𝑣𝑃𝑑𝑀,j > 0 (2)

𝑗=1
 6.3 Model Principles 99

 𝑆𝑎𝑣𝑃𝑑𝑚 [€⁄𝑦] (Savings Predictive Maintenance): Annual savings achievable

overall considering the ones of clusters that predictive maintenance change

results to be positive. This calculation is repeated year by year;

 𝑆𝑎𝑣𝑃𝑑𝑚,𝑗 [€⁄𝑦] (Savings Predictive Maintenance per Cluster): Annual savings per

cluster, which has its specific form in case the as-is situation is corrective (3)

or preventive maintenance (4);

 𝑘 [#𝑐𝑙] (Cluster Number): The number of clusters taken into account per

machinery (𝑗 is the variable used to move from cluster 1 to 𝑘).

As anticipated, the structure of the savings for each cluster is a function of the

maintenance strategy. The resulting formula for corrective maintenance is:

𝑆𝑎𝑣𝑃𝑑𝑀−𝐶𝑟𝑀 = 𝐼𝐶𝑢𝑛 ∙ 𝑇𝑃 − 𝐼𝐶𝑠𝑐 ∙ (𝐹𝑃 + 𝑇𝑃) + 𝐸𝑋𝐶𝐶𝑟𝑀 (3)

In case of preventive maintenance, the formulation changes in:

𝑆𝑎𝑣𝑃𝑑𝑀−𝑃𝑣𝑀 = 𝐼𝐶𝑢𝑛 ∙ (𝐵 − 𝐹𝑁) + 𝐼𝐶𝑠𝑐 ∙ (𝑆 − 𝐹𝑃 − 𝑇𝑃) + 𝐸𝑋𝐶𝑃𝑣𝑀 (4)

In both cases, the structure results to be similar, presenting the estimate of the

number of times an unscheduled or scheduled intervention is necessary during the

year, multiplied by the cost per intervention of that type, plus costs directly

attributable to that maintenance approach. The shared variables are related to the

intervention cost and the predictive algorithm performance.

 𝐼𝐶𝑢𝑛 [€⁄𝑖𝑛𝑡 ] and 𝐼𝐶𝑠𝑐 [€⁄𝑖𝑛𝑡 ] (Cost per intervention): they represent the cost per

unscheduled and scheduled intervention respectively;

 𝑇𝑃 [𝑖𝑛𝑡⁄𝑦]: True Positive;

 𝐹𝑃 [𝑖𝑛𝑡⁄𝑦]: False Positive;

 𝐹𝑁 [𝑖𝑛𝑡⁄𝑦]: False Negative.

 100 Chapter 6. Model Design

True positive, false positive and false negative are variables indicating the ability

of the predictive maintenance algorithm to provide the right forecast of the failure.

They will be declined in the concept of Precision and Recall (section 6.9). For the

comparison with corrective and preventive maintenance, other variables were

declared.

In case of corrective maintenance, the specific variable is:

 𝐸𝑋𝐶𝐶𝑟𝑀 [€⁄𝑖𝑛𝑡 ] (Exclusive Cost related to corrective maintenance): It takes into

account the cost of service intervention (internal or external) and an explicit

warehouse cost.

In the preventive maintenance approach, the specific variables are:

 𝐸𝑋𝐶𝑃𝑣𝑀 [€⁄𝑖𝑛𝑡 ] (Exclusive Cost related to preventive maintenance): In this case, it

considers the service cost as fix subscription or a portion of the internal

maintenance department cost;

 𝑆 [𝑖𝑛𝑡⁄𝑦]: Number of service intervention per year;

 𝐵 [𝑖𝑛𝑡⁄𝑦]: Estimate of the number of corrective interventions still required in

case of preventive maintenance.

It is essential to understand how the cost per intervention 𝐼𝐶 is calculated. In the

model, it has a structure shared either the intervention is unscheduled or scheduled.

The specific cost variables are mutual between the types of intervention. When the

model is applied, those same variables acquire a different value. We could identify

three large cost groups:

𝐼𝐶 = 𝐷𝑇𝐶 + 𝑆𝑝𝐶 + 𝑃𝑛𝑙𝑡 (5)

 𝐷𝑇𝐶 [€⁄𝑖𝑛𝑡] (Downtime Cost): The downtime-related cost considering the

impact of a stop of the considered machinery;

 𝑆𝑝𝐶 [€⁄𝑖𝑛𝑡] (Spare part Cost): The cost to replace or repair the

component/system;
 6.3 Model Principles 101

 𝑃𝑛𝑙𝑡 [€⁄𝑖𝑛𝑡 ] (Penalties): The penalty cost in case of miss client agreement.

These variables will be further explored in the following chapters. Ultimately, a

more general consideration wants to widen the scope of this formulation, which

leaves, if necessary, the possibility to insert additional costs to fit the specific case of

application. This flexible structure can be considered a salient point of the goodness of

the model.

Dynamic Payment Method

Pay-per-performance agreement is an ever-existing payment method, and in this

last period, it is experiencing a new popularity in the BtB sector. The re-discovery of

the term can be related to the digital innovation that, through the Industrial IoT,

provides increasingly advanced data collection and analytical methods, making

possible more immediate and easy-to-build analyses, from establishing system

performance onwards. The transition to a dynamic payment agreement will be taken

into consideration even if not all the clusters of the machinery move from the as-is to

the predictive maintenance type. This assumption relies on hypothesis H1a in which

the Industrial IoT investment on the machine is unique and leads to a complete

acquisition of the operating data. The different declination of the investment (not only

for a predictive maintenance application) wants to enhance the multi-use

characteristics of Industrial IoT solutions. Below the cost of the Pay-per-Performance

subscription is presented.

𝐷𝑖𝑓𝑓𝑀𝑆𝐶 = 𝑃𝑝𝑃𝑆𝑢𝑏 − 𝑀𝑆𝐶𝑥 (6)

𝑃𝑝𝑃𝑆𝑢𝑏 = 𝐹𝑒𝑒 ∙ (𝑊ℎ − 𝐷𝑜𝑤𝑛𝑇𝑖𝑚𝑒) (7)

 𝐷𝑖𝑓𝑓𝑀𝑆𝐶 [€⁄𝑦𝑒𝑎𝑟 ] (Difference in Maintenance Service Cost): The difference

between the cost of the Pay-per-Performance subscription and the previous

 102 Chapter 6. Model Design

fixed annual one (not already accounted in the savings from the shift from as-

is to predictive maintenance);

 𝑃𝑝𝑃𝑆𝑢𝑏 [€⁄𝑦𝑒𝑎𝑟 ] (Pay-per-Performance Subscription): It represents the cost of

the Pay-per-Performance subscription. A possible formulation of this cost is

represented by the cost per ensured hours established from the vendor ( 𝐹𝑒𝑒

[€⁄ℎ𝑜𝑢𝑟 ] ) multiplied by the possible UpTime of machinery operation,

represented by the difference between agreed working hours and the

machinery downtime ( 𝑊ℎ − 𝐷𝑜𝑤𝑛𝑇𝑖𝑚𝑒 [ℎ𝑜𝑢𝑟⁄𝑦] );

 𝑀𝑆𝐶𝑥 [€⁄𝑦] (Previous Maintenance Service Cost): it represents the previous

fixed annual subscription or service intervention cost not already accounted

in the savings (2), which means when the cluster of failure is not taken into

account for the shift to predictive maintenance approach. It could be

formulated as follows:

𝑀𝑆𝐶𝑥 = ∑ 𝐸𝑋𝐶𝑗 𝑤𝑖𝑡ℎ 𝑆𝑎𝑣𝑃𝑑𝑀,j < 0 (8)

𝑗=1

The shift to a Pay-per-Performance agreement will create a closer relationship

between the company owning the machinery and the vendor, giving to the

maintenance the typical flexibility of a service. To conclude the chapter model

principles, the formula of the DownTime (the dynamic part of the subscription cost) is

presented:

𝐷𝑜𝑤𝑛𝑇𝑖𝑚𝑒 = ∑(𝑀𝑇𝑇𝑅𝑢𝑛,𝑗 ∙ 𝐼𝑢𝑛,𝑗 + 𝑀𝑇𝑇𝑅𝑠𝑐,𝑗 ∙ 𝐼𝑠𝑐,𝑗 ) (9)

𝑗=1

 𝐷𝑜𝑤𝑛𝑇𝑖𝑚𝑒 [ℎ𝑜𝑢𝑟⁄𝑦] (Downtime): The downtime represents one of the key

variables in the maintenance sector, with the Mean Time To Repair declined

in 𝑀𝑇𝑇𝑅𝑢𝑛,𝑗 and 𝑀𝑇𝑇𝑅𝑠𝑐,𝑗 , in case we considered an unscheduled or

 6.4 Flowchart of Model Functioning 103

scheduled intervention respectively (𝑗 is used to indicate which cluster has

been referred to, moving from 1 to 𝑘)

 𝐼𝑢𝑛,𝑗 [𝑖𝑛𝑡] and 𝐼𝑠𝑐,𝑗 [𝑖𝑛𝑡]: These variables refer respectively to the estimated

number of corrective and preventive interventions of the specific cluster with

the type of maintenance chosen after the evaluation performed in formula (2).

6.4 Flowchart of Model Functioning

In the previous paragraph, we provided an overview of the model and the more

general formulas has been presented. In this part, we will analyse step by step the

model in order to provide a more detailed guide and enhance comprehension. The

diagram in Figure 6.1 shows the macro areas that characterise the model and define its

functioning. First, it is necessary to understand the production plant and machine

characteristics on which the model is to be applied. Depending on these characteristics,

the initial costs change and consequently, the convenience to activate a predictive

maintenance solution with the use of Industrial IoT. Among these variables, a separate

discussion is dedicated to those relating to the failure rate and the failure distribution.

Subsequently, in section 6.8, the possible IoT systems that can make the failure

prediction system smart are presented. Finally, in section 6.9, any benefits deriving

from the predictive maintenance solution are assessed. After the definition of some

variables by the user, the NPV of the solution is calculated. In the following sections,

we will deepen all of these parts to explain in detail the formulas and the variables that

must be considered.

Figure 6.1: Flowchart of model functioning

 104 Chapter 6. Model Design

6.5 Plant Characteristics

The first step in making the model work is the definition of the general initial

conditions, the as-is situation. The first values to insert in the model are the number of

clusters and the correspondent maintenance policies according to the variables

expressed in section Maintenance Strategy Comparison. Many variables contribute to

the definition of the total maintenance cost: the costs deriving from downtime, those

deriving from the personnel assigned to the interventions and the necessary spare

parts represent only a part of the total which according to Bevilacqua & Braglia (2000)

represents between 15 and 70% of total production cost (Bevilacqua & Braglia, 2000).

An estimate that has a vast range but that makes the idea of the value at stake. In this

section, the first variables to be taken into consideration and inserted in the model are

explained.

To calculate in detail the total cost of maintenance, it is, therefore, necessary to

first define the total cost of maintenance intervention. In this regard, we start from the

general formula 10, which will then be declined, in the different types of maintenance

in question.

The cost formula is:

𝐼𝐶 = 𝐷𝑇𝐶 + 𝑆𝑝𝐶 + 𝑃𝑛𝑙𝑡 (10)

Where IC is the cost activated every time there is a maintenance intervention (i.e.

Intervention Cost). It is composed by a variable part that depends on the time the

machine is not working because of the intervention (i.e. downtime cost - 𝐷𝑇𝐶) and a

fixed part: the cost of spare parts the technician has to replace (𝑆𝑝𝐶) and the penalties

that must be paid in case of delay (𝑆𝑝𝐶).

As stated previously in the hypothesis, the model base is functioning on the

application of this formula for the computation of the cost per unscheduled and

scheduled intervention. Apart from the corrective maintenance, which considers only
 6.5 Plant Characteristics 105

the cost of the unscheduled intervention, both the preventive and the predictive are a

composition of the two. Therefore, whatever type of maintenance the company is

implementing, it is fundamental to underline that both, the unscheduled and the

scheduled cost must be completed to let the tool work.

Several other costs contribute to the overall cost of the intervention. However,

some of them will be considered later since they do not vary according to the number

of interventions. Some other costs are intentionally left aside because we believe that

their calculation is of marginal importance. For instance, the cost of non-compliant

pieces produced after a breakdown occurs, that can be considered the sum of the cost

of the pieces worked while the machine is working in bad condition. The cost of re-

working these pieces is minimal respect to the overall amount of pieces that cannot be

produced while the machine is not working. Moreover, this value strongly depends

on the situation and can have huge differences from time to time, making his estimate

highly inaccurate and therefore adding uncertainty to the system.

Downtime Costs

Downtime is the time when the machine is not working, which means that no

value is being produced for the company. Process managers struggle to try to make

their impact as lower as possible. However, even if companies are aware of the effect

of the downtime, 80% of the companies are unable to calculate their exact downtime

costs correctly (Immerman, 2018). A distinction between planned and unplanned

downtime must be made because the greatest expenses come from the unplanned one

(for example, excessive tool changeover, excessive job changeover, lack of operator,

and unplanned machine maintenance). Besides, it is important to specify that

unscheduled interventions are entirely unplanned, while scheduled one is planned

intervention even if they can generate unplanned downtime if not carried out

correctly. Since the focus of our model is on maintenance, we will only consider the

downtimes linked with these type of interventions and therefore, their costs.
 106 Chapter 6. Model Design

Since a downtime generate not only the stoppage of the faulty machine but also

it produces efficiency losses in the production lines or the machines nearby, the

formula sum both of them as follow:

𝐷𝑇𝐶 = 𝐷𝑇𝐶𝑚𝑎𝑐ℎ𝑖𝑛𝑒 + 𝐷𝑇𝐶𝑝𝑙𝑎𝑛𝑡 (11)

The downtime cost of the machine is:

𝐷𝑇𝐶𝑚𝑎𝑐ℎ𝑖𝑛𝑒 = 𝑀𝑇𝑇𝑅 ∙ ∑(𝑅𝑃𝑅ℎ ∙ 𝐵𝐾%,ℎ ∙ 𝐺𝑃𝑢,ℎ ) (12)

ℎ=1

Where:

 𝑀𝑇𝑇𝑅 [ℎ⁄𝑖𝑛𝑡 ] (Mean Time To Repair) 25: It corresponds to a metric used by

the maintenance departments to measure the average time needed to

determine the cause and repair the failed equipment. Il includes:

o Notification to maintenance technicians and, since case the

intervention is made in outsource, the time to intervene on the

spot;

o Diagnosis of the problem;

o Solve the problem;

o Reassembly, alignment and validation of equipment;

o Restore, test and start the equipment or system for production.

In addition to these standard values, in the corrective case for our

model, it is important to also include any other time that the machine stops

due to the fault (e.g. lead time spare parts). Furthermore, it is important to

specify that even if in theory the value is used exclusively to respond and

25 In this paper, Mean Time To Repair and Mean Time To Recovery are assumed as
synonymous.
6.5 Plant Characteristics 107

repair unplanned failures, in this model the MTTR is also used for planned

maintenance interventions with the necessary arrangements.

In particular, to have a rough estimation of the MTTR in case of

scheduled intervention, it is possible to consider only the time needed to

solve the problem and restart the machine. All the other time can be

considered wasted and therefore eliminated in a scheduled maintenance

activity. In case it is not possible to have an estimation with this level of detail,

the Mean Time To Repair in scheduled and unscheduled situations can be

obtained from the historical series applying the following formula:

𝑇𝑜𝑡𝑎𝑙 𝑚𝑎𝑖𝑛𝑡𝑒𝑛𝑛𝑐𝑒 𝑡𝑖𝑚𝑒

𝑀𝑇𝑇𝑅 =
𝑁𝑢𝑚𝑏𝑒𝑟 𝑜𝑓 𝑟𝑒𝑝𝑎𝑖𝑟𝑠

In case the scheduled time wants to be obtained, the ratio has to be done

between the total maintenance time of the scheduled intervention and the

total number of scheduled repairs; for the unscheduled MTTR, it has to be

calculated as the ratio of the unscheduled time and the following repairs.

In case the company only performs unscheduled maintenance and does

not have any data about scheduled interventions, or vice versa, a good

approximation can be obtained using a percentage reduction (or increase) of

the unscheduled MTTR. According to the Federal Energy Management

Program publication, this value is around 35-45% less than the corrective one

(U.S Department of Energy, 2010), while a McKinsey analysis of 2015 estimate

that the gain can be up to 50%.

𝑝𝑐
 𝑅𝑃𝑅ℎ [ ⁄ℎ] (Real Production Rate): The real production rate refers to the

number of goods that can be produced in an hour considering the mix,

the availability of the machine and all the other aspects that can reduce

the overall output. The subscript ℎ indicate the products (from 1 to 𝑚)

that are worked on the faulty machine. For example, supposing that

machine 1 produces products A, B and C with a production rate (that

already considers all the operative downtime of the machine) of 100 pc/h,

50 pc/h and 200 pc/h respectively. The machine works 2 hours a day on

product A, 3 hours a day on product B and 3 hours a day on product C.

108 Chapter 6. Model Design

To calculate the hourly mix of three it must be multiplicate the number

of hours the machine works each piece by its production rate, and the

result divided by 8, the total number of hours a machine work a day.

This value is an estimation of the hourly mix produced by machines since

it can vary from year to year or even every day. These values should be

well known to the COO of the companies or even to the owner itself.

 𝐵𝐾%,ℎ [%] (Percentage of slowdown due to the Breakdown): This value is used

to adjust the effect of a breakdown. It is a measure of how much a

breakdown slow down the production of the product ℎ produced by the

faulty machine. This percentage can vary according to different

characteristics of the plant. For example, assuming to be in a job shop

layout, where piece A, B, C, D and E are worked on machine 1 in a given

mix. Supposing no other machine in the plant can process products A, B

and C and no other time windows where recover the loss production,

the breakdown of machine 1 stop their production by 100%. Therefore,

the 𝐵𝐾%,𝐴 , 𝐵𝐾%,𝐵 and 𝐵𝐾%,𝐶 is equal to 100%. Slightly different is the case

of D and E. These two products can also be processed by machine 2.

Machine 2 processes product F, G and H, and it is fully saturated with

their production. In this case, if the company want to continue to

produce product D and E has to decrease the quantity of the products

already processed on the machine26. Therefore, the slowdown due to the

break of machine 1 on product D and E is not total and their 𝐵𝐾%,ℎ is to

be considered lower than 100%.

When the value is different than 100%, it is difficult to provide a precise

and general formula to evaluate it because it depends on too many variables.

Therefore, the value to be entered is left to the interpretation of the user

machinery based on his experience.

26 The algorithm does not provide a logic to follow to redirect the flow.
 6.5 Plant Characteristics 109

 𝐺𝑃𝑢,ℎ [€⁄𝑝𝑐 ] (Unitary Gross Profit): It represents the unitary gross profit

of the good ℎ processes by the faulty machine. It is the difference between

the price and the production cost of the good.

The formula of the downtime cost for the plan has the same appearance as the

one of the machine:

𝐷𝑇𝐶𝑝𝑙𝑎𝑛𝑡 = 𝑀𝑇𝑇𝑅 ∙ ∑(𝑅𝑃𝑅𝑂ℎ ∙ 𝑂𝐸%,ℎ ∙ 𝐺𝑃𝑂𝑢,ℎ ) (13)

ℎ=1

Where:

 𝑀𝑇𝑇𝑅 [ℎ⁄𝑖𝑛𝑡 ] (Mean Time To Repair): It corresponds exactly to the same

amount considered in the formula (12).

𝑝𝑐
 𝑅𝑃𝑅𝑂ℎ [ ⁄ℎ] (Real Production Rate of the plant): As the previous

production rate, this value corresponds to the real number of goods that

can be produced in an hour considering the mix, the availability of the

machine and all the other aspects that can reduce the overall output.

However, contrary to the other, it is not focused on a single machine, but

it considers all the other machines that are affected by the faulty machine.

 𝑂𝐸%,ℎ [%] (Percentage of slowdown due to the breakdown in other machines):

This variable corresponds to the percentage of slowdowns of all the other

products that are produced on machines that are influenced by the

breakdown. 𝑂𝐸% varies according to the same characteristics as the 𝐵𝐾% ,

and it is difficult to provide a single formula to calculate it. Taking the

previous example, if the production manager decides that product D and

E cannot stop, they must divert them on machine 2. However, machine

2 is already fully saturated by product F, G and H and therefore to move

the production on machine 2 their amount must be reduced. This second

part of the formula for the downtime cost take into account this aspect,

all the products that are not produced on the faulty machine but are still

affected are considered, and 𝑂𝐸% is used as the percentage reduction.

 110 Chapter 6. Model Design

Even if the value could be considered in the machine downtime cost and

only one formula can be computed, we separated them for two reasons.

First, it is clearer what the contribution of each part is; secondly because

this formula can be used to take into account eventual loss in

productivity of other machines due to other indirect costs that are linked

with the lower production rate due to the breakdown. From the example,

it is evident that provide a formula for the calculation of 𝑂𝐸% has the

same obstacles of the formula of 𝐵𝐾% because it is not easy to calculate

mathematically how the breakdown of one machine can influence the

others. Therefore, since too many variables influence this value, it is left

to be estimated by the end-user of the model based on his experience.

 𝐺𝑃𝑂𝑢,ℎ [€⁄𝑝𝑐 ] (Average Unitary Gross Profit): This value corresponds to

the average gross profit of the entire products whose production is slow

down by the breakdown. It is calculated with the same logic of the

previous.

Spare Parts

A spare part is an interchangeable part that is stored in a warehouse and used for

the repair or replacement of failed units. Under this denomination, many costs could

be considered. However, for the calculation of the intervention cost, only the variable

part (i.e. the pieces that are replaced in the intervention) are considered. Other costs as

the rent of the space, the salaries of the warehouse staff, all the handling costs and

other “out-of-pocket” expenses are not considered in this calculation. While in the next

sections we will deepen the opportunity cost of the invested capital in the spare parts

warehouse, we did not include in the model the other costs related to the warehouse

storage previously mentioned due to their high uncertainty in the allocation and the

low incidence at a single piece level.

 6.5 Plant Characteristics 111

The formula of the spare part is:

𝑆𝑝𝐶 = 𝐷𝑆𝑝𝐶 + 𝐴𝑆𝑝𝐶 (14)

Where:

 𝐷𝑆𝑝𝐶 [€⁄𝑖𝑛𝑡 ] (Direct Spare part Cost): This variable represents the cost of

the spare part that technicians have to change after the breakdown occurs

(or when they perform preventive interventions) to fix the machine. To

calculate it, the total number of spare parts replaced in one year must be

divided by the overall amount of the interventions conducted.

 𝐴𝑆𝑝𝐶 [€⁄𝑖𝑛𝑡 ] (Additional Spare part Cost): When no preventive action is

taken to avoid fault, the breakdown that occurs, if not detected promptly,

can cause the breakdown of other pieces. This variable takes into account

this aspect, and therefore, it is different to zero only in case of

unscheduled interventions. The value is calculated as the previous one.

Penalties

The agreement between a supplier and a buyer for the delivery of goods

generally provides for an attractive delivery date for the seller. Since a failure to deliver

the goods within the established times could cause a loss for the customer, to

encourage the supplier to meet the deadline and divide the risk, it is possible to specify

in the contract the application of penalties for delayed delivery. This variable is

included to consider this aspect in the model. Indeed, a failure in the machine, if not

fixed in time can generate delay in the delivery. While the impact of this variable is

extremely high in the case of an unscheduled intervention, in case of scheduled we

assume it equal to zero, since there is the possibility indeed to schedule the

intervention in the best moment possible. It is difficult to provide a univocal way to

calculate the formula since it varies enormously from case to case; different subjective
 112 Chapter 6. Model Design

aspects could be taken into account. In the GDO industry, for example, a delay in the

delivery of the products can cost the company to be substitute with another who sells

the same products or in the machine production industry can be a penalty caused by

the fact that the client cannot produce and sell a specific good. Therefore, because of

these conditions, this variable is left to be estimated to the end-user of the model,

according to the contract they have.

Additional Costs

Although at a glance, it may seem that the downtime costs are exclusively costs

related to the repair of the machine, we have seen in the previous paragraphs that

several others must be considered. When a machine stops working due to a

breakdown, the costs of spare parts must be considered as the cost of non-production,

(the cost of loss of efficiency of the workers) and the cost of possible breaches of the

contract with the customer (for example, delayed delivery).

Other costs can be taken into account when calculating the total cost of

downtime. There may be recovery costs, such as the cost of employees working

overtime and the costs of data recovery, or there may be the costs of reworking some

pieces that have gone wrong and the costs of the pieces to be discarded. However, as

mentioned above, these costs are not taken into account in the model because they

have a reduced impact compared to those considered.

Besides these, there are other more difficult costs to determine (i.e. intangible

costs), such as any damage to reputation or brand, stress. Intangible costs may be

difficult to predict, but having a thorough understanding of the potential long-term

impact on future sales and customer retention can help entrepreneurs and managers

to have a clear view of the real cost of downtime and take countermeasures to stem it.

Our model does not provide a specific cost item for each of these, because as we have

seen, they can have a huge impact in certain industries and be completely irrelevant in

others. However, if one or more of these intangible costs that have a notable impact on
 6.6 Failure Rate 113

the overall cost and they are known or can be estimated, these can be simply

considered in the intervention costs summing them in the penalties variable (𝑃𝑛𝑙𝑡).

6.6 Failure Rate

The failure of the machine could be caused by several reasons, from wear and

corrosion to fatigue, fracture and instability. Failures can be divided into complete and

partial, both of which are classified into sudden and gradual (Nakagawa, 2006). The

progress of the unavoidable natural phenomena listed before, can be represented by a

health index function of time. Different patterns can be identified depending on the

degradation mechanism involved:

a) A single-stage degradation pattern, associated with a monotonic and

continuous degradation process until machine failure or stop; the curve

profile is usually represented by linear or exponential function (Figure 6.2);

b) A two-stage degradation pattern, characterized by an initial healthy stage

with no evidence of fault and a second stage leading back to the described

pattern 1 (Figure 6.2);

c) A three-stage degradation pattern, happening when, next to a worsening of

the health index, it can be observed a decrease of the defects in happening,

after which an increase in size will manifest again (an example, the rolling

bearings inner surface fault: initially characterized by many vibrations due to

the bumps of the rolling elements on the surface; then the defects is smoothed

by the continuous impacts before starting to increase again) (Figure 6.2);

d) A multi-stage degradation pattern, it is often the case of a complex system by

multiple, interactive and concurrent faults (Baur, Albertelli, & Monno, 2019).
 114 Chapter 6. Model Design

(Baur, Albertelli, & Monno, 2020)

Figure 6.2: Examples of health index time profiles.

The health (or degradation) curve can also be used to define the right time of

intervention in case of condition-based or predictive maintenance. The object is to

minimize the risks associated with the failure (Basson, 2017). In this case, the curve

changes its name in the P-F curve. As for the health curve, asset performance or

condition declines over time, leading to the loss of function for which it was intended.

The transition from a healthy to an unhealthy stage, could vary from being instant to

a long degradation process (Bousdekis, Apostolou, & Mentzas, 2020). The time

windows between the potential failure P (representing in the P-F curve the time when

it is possible to detect that the equipment is in the process of failing), and the functional

failure F (the time where satisfactory performance standard are no longer met) will be

key in understanding the ability of predictive maintenance to arrange on time an

intervention. An example that results challenging to predict is often consider the

electronic component faults with degradations not progressive and observable (Baur,

Albertelli, & Monno, 2019).

In this scenario, the definitions of the primary variables are required. How long

can a unit/system operate without failure? Reliability is the answer to this question

defined as “the probability that the system/unit will perform a required function under
 6.6 Failure Rate 115

stated conditions for a stated period”27. Failure is usually represented by mechanical

breakdown, deterioration beyond a threshold or, looking to system performance,

appearance of certain defects or decrease beyond a critical level; the failure rate is the

most important quantity in maintenance theory, used as a measure for the aging

phenomenon (how a unit/system deteriorates with its age) (Nakagawa, 2006). The

Weibull statistical distribution (with parameter 𝜂 and 𝛽) is the used function to

represent the behaviour of reliability through the life of the asset (Durán, Afonso, &

Durán, 2019). With the Weibull parameters, it is possible to estimate the failure rate 𝜆

for a given component in time:

𝛽 𝑡 𝛽−1
𝜆(𝑡) = ∙( )
𝜂 𝜂

Setting different specific values of 𝛽 28, the profile obtained will result to be the

bathtub curve (in

Figure 6.3, it is represented by the SP3 pattern), which is used to describe the

behaviour of failure rate over time. Three phases can be described: (1) the infant

mortality, where the failure rate 𝜆 decreases over time, (2) the useful life where 𝜆

remains constant and (3) the wear-out phase, 𝜆 increases until the equipment will be

discarded (Durán, Afonso, & Durán, 2019).

27 The reliability presents a mathematical formulation passing through the

integral of the failure density function from t to infinity. For the rigorous structure of the
formula we referred to Nakagawa (2006).

28 𝛽 is called the shape parameter due to their effect on the 𝜆 profile. The other
parameter 𝜂 is representative of the scale and of the characteristic life of the equipment;
its value corresponds to the time in which 63,2% of the failures are expected to occur.
 116 Chapter 6. Model Design

(Durán, Afonso, & Durán, 2019)

Figure 6.3: Examples of three-failure rate pattern

Different sub-optimal environment conditions (e.g., reactive agents,

temperature, vibration) will affect the parameters of the equation, moving them

dynamically. The proposed model needs as input a single value per year that could be

represented by the sum of the failure rate of the specific cluster. A fixed/constant

failure rate assumption could be considered a good approximation of reality, when

different distribution functions for a variety of components are combined (Geitner &

Heinz, 2006).

Failure rates are usually expressed in events per million hours. Machinery

component failures will lie mostly between 1 and 100 failures per million hours

(Geitner & Heinz, 2006). A failure rate of 𝜆 = 123 × 10−6 ℎ𝑟 corresponding to almost

one failure yearly (over 8000 hours per year). The estimate of the failure rate could be

calculated at the component level until considering the entire machinery. The

complexity of a multi-component system in a changing environment represents a

crucial aspect if it also considered the correlation between different failure

mechanisms. The information about the failure rate is expected to be retrieved directly

from the vendor, who can leverage the design engineer capabilities. The effect of the

environment could be considered evaluating the best and worst scenario. If the

manufacturer has operating experience to rely on, he can estimate the failure rates

starting from field statistics. An example taken from Geitner & Heinz (2006) is reported

below in the table: from a recorded percentage of failure regarding an element, the

failure rate can be found as the multiplication of the percentage by the equivalent

failure rate of two incidents yearly on the machinery.

 6.7 Initial Output 117

(Geitner & Heinz, 2006)

Table 6.1: Example of how failure rate data for machinery components can be obtained
from field statistics starting by the total number of the registered failure

From a broader perspective, the model will consider describable by a failure rate

every condition that will be subject to maintenance intervention, from example, the

machinery level of oil, the corrosion of some components exposed to adverse

environmental condition or the crack of the mechanical component.

6.7 Initial Output

At this point, we could calculate an initial cost value of the situation as-is for each

cluster. This step allows to highlight how the costs of the intervention explained in

chapter 6.5 are weighed and to present the other cost variables not yet covered (specific

to the maintenance considered). As anticipated in the hypotheses, the as-is is

represented by two starting scenarios, corrective and preventive maintenance, which

therefore need to be treated in separate locations. Regardless, the structure in the two

assessments presents some common points that could be noted. The modelling field

(i.e., the manufacturing field) appears complex, and it would be an ambitious claim

stating to be able to manage it completely. Therefore, we need to present a flexible

structure that remains valid beyond the various estimates that will be presented, and

that allows us to insert new/other cost items if necessary. As anticipated, the

calculations of maintenance costs are performed for each cluster for which it was

decided to examine the machinery.

 118 Chapter 6. Model Design

Corrective Maintenance Charge

Corrective maintenance is an approach undertaken when an obvious fault has

been located. This approach is still a common practice29, especially for equipment,

neither crucial nor safety-critical (Zio & Compare, 2012). However, the high

percentage of corrective activities can be considered an indicator of possible

improvements. In the model, the failure rate is assumed as an estimate of the expected

number of interventions, modifying its dimensions from failure per million of hours

to failure per year.

The formula of the cost of corrective maintenance is:

𝐶ℎ𝑎𝑟𝑔𝑒𝐶𝑟𝑀 = 𝐼𝐶𝑢𝑛 ∙ 𝜆 + 𝐸𝑋𝐶𝐶𝑟𝑀 (15)

The cost of unscheduled intervention is multiplied by the failure rate 𝜆, and the

exclusive cost related to could be formulated as follow:

𝐸𝑋𝐶𝐶𝑟𝑀 = 𝐸𝐶𝑆𝐼 ∙ 𝜆 + 𝑊𝐶 (16)

Where:

 𝐸𝐶𝑆𝐼 [€⁄𝑖𝑛𝑡 ] (Extra Cost for Service Intervention): It has the dimension of cost
𝑓𝑎𝑖𝑙⁄
per intervention (it is necessary to multiply it by 𝜆 [ 𝑦] (Failure rate). This

29 More than 55% of maintenance activities of an average facility are still

corrective/reactive (U.S Department of Energy, 2010). This value was confirmed also
by a survey conducted in Sweden during a condition-based maintenance seminar,
showing a 56% of a mean value of the utilization grade of corrective maintenance
(Bengtsson, 2004).
 6.7 Initial Output 119

value wants to take into account the cost of the service. For an evaluation, a

distinction between internal or external intervention has been considered:

o Internally: In case of internal intervention, the formula for 𝐸𝐶𝑆𝐼 is

the multiplication of the number of people, on average, who must

be involved for the intervention by their hourly cost and the hours

of intervention needed (Mean Time To Repair of the corrective

intervention);

o Externally: If the maintenance intervention is managed externally,

an estimation of the cost of the urgent call starting from descriptive

variables of the failure could be misaligned, and what can be

recommended is to take the invoices made by the maintenance

provider to understand the cost per year.

 𝑊𝐶 [€⁄𝑦] (Warehouse Cost): The cost of maintaining stock, taken into account

as a cost of capital. An estimate of the spare parts kept in stock is required,

multiplied by the unit cost and the weighted average cost of capital.

Preventive Maintenance Charge

Preventive scheduled maintenance is set as a fixed time, meter interval, or a

combination of the two (the intervention will be performed at the first expired term).

As previously specified, the former has been considered for the comparison. The

predictable and frequent failure modes are usually tackled with a preventive

maintenance approach. The problem highlighted regarding preventive maintenance is

the high number of interventions carried out, which could be executed without a real

need. In the model, the number of preventive interventions is enclosed in the variable

𝑆, determined from the vendor’s guidance mostly or experience. By carrying out

regular preventive maintenance, the intent is to try to intervene before the breakdown

occurs; however, not all the scheduled interventions are carried out at the most

appropriate time, and unscheduled interventions could still happen. The variable 𝐵 is

formalized to keep into account this phenomenon.

 120 Chapter 6. Model Design

𝜆 − (1 − 𝑅𝑏 ) ∙ 𝑆, 𝜆 > (1 − 𝑅𝑏 ) ∙ 𝑆
𝐵={ (17)
0, 𝑒𝑙𝑠𝑒

#𝑓𝑎𝑖𝑙⁄
An example of the function with 𝜆 = 4 [ 𝑦] and 𝑅𝑏 = 33% is reported in
Figure 6.4, in which, also from a graphical interpretation, it can be deduced the need to

carry out six preventive fixed time intervention not to have unscheduled intervention.

𝑅𝑏 can be valued from considerations about the variability of the occurrence of the

failure, or more roughly taking as reference that generally 30% of preventive fixed time

interventions are performed unnecessary (IBM, 2016), assuming the difficulty to

evaluate the right timing to intervene. The proposed scenario, evaluated in the model

functioning chapter, will adopt this second method.

B 2

0
0 1 2 3 4 5 6 7 8
S

Figure 6.4: Plot of the function combining variable S to B.

After this introduction, the formula for the calculation of the cost of preventive

maintenance is presented:

𝐶ℎ𝑎𝑟𝑔𝑒𝑃𝑣𝑀 = 𝐼𝐶𝑢𝑛 ∙ 𝐵 + 𝐼𝐶𝑠𝑐 ∙ 𝑆 + 𝐸𝑋𝐶𝑃𝑣𝑀 (18)

The costs of unscheduled and scheduled actions are multiplied by the two

variables previously presented. The exclusive cost is represented as:

 6.7 Initial Output 121

𝐸𝑋𝐶𝑃𝑣𝑀 = 𝑆𝑢𝑏𝑃𝑢 + 𝑀𝐷𝐶𝑢 (19)

Where:

 𝑆𝑢𝑏𝑃𝑢 [€⁄𝑦] (Preventive Subscription Cost per cluster): The subscription on that

particular cluster is taken into account. The subscript 𝑢 denotes the division

of this cost per cluster. As for the evaluation of the external cost per corrective

maintenance, what can be recommended is to take the invoices made by the

maintenance provider to estimate the general cost per year of the machinery.

A division for the number of preventive maintenance clusters brings to the

specific cost to allocate.

 𝑀𝐷𝐶𝑢 [€⁄𝑦] (Maintenance Department Cost): In case there are linked internal

maintenance costs, a possible hypothesis to take them into account is to

divide the costs of the maintenance department by the number of machines

present in the plant. Providing an average estimate might be considered too

naive, but it could result appropriate in case of similar machines. Ultimately,

as before, a division for the number of preventive maintenance clusters brings

to specific cost to allocate.

We wanted to present a modelling of the possible more relevant factors, bearing

in mind, as already mentioned, the possibility to insert additional costs to fit the

specific case of application if necessary. In the next chapter, the introduction of the

Industrial IoT solution is discussed.

 122 Chapter 6. Model Design

6.8 Industrial IoT Solutions

This part represents the turning point of the model. After calculating the initial

output (section 6.7), the traditional approaches are abandoned, and the whole system

is moved toward the computation of the differential advantage of the predictive

maintenance solution. However, the aim of this paragraph is not to present the

formulas to calculate these benefits, instead expose the Industrial IoT solutions to

achieve better results in this type of maintenance, from the sensor to the analytics.

A predictive maintenance programme consists of three main steps: data

acquisition, data processing and maintenance decision-making (Jardine, Lin, &

Banjevic, 2006). From its definition onward, better ways of collecting, transmitting and

processing data have been sought. Today, Industrial IoT provides a very high amount

of data collected by low-cost sensors, transmitted, analysed and made available to all

users in real-time. Therefore, we can assess that Industrial IoT is the disruptive

technology that shifts the S-curve of predictive maintenance upwards, promising

improvements and increasingly widespread applicability in the near future, making it

more efficient and more effective.

The Initial Investment

Up to now, we spoke about Industrial IoT, predictive maintenance and the

benefits that follow. However, an important and often painful aspect to talk about is

the cost of the investment necessary to make the system functioning. Traditional

solutions such as SCADA, MES or CMMS systems have always allowed the collection

and analysis of data for maintenance purposes. Therefore, it seems logical to ask why

an Industrial IoT system should be implemented.

As previously mentioned, Industrial IoT is the technology that makes predictive

maintenance efficient, making prediction more reliable and precise. To improve

reliability, large quantities of data must be collected, transmitted and processed by

 6.8 Industrial IoT Solutions 123

sophisticated algorithms that cannot be implemented on traditional systems. For a

robust IoT-based predictive maintenance solution, an architecture designed

specifically for the production plant and specific machinery is necessary.

Consequently, the necessary investments for the sensors, the infrastructure and

software may differ from case to case. For the estimation of this variable, it is important

to underline that it must be considered both in the case of new machinery to be

purchased and in the case of machinery already in use in the plant. The estimation of

the two is slightly different. In one case, there is an overall value to improve the

production plant and make it smart, while in the case of purchasing of new machinery

must be considered the difference between how much the machine costs and how

much it would have cost without an IoT instrumentation.

Remote Monitoring

Another fundamental aspect of Industrial IoT and a key element of our model is

remote monitoring. The Industrial IoT allows the collection and transmission of data

anywhere. In this way, the vendor, or more generally OEM of a specific machine, can

offer new services and thus open new avenues for the business that they had never

had before. Specifically for our model, a remote monitoring system allows OEMs to

collect data from anywhere and provide predictive maintenance service to their

customers. In this way, the OEM can exploit scale economies gathering data from all

its machinery to train its algorithms and therefore make them more effective.

Moreover, thanks to the data collected, if an unexpected error occurs, the OEM can

determine what was happening before the problem occur and where the problem is,

speeding up recovery operations30.

30 In the model, the speeding up of the recovery operations is considered only in

the case the predictive maintenance solution is implemented. Therefore, the benefits
of remote monitoring can be slightly underestimated.
 124 Chapter 6. Model Design

This new way of performing predictive maintenance should not only be seen as

a new source of revenue stream for OEMs. A Bain survey of 2019 underlines that

barriers to adoptions to predictive maintenance has declined its attractivity to the

customers. Therefore, vendors that want to boost adoption needs to focus on helping

customers implement this practice, and remote monitoring is the solution. Moreover,

the United States Department of Energy, assessed that the equipment needed to

perform predictive maintenance “should not be purchased for in-house use if there is

not a serious commitment to proper implementation, operator training, and

equipment monitoring and repair.” Consequently, it specifies, “if such a commitment

cannot be made, a site is well advised to seek other methods of program

implementation—a preferable option may be to contract for these services with an

outside vendor and rely on their equipment and expertise.” (U.S Department of

Energy, 2010). Back in 2010, when the paper was written, the term Industrial IoT was

not dubbed yet, and the technology had almost no application in the business sector.

However, the manual implicitly underlines the need for this communication system

that the Industrial IoT allows and highlights the need for the adoption of this service

for all those companies that do not want or cannot support a huge investment in the

matter.

In the next section, the concept of remote monitoring will be presented in the

form of a pay-per-performance subscription that the manufacturer agrees with the

vendor. Given the high level of monitoring that the technology allows, the vendor can

align his business model with the manufacturer and share with him the risk of machine

downtime, agreeing to pay only for the uptime. In this way, the vendor not only

demonstrates his confidence in the reliability of the system convincing more easily

manufacturers to shift to the pay-per-performance solution, but he also improves the

customer satisfaction and, therefore, his relation. Thanks to the Industrial IoT, the

OEM can transform a costly activity, usually seen from the manufacturer as a waste of

money, to a source of competitive advantage.

 6.9 Final output 125

6.9 Final output

In this last section, the conclusion of the cost and benefit evaluation process of an

Industrial IoT application is presented, thus arriving to estimate all the latter necessary

terms for the NPV formula.

The computation of the differential advantage of the predictive maintenance

solution is performed for each cluster for which it was decided to model the machinery

(the variables will be populated k-times). Here below, the formulation in case of the

comparison of predictive maintenance respect to corrective and preventive approach:

𝑆𝑎𝑣𝑃𝑑𝑀−𝐶𝑟𝑀 = 𝐼𝐶𝑢𝑛 ∙ 𝑇𝑃 − 𝐼𝐶𝑠𝑐 ∙ (𝐹𝑃 + 𝑇𝑃) + 𝐸𝑋𝐶𝐶𝑟𝑀 (3)

𝑆𝑎𝑣𝑃𝑑𝑀−𝑃𝑣𝑀 = 𝐼𝐶𝑢𝑛 ∙ (𝐵 − 𝐹𝑁) + 𝐼𝐶𝑠𝑐 ∙ [𝑆 − 𝐹𝑃 − 𝑇𝑃] + 𝐸𝑋𝐶𝑃𝑣𝑀 (4)

Regarding these formulations, the majority of the variables have been already

presented in the previous subchapters. What represents novelty is the insertion of the

variables TP (True Positive), FP (False Positive) and FN (False Negative). They refer to

the confusion matrix (Figure 6.5), a method for summarizing the performance of a

classification algorithm. What it is necessary to define is the ability of the algorithm to

forecast a failure in time to be able to schedule the intervention. This result assumes

the form of a binary outcome, positive if it is possible to schedule, and negative, the

opposite31. With this method, instead of providing the life time remains, it will be

predicted whether a machine will fail within the next time T.

31 The other predictive maintenance method is focused on the regression

algorithm, predicting how much time is left before the next failure.
 126 Chapter 6. Model Design

Figure 6.5: Confusion Matrix

The valorisation of these variables could represent the grey field of the model,

where a tiny difference in time will result in a possible big difference in the cost per

intervention. If, as reported, 99% of machine failures are preceded by some indicators

(Bloch & Geitner, 2012), accurate estimates are not present on how much time before

the algorithm can give warnings of the failure. The aim is not only to predict the failure

event but also the time interval of its occurrence. This capability is also a function of

what kind of analysis is used; for example, there are cases in which vibration analysis

can spot failure months earlier (Selcuk, 2015). The P-F curve, already presented in

chapter 6.6, represents this concept in the time interval width between potential and

functional failure. In general, to have an evaluation of the algorithm, two relevant

variables can be introduced, the precision and the recall, defined as follow:

𝑇𝑃
𝑃=
𝑇𝑃 + 𝐹𝑃

 𝑃 (Precision): it represents how many right suggested interventions are

accurate respect the actual one.

𝑇𝑃
𝑅=
𝑇𝑃 + 𝐹𝑁

 𝑅 (Recall): it is a metric of the proportion of the failure that is going to happen

without being intercepted. In general speaking, what has been predicted

against what it should be.

 6.9 Final output 127

In a perfect algorithm, precision and recall will assume values close to one. In

reality, these two entities are in balance between them, as it will be reflected in the

model.

Figure 6.6: Perfect versus realistic prediction

From the system formed by the formulas of 𝑅 and 𝑃 plus the interpretation of

the failure rate as 𝜆 = 𝑇𝑃 + 𝐹𝑁, it is possible to calculate the missing variables of the

model:

𝐹𝑁 = (1 − 𝑅) ∙ 𝜆 (20)

𝐹𝑃 + 𝑇𝑃 = 𝑅⁄𝑃 ∙ 𝜆 (21)

𝑇𝑃 = 𝜆 − 𝐹𝑁 (22)

The criticality of the evaluation is closed in the variables 𝑅 and 𝑆; however,

considering the amount of data available by the vendor offering the predictive

maintenance service, it can be assumed that this information could be provided from

him. For a first estimation (as taken into account also in the chapter of model findings),

it could be considered the precision and recall of the algorithm present in the literature

referring to the pattern classification performance.

After having managed this aspect, the savings respect the as-is maintenance

strategy could be calculated. The model is structured considering only the positive

ones: not always, a shift to the predictive maintenance is advantageous.

 128 Chapter 6. Model Design

𝑆𝑎𝑣𝑃𝑑𝑀 = ∑ 𝑆𝑎𝑣𝑃𝑑𝑀,j 𝑤𝑖𝑡ℎ 𝑆𝑎𝑣𝑃𝑑𝑀,j > 0 (2)

𝑗=1

The second term to be analysed is the cost of the new subscription.

𝐷𝑖𝑓𝑓𝑀𝑆𝐶 = 𝑃𝑝𝑃𝑆𝑢𝑏 − 𝑀𝑆𝐶𝑥 (6)

𝑀𝑆𝐶𝑥 variable needs some further explanation It takes into account the exclusive

cost not considered in savings. In fact, for the cluster where savings are negative, the

exclusive cost (which can be seen as the cost of the maintenance service) will be

summed in this term. A change of subscription to pay-per-performance will be

evaluated on the whole machine. For this reason, the exclusive costs (that correspond

to the cost of maintenance that will be substituted by the new subscription) need to be

included.

𝑀𝑆𝐶𝑥 = ∑ 𝐸𝑋𝐶𝑗 𝑤𝑖𝑡ℎ 𝑆𝑎𝑣𝑃𝑑𝑀,𝑘 < 0 (8)

𝑗=1

For the calculation of the cost of the new payment agreement, the formula is

defined as follows.

𝑃𝑝𝑃𝑆𝑢𝑏 = 𝐹𝑒𝑒 ∙ (𝑊ℎ − 𝐷𝑜𝑤𝑛𝑇𝑖𝑚𝑒) (7)

Where:

 𝐹𝑒𝑒 [€⁄𝑦] (Fee): The fee is entrusted by the vendor, and, as a starting point, it

may be calculated from the previous cost with the downtime related. A shift

to the predictive maintenance will lead a drop of the downtime; the vendor
 6.9 Final output 129

(or the OEM providing the maintenance service) will obtain a gain for the

better performance achieved by the manufacturer;

 𝑊ℎ [ℎ⁄𝑦] (Working Hours): it can be established by the number of shifts

multiplied by the working days per year;

 𝐷𝑜𝑤𝑛𝑇𝑖𝑚𝑒 [ℎ⁄𝑦] (Downtime): The estimate of the downtime can be calculated

using the summation below:

𝐷𝑜𝑤𝑛𝑇𝑖𝑚𝑒 = ∑(𝑀𝑇𝑇𝑅𝑢𝑛,𝑗 ∙ 𝐼𝑢𝑛,𝑗 + 𝑀𝑇𝑇𝑅𝑠𝑐,𝑗 ∙ 𝐼𝑠𝑐,𝑗 )

𝑗=1 (9)

Where:

o 𝐼𝑢𝑛,𝑗 [#𝑖𝑛𝑡⁄𝑦] and 𝐼𝑠𝑐,𝑗 [#𝑖𝑛𝑡⁄𝑦] (number of interventions): they

represent the estimate of the number of interventions expected

divided per each cluster, taking into account if the cluster moved to

a predictive maintenance approach;

o 𝑀𝑇𝑇𝑅𝑢𝑛,𝑗 [ℎ⁄#𝑖𝑛𝑡 ] and 𝑀𝑇𝑇𝑅𝑠𝑐,𝑗 [ℎ⁄#𝑖𝑛𝑡 ] (Mean Time To Repair): they

measure per each cluster the average time needed to determine the

cause and repair the failed equipment in case of unscheduled and

scheduled intervention, respectively.

Finally, all the data can be inserted in the Net Present Value formulation that will

evaluate the investment over 𝑛 year, the useful life of the machinery.

𝑛
𝑆𝑎𝑣𝑃𝑑𝑀,𝑖 − 𝐷𝑖𝑓𝑓𝑀𝑆𝐶𝑖
𝑁𝑃𝑉 = −𝐼𝑛𝑣0 + ∑
(1 + 𝑊𝐴𝐶𝐶)𝑖 (1)
𝑖=1
 Chapter 7

7 Model Application

The structure of the model is adaptable and gives several possibilities in

personalisation. In this perspective, comprehension is vital to be able to tailor the

model to a specific case. To give the reader a complete insight of its functioning, this

chapter will present first a concept scenario and then an application to a real case. The

former is about an invented manufacturing plant, with a specific layout and product

flow. The variables Recall and Precision will be subjected to a sensitivity analysis, to

see a possible effect of the propagation of the uncertainty to the results. One real case

is then presented, fundamental to understand effectively how the model could work

inside a real context. Some decisions regarding a simplification of the model have been

taken in the first part of this chapter.

 7.1 Simplifications Adopted 131

7.1 Simplifications Adopted

The model leaves the possibility to enter different values year by year, and

obviously, this multiplies the necessary implementation effort. In this section we

present some difficulties in proceeding in this direction, arriving at the end to declare

some further hypotheses, most of which regard in keeping the variables constant year

by year (and so the cash flows). However, as will be pointed out in the conclusions, it

remains the possibility (if data are available) to use the model considering different

entries year by year. This step is necessary when the data available are limited. In the

concept scenario, the same decisions were taken to present clear results and to let

understand better the functioning of the model.

A general scheme describing the maintenance approach is presented (Figure 7.1)

to give a point of view regarding the difficulties in estimate the variables necessary for

the model.

(Zio & Compare, 2012)

Figure 7.1: Maintenance explanation schema

Corrective maintenance can be described taking into consideration the boxes

Failure and Maintenance. Preventive approach will add the Decision part, focus on

balance the costs of maintenance actions with the safety and production benefits

obtained from substituting/repairing a component before experience a failure. In case

of predictive the other two quadrants are added: a monitoring part (Observed
 132 Chapter 7. Model Application

Conditions) where the real-time data are a necessary condition, and a Prognostics

module where the actual state of the equipment is projected in the future to predict the

Remain Useful Life (RUL). The Failure rate, as presented in section (6.6), has a complex

behaviour and its complexity (in the estimation of the parameters) enlarged in case of

multi-component or multi-degradation system, assessing not only the parameter of

each single degradation path but also the correlation between the different mechanism

taking into account synergies and dependencies effects. In these cases, physical and

stochastic methods leave space for the experience-based ones to try to describe the

failure behaviour.

Another relevant point to be considered is the uncertainty derives from factors

like the maintenance action and the human reliability regarding its effectiveness;

model the effect of a maintenance policy is a critical point. Is it a minimal repair (As

Bad As Old, ABAO) or a perfect repair (As Good As New, AGAN)? Reality is

reasonably between these two extremes. As said before, with the preventive

maintenance, the Decision box (referring to Figure 7.1) is introduced. In literature, the

right optimisation in setting the variable 𝑆 (Number of service intervention per year)

is widely discussed. In the model, we did a step towards preventive maintenance

formalisation introducing the variable 𝑅𝑏 to take into account maintenance action

reliability/effectiveness. The variable 𝑆, in these applications, will be estimated

considering the number of actions performed in the last years. This number is

generally considered higher than the necessary. Observed conditions (or monitoring)

encloses some difficult in modelling too, considering that in some cases further

investigation of the warning could take place without stopping the equipment

(improving the performance of a preventive maintenance). The assumption here it will

be the ability to provide an overall estimate to consider the correct cost. At the end,

prognostic remains a challenging task, requiring the prediction of the evolution of the

failure behaviour. The performance of the model is expected to improve along the

years, considering either the use of traditional statistical approach or artificial

intelligence. Every algorithm has different performance that vary along time, and what

has been considered in the application was a value of 80% both for Recall and

Precision, the performance algorithm indicator adopted. The reference derives from

Zhang, Liu, Su, Han, & Li (2018) that obtain positive results in classifying high-
 7.2 Concept Scenario 133

dimensional data in a power plant, with precision and recall vary from 80% to 95%.

After this starting point, a sensitivity analysis varying these variables from 50% to

100% has been conducted to have a better evaluation of the benefits related. What is

interesting also to consider is an update/fix of the equipment by design for frequent

predicted failure, exploiting the possibilities of smart life cycle concept.

In addition to this part, the complexity in understanding how the flows of the

products are subjected to a stop due to a failure is hidden inside the variable 𝐵𝐾%, .

Simplifications could result in not considering possible buffer to absorb the stop of the

production. An analysis of a bottleneck machinery simplifies the considerations of

𝐵𝐾%, . To conclude, we assume all the variable introduced constant along the lifetime

of the machine, and this will bring constant cash flow.

7.2 Concept Scenario

A fictional manufacturing plant is presented, useful for better comprehension.

The company is in the mechanical engineering sector, hot forging of non-ferrous

materials (e.g. aluminium, copper, etc.); a turnover of 20 million per year, characterised

by orders of various sizes, generally small lots. In the case of some customers from the

automotive sector, large lots are present as well. In the next subchapters, the layout

and the division into clusters of the chosen machine will be defined. The calculation of

the various variables follow. The presentation of the benefits is associated with the

sensitivity analysis of variables 𝑃 and 𝑅.

Layout

The layout presents a division into four areas, a logical sequence followed by the

flow of products: the cutting department (3 machines used to cut extruded bar at the

desired length), the hot forging area (10 screw presses with different operating power
 134 Chapter 7. Model Application

and possibility to use specific moulds), the heat treatment zone (for solubilisation,

quenching and age of aluminium products) and the trimming department (where the

flash generating by the forging process is removed) (Figure 7.2). For the cost and benefit

analysis, a screw press that processes two product codes per month is taken into

consideration; the product codes are: P1, a component of the car shock absorber, and

P2, the motorcycle brake lever. An assessment of the criticality of the machine is not

explicitly presented: the starting situation is simply represented by the machine

vendor who is proposing a predictive maintenance service in pay-per-performance

agreement. An assessment of this proposal is conducted.

The company works on a single shift, 20 days a month for 12 months, resulting

in 1,920 hours per year. The production of P1 and P2 is divided into 15 and 5 days per

month, respectively. The effective production rate for P1 is 160 pieces per hour; 180

pieces per hour for P2. Given this data, it can be estimated that 120 pieces of P1 and 45

pieces of P2 representing in a sample hour. The sample hour is a significant concept in

the model, and not knowing the time of the failure leads to the need to summarise the

production mix (of the year) in a single hour. Other considerations regarding the

product flow concern P2, which can only be processed on that specific press (press A),

the production cannot be rescheduled; P1 does not have these limitations, and it could

be worked on the nearby press (press B), where a generic product, P3, is worked as

well. During a stop due to a machine failure, this flexibility will be used to give

continuity to the production.

Figure 7.2: Production cycle in the concept scenario

 7.2 Concept Scenario 135

Definition of the Maintenance Strategy

After an analysis most convenient of the machine failures, for this conceptual

model, we divided them into two primary clusters: the failures addressed through a

corrective approach (cluster 1) and those with a preventive approach (cluster 2). As

reported in the model design chapter, the creation of clusters can also be performed

looking to the economic evaluation of the failure impact, to the performances of the

predictive algorithm, or other characteristics of the failure. Here following the analysis

of the necessary variables computed for the two identified cluster.

Cluster 1

Different parts compose the structure of the cost per intervention. Starting from

the analysis of the downtime cost, and specifically the downtime of the machine, we

can estimate the following variables32:

Type of 𝑀𝑇𝑇𝑅1 𝑅𝑃𝑅𝑃1,1 𝐵𝐾%,𝑃1,1 𝐺𝑃𝑢,𝑃1 𝑅𝑃𝑅𝑃2,1 𝐵𝐾%,𝑃2,1 𝐺𝑃𝑢,𝑃2

𝑝𝑧 𝑝𝑧
intervention [ℎ] [ ⁄ℎ] [%] [€⁄𝑝𝑧] [ ⁄ℎ] [%] [€⁄𝑝𝑧]

intervention 40 120 50% 10 45 100% 6

Scheduled 20 120 0% 10 45 100% 6

Table 7.1: Variables of the downtime cost – Cluster 1

The value of Mean Time To Repair in case of unscheduled is retrieved by

averaging the historical series. For the scheduled 𝑀𝑇𝑇𝑅 a 50% reduction is considered

in line to what reported in McKinsey's report (McKinsey&Company, 2015). For the

percentage of 𝐵𝐾% , as anticipated in 7.1, various assumptions can be made to simplify

32 All the variables related specifically to the cluster 1 will present the subscript 1.
 136 Chapter 7. Model Application

its calculation. The fact that there is no idle production time is one of them. Several

considerations follow: P1 can be processed on another machine that at that time of

failure shows production capacity availability. This operation requires planning, that

means time, and in the case of an unscheduled production a quick turnaround. This

aspect is reported in the 𝐵𝐾% difference between scheduled and unscheduled. The

other product code P2 can not be worked on another machine, and therefore a failure

leads to a complete production stop. As already mentioned, we have decided to keep

the gross profit as the economic multiplier. In this way, the production costs are

assumed recoverable. By giving an example, the workforce in the case of a failure can

be immediately shifted to another occupation. If this does not happen, this term can

be used to add any production costs that have been removed. The downtime cost of

the machine can be calculated.

𝐷𝑇𝐶𝑚𝑎𝑐ℎ𝑖𝑛𝑒 = 𝑀𝑇𝑇𝑅 ∙ ∑(𝑅𝑃𝑅ℎ ∙ 𝐵𝐾%,ℎ ∙ 𝐺𝑃𝑢,ℎ )

ℎ=1

𝐷𝑇𝐶𝑚𝑎𝑐ℎ𝑖𝑛𝑒,𝑢𝑛,1 = 34,800 €/𝑖𝑛𝑡

𝐷𝑇𝐶𝑚𝑎𝑐ℎ𝑖𝑛𝑒,𝑠𝑐,1 = 5,400 €/𝑖𝑛𝑡

The passage of P1 to the second machine can lead to a slowdown of the other

production. This consideration is taken into account in the general downtime of the

plant. The formulas for the calculation have the same structure of the downtime cost

of the machine. In this case, only P3 is impacted.

𝐷𝑇𝐶𝑝𝑙𝑎𝑛𝑡,𝑢𝑛,1 = 𝑀𝑇𝑇𝑅𝑢𝑛 ∙ 𝑅𝑃𝑅𝑂𝑃3 ∙ 𝑂𝐸%,𝑢𝑛,P3 ∙ 𝐺𝑃𝑂𝑢,P3 = 40 ∙ 300 ∙ 10% ∙ 5

= 6,000 €/𝑖𝑛𝑡

𝐷𝑇𝐶𝑝𝑙𝑎𝑛𝑡,𝑠𝑐,1 = 𝑀𝑇𝑇𝑅𝑠𝑐 ∙ 𝑅𝑃𝑅𝑂𝑃3 ∙ 𝑂𝐸%,𝑠𝑐,P3 ∙ 𝐺𝑃𝑂𝑢,P3 = 20 ∙ 300 ∙ 15% ∙ 5

= 4,500 €/𝑖𝑛𝑡
 7.2 Concept Scenario 137

In these first variables lies the ability to react quickly to the failure, whether it has

been scheduled or not. Summed together, it turns out:

𝐷𝑇𝐶𝑢𝑛,1 = 𝐷𝑇𝐶𝑚𝑎𝑐ℎ𝑖𝑛𝑒,𝑢𝑛,1 + 𝐷𝑇𝐶𝑝𝑙𝑎𝑛𝑡,𝑢𝑛,1 = 34,800 + 6,000 = 40,800 €/𝑖𝑛𝑡

𝐷𝑇𝐶𝑠𝑐,1 = 𝐷𝑇𝐶𝑚𝑎𝑐ℎ𝑖𝑛𝑒,𝑠𝑐,1 + 𝐷𝑇𝐶𝑝𝑙𝑎𝑛𝑡,𝑠𝑐,1 = 5,400 + 4,500 = 9,900 €/𝑖𝑛𝑡

As regards the other two costs linked to the interventions, they are represented

by the cost of the spare parts and the cost of penalties for late delivery/non-production.

The former is considered an average of the annual historical cost divided by the

number of interventions executed (which amounted to 5,000 €/int). An extra cost of

3,500 €/int has been added in the case of unscheduled intervention; indeed, a delayed

failure detection can lead to the replacements of several other parts. There is a cost

linked to the penalty for non-delivery within the time agreed with the customer (a

practice present in the automotive sector). This cost was estimated to be 10,000 €, only

considering in case of unscheduled intervention. The total cost per intervention can be

calculated:

𝐼𝐶𝑢𝑛,1 = 𝐷𝑇𝐶𝑢𝑛,1 + 𝑆𝑝𝐶𝑢𝑛,1 + 𝑃𝑛𝑙𝑡𝑢𝑛,1 = 40,800 + 3,500 + 10,000 = 59,300 €/#𝑖𝑛𝑡

𝐼𝐶𝑠𝑐,1 = 𝐷𝑇𝐶𝑠𝑐,1 + 𝑆𝑝𝐶𝑠𝑐,1 + 𝑃𝑛𝑙𝑡𝑠𝑐,1 = 9,900 + 5,000 + 0 = 14,900 €/#𝑖𝑛𝑡

The unscheduled intervention cost is five times the scheduled one for this cluster.

As we have seen, the amount of the downtime cost, composed of the opportunity cost,

represents the highest percentage (78.9% in the unscheduled and 95.2% in the

scheduled).

The missing variable for the computation of the true positive and false positive,

is the failure rate, having already assumed a precision and recall of 80%. If the machine

is not represented by an already existing one, an analysis of the faults had in the last

years can be conducted (). Besides, there is always the possibility to ask direct
 138 Chapter 7. Model Application

information to the vendor. In this application, we assumed an average of 1.2 faults per

year (considering a single shift, 20 days a month for 12 months, a year corresponds to

1,920 hours). With these variables, we could move on to calculating the true positive

TP and the false positive FP.

𝑇𝑃1 = 𝑅1 ∙ 𝜆1 = 80% ∗ 1.2 = 0.96 𝑖𝑛𝑡/𝑦

𝑅1
𝐹𝑃1 + 𝑇𝑃1 = ⁄𝑃 ∙ 𝜆1 = 80%⁄80% ∙ 1.2 = 1.2 𝑖𝑛𝑡/𝑦
1

Service cost of the maintenance in the as-is misses to be calculated (the evaluation

of the new expense of the pay-per-performance subscription is considered in the next

section). The service cost variables of the corrective are the cost of the urgent

intervention and any costs related to the spare parts warehouse. In this application,

the maintenance is assumed external. The cost of the service request to the

maintenance technician(s) has been entered at the cost of 1,000 €/int (the other

possibility is represented by the cost of internal maintenance, calculated as the hourly

cost for the downtime). The cost for the service request must be multiplied by the

failure rate, to reach the estimated on annual; 1,200 €/y is the result to consider.

Regarding the warehouse, the cost of capital has been calculated: an average

stock of about ten spare parts with an average cost equal to the one previously

considered (5,000 € per piece). The fact of having ten pieces in stock with only one

failure per year should not mislead to think of an overestimation error: the failure rate

is representative of the whole cluster covering several failures, therefore the need to

have several spare parts. Considering the WACC of 5%, the cost of capital in stock is

2,500 €/y. Below, the formula for the annual cost estimate of the corrective service

followed by the sensitivity analysis letting the variable recall and precision vary from

50% to 100%.

𝑆𝑎𝑣𝑃𝑑𝑀−𝐶𝑟𝑀 = 𝐼𝐶𝑢𝑛,1 ∙ 𝑇𝑃1 − 𝐼𝐶𝑠𝑐,1 ∙ (𝐹𝑃1 + 𝑇𝑃1 ) + 𝐸𝑋𝐶𝐶𝑟𝑀,1

= 59,300 ∙ 0.96 − 14,900 ∙ 1,2 + 1,200 + 2,500 = 42,748 €/𝑦
7.2 Concept Scenario 139
 140 Chapter 7. Model Application

As we might have expected, the result is strongly influenced by Recall R and

Precision P. A variation of the recall has a greater effect on the final output than the

precision. This is due to the fact that the cost of unscheduled intervention has a more

significant impact. In this cluster it results to be about four times greater than the

scheduled one. Assuming 80% for both Recall R and Precision P, the savings obtained

compared with the as-is costs (74,860 €, referring to the initial output formula)

highlights an improvement rate of 57,1%.

Cluster 2

The second cluster groups all the failure addressed by preventive maintenance.

For the calculation of the costs of the intervention, the steps seen above are replicated.

We reported the summary table of the variables for the downtime cost estimation.

Type of 𝑀𝑇𝑇𝑅2 𝑅𝑃𝑅𝑃1 𝐵𝐾%,2,𝑃1 𝐺𝑃𝑢,𝑃1 𝑅𝑃𝑅𝑃2 𝐵𝐾%,2,𝑃2 𝐺𝑃𝑢,𝑃2

𝑝𝑧 𝑝𝑧
intervention [ℎ] [ ⁄ℎ ] [%] [€⁄𝑝𝑧] [ ⁄ℎ ] [%] [€⁄𝑝𝑧]

Unscheduled 6 120 50% 10 45 100% 6

Scheduled 4 120 0% 10 45 100% 6

Table 7.2: Variables of the downtime cost– Cluster 2

The previous reasoning about estimating the variables 𝐵𝐾%, remain valid. The

𝑅𝑃𝑅 and 𝐺𝑃𝑢 variables are independent of the cluster considered, and therefore these

remain the same. What is varied in this case is the Mean Time To Repair 𝑀𝑇𝑇𝑅.

Whether the intervention is planned or unplanned the repair times was chosen much

shorter than those of cluster 1. Looking to the internal differences, the scheduled

interventions was thought smaller for the ability to arrive on the machine and already

know what to fix; apart of this, a detection delay does not lead to a significant

worsening of the situation and the 𝑀𝑇𝑇𝑅 remains at low value even for the unplanned
 7.2 Concept Scenario 141

intervention. For 𝑀𝑇𝑇𝑅 estimation, we can assume the possibility to retrieve it from

historical data.

𝐷𝑇𝐶𝑚𝑎𝑐ℎ𝑖𝑛𝑒,𝑢𝑛,2 = 5,220 €/#𝑖𝑛𝑡

𝐷𝑇𝐶𝑚𝑎𝑐ℎ𝑖𝑛𝑒,𝑠𝑐,2 = 1,080 €/#𝑖𝑛𝑡

𝐷𝑇𝐶𝑝𝑙𝑎𝑛𝑡,𝑢𝑛,2 = 900 €/#𝑖𝑛𝑡

𝐷𝑇𝐶𝑝𝑙𝑎𝑛𝑡,𝑠𝑐,2 = 900 €/#𝑖𝑛𝑡

For the costs related to spare parts, 200€ have been added for each intervention

and 400€ additional in case of unscheduled ones. There are no penalty costs. The

intervention cost can be calculated:

𝐼𝐶𝑢𝑛,2 = 𝐷𝑇𝐶𝑢𝑛,2 + 𝑆𝑝𝐶𝑢𝑛,2 = 6,120 + 600 = 6,720 €/#𝑖𝑛𝑡

𝐼𝐶𝑠𝑐,2 = 𝐷𝑇𝐶𝑠𝑐,2 + 𝑆𝑝𝐶𝑠𝑐,2 = 1,980 + 200 = 2,180 €/#𝑖𝑛𝑡

The failure rate has been set at 6.4 and the number of annual interventions by

operators 𝑆 at 10.2 (in line with the simplification decisions taken in the first part of

this chapter).

𝐹𝑁2 = (1 − 𝑅2 ) ∙ 𝜆2 = (1 − 80%) ∙ 6.8 = 1.36 𝑖𝑛𝑡/𝑦

𝑅2
𝐹𝑃2 + 𝑇𝑃2 = ⁄𝑃 ∙ 𝜆2 = 80%⁄80% ∙ 6.8 = 6.8 𝑖𝑛𝑡/𝑦
2

The cost of preventive maintenance service is the cost of the internal maintenance

department divided by the number of machines monitored. Assuming an initial cost

of 48,500 €/y and 15 machines involved, the cost allocated per machine is 3,233 €/y. The

final savings for the cluster 2 can be calculated:

𝑆𝑎𝑣𝑃𝑑𝑀−𝑃𝑣𝑀 = 𝐼𝐶𝑢𝑛 ∙ (𝐵 − 𝐹𝑁) + 𝐼𝐶𝑠𝑐 ∙ [𝑆 − 𝐹𝑃 − 𝑇𝑃] + 𝐸𝑋𝐶𝑃𝑣𝑀

= 6,720 ∙ (0 − 1.36) + 2,180 ∙ (10.2 − 6.8) + 3,233 = 1,506 €/𝑦
142 Chapter 7. Model Application
 7.2 Concept Scenario 143

A sensitivity analysis regarding the annual savings achievable has been realised

in this second cluster (characterised by a preventive maintenance approach described

previously). A switch to the predictive maintenance is not always advantageous,

considering also having to add the cost of the new subscription in a second step. For

many combinations of precision and recall, the resulting savings are negative. This

depends from case to case, regarding the cost variables considered. The potentiality to

bring advantages is anyway always present, and it need to be discussed. Assuming

80% for both Recall R and Precision P, the comparison between savings obtained and

as-is maintenance costs (25,469 €/y, referring to the initial output formula) brings to a

percentage of improvement of 5,9%. In this concept scenario the unscheduled

intervention cost results three times greater than the scheduled one

Benefits Evaluation

The first formula presented summed the savings of the cluster that result to be

positive:

𝑆𝑎𝑣𝑃𝑑𝑀 = ∑ 𝑆𝑎𝑣𝑃𝑑𝑀,𝑗 𝑤𝑖𝑡ℎ 𝑆𝑎𝑣𝑃𝑑𝑀,𝑗 > 0 = 38,698 + 2,915 = 41,623 €/𝑦

𝑗=1

If there are negative clusters (therefore a switch to predictive maintenance is

assessed as economically disadvantageous), the as-is approach is maintained. The

switch to a pay-per-performance service is however evaluated, considering the related

propose on the whole machine33. For this reason, as it is correct to choose the more

33 In the model, to otherwise obtain an evaluation of the shift to only a predictive

maintenance approach without the pay-per-performance agreement, the cost of the
new subscription can be set to zero. Consider evaluating zero also the variables relative
to as-is service costs, to not overestimate the benefits obtained.
 144 Chapter 7. Model Application

convenient maintenance approach, it is necessary to recover all those service costs that

will be replaced by the new subscription. In case of a cluster switching to a predictive

maintenance approach, those costs will be already considered in the savings.

Otherwise, those costs, represented by the exclusive maintenance cost term introduced

previously, will be considered in the 𝑀𝑆𝐶𝑥 . In our application (recall and precision set

both to 80%), we obtained a positive savings for both the cluster, so the 𝑀𝑆𝐶𝑥 variable

results to be zero.

𝑀𝑆𝐶𝑥 = ∑ 𝐸𝑋𝐶𝑗 𝑤𝑖𝑡ℎ 𝑆𝑎𝑣𝑃𝑑𝑀,𝑗 < 0 = 0 €/𝑦

𝑗=1

The cost of the pay-per-performance subscription is based on the uptime of the

machine, and therefore, it reflects the downtime assessment. In the annual downtime

estimate, the repair time established for each cluster is taken into account multiplied it

by the number of interventions expected in the year. The formula is here presented:

𝐷𝑜𝑤𝑛𝑇𝑖𝑚𝑒 = ∑(𝑀𝑇𝑇𝑅𝑢𝑛,𝑗 ∙ 𝐼𝑢𝑛,𝑗 + 𝑀𝑇𝑇𝑅𝑠𝑐,𝑗 ∙ 𝐼𝑠𝑐,𝑗 ) = (40 ∙ 0.24 + 20 ∙ 1.2)

𝑗=1

+ (6 ∙ 1.36 + 4 ∙ 6.8) = 69 ℎ/𝑦

The more the vendor will reduce this downtime, the higher his profit will be.

Obviously, a central aspect for the vendor is fixing the value of the fee. An initial

estimate can be obtained from the previous total service cost divided by the established

number of working hours (one shift and 240 working days per year bring to 1960 hours

per year). An initial value of the fee results to be 3,61 €/h. Then, it is legitimate to

request some extra payments by evaluating the benefits that the manufacturer could

obtain. What has been considered is a fee represented by the initial estimate increased

by 20%, arriving at 4,33 €/h. The resulting subscription cost is:

 7.2 Concept Scenario 145

𝑃𝑝𝑃𝑆𝑢𝑏 = 𝐹𝑒𝑒 ∙ (𝑊ℎ − 𝐷𝑜𝑤𝑛𝑇𝑖𝑚𝑒) = 4,33 ∙ (1920 − 69) = 8,015 €/𝑦

Remember that the vendor's benefits may not only be economical, but he may be

more interested in the possibility of introducing smart life cycle logics that can be

obtained from the analysis of the acquired data. Considerations on lowering the cost

of the fee to make the investment more attractive for the manufacturer can be

evaluated. Another case, in which to lower the fee, is represented by the period of the

first evaluations of the predictive maintenance service, which can be done in

collaboration with specific manufacturers sharing the maintenance service costs.

Below, it is reported a formality step regarding where to take into account the

term 𝑀𝑆𝐶𝑥 (presented before), for then be ready to evaluate the overall Net Present

Value.

𝐷𝑖𝑓𝑓𝑀𝑆𝐶 = 𝑃𝑝𝑃𝑆𝑢𝑏 − 𝑀𝑆𝐶𝑥 = 8,015 − 0 = 8,015 €/𝑦

For the calculation of the Net Present Value, a remaining useful life for the

machinery of 20 years was considered, and as the weighted average cost of capital a

value of 5%. The initial investment has been valued at 50,000 € (U.S Department of

Energy, 2010).

𝑛
𝑆𝑎𝑣𝑃𝑑𝑀,𝑖 − 𝐷𝑖𝑓𝑓𝑀𝑆𝐶𝑖
𝑁𝑃𝑉 = −𝐼𝑛𝑣0 + ∑ = 401,620€
(1 + 𝑊𝐴𝐶𝐶)𝑖
𝑖=1

Payback time is less than two years considering an algorithm with 80% precision

and recall. By entering the extreme situations of 50% and 100% for both variables, we

obtained an NPV of 158,172 € and 691,918 € respectively. We would like to remind that
 146 Chapter 7. Model Application

these results are achievable by respecting the various starting hypotheses. If the

application of the Industrial IoT predictive maintenance solution was implemented

independently from the manufacturer, the costs he may incur will be different. For

example, it would require the training of internal staff on the new technology and a

continuous investment along the years. The other major hypothesis made is not to

considerate any buffer between the machines: a lack of production leads to a loss of

sales (unless flexibility given by other production lines, managed with the term 𝐵𝐾%

and 𝑂𝐸% ). This consideration is part of the downtime cost, represented then by an

opportunity cost and by the Mean Time To Repair, which are obviously central

variables in the estimates. On the costs of the intervention considered in this concept

scenario, the downtime cost has a weight from 65 to 90%.

The variability given by the difficulty in defining the Precision and Recall was

studied in a sensitivity analysis. Generally, we can expect that the algorithms will

become more and more accurate as the quantity and quality of the available data

increases.

In this concept scenario, a transition to predictive maintenance from a corrective

as-is is certainly to be taken into account. On the other hand, when as-is is preventive,

various evaluations will have to be examined as the economic benefits do not allow

any margin of error.

The validation of these results comes both from the literature and from the

meeting with managers. In the former, although analytical models are almost absent,

some estimates are present: savings provided by the shift from corrective and

preventive to predictive (40%-50%, 8-12% respectively), return on investment (10

times) and downtime reduction (up to 35%) (U.S Department of Energy, 2010). These

values are comparable to those obtained in the two applications conducted. As far as

managers are concerned, a positive opinion was received regarding the evaluation of

the cost structure and the final NPV obtainable.

 7.3 Real Cases Application: ConBio 147

7.3 Real Cases Application: ConBio

In this section, the model is tested in a real situation. We conducted several

interviews with a manager from ConBio srl. to obtain all the data we needed to

implement the cases that reproduce reality as accurately as possible.

About the Company

Recently merged into the Granarolo S.p.A group, ConBio was born as a

pioneering activity in the production of biological and vegan ready meals in 1998

thanks to the intuition of its founders who advocate and encourage what is today a

thriving and booming market. ConBio products respond to the need for human well-

being and environmental sustainability: the company uses only 100% organic and

vegetable ingredients, preferably of Italian origin. All products are produced

internally, directly following all the steps from the raw material to the packaged

product, establishing deep and lasting relationships with all the players in the supply

chain, in particular for basic raw materials such as soy and flours. The company

distributes its products under three brands and manufactures private label

productions for some of the main players in the large-scale retail trade (GDO).

Layout

ConBio has its headquarter situated in Rimini. They produce many different

products in small batches to satisfy a very fragmented market demand among the

different actors. Therefore, the production layout must adapt to this high degree of

flexibility imposed by the market, react quickly to any changes in the quantities

requested and in the mix and be able to support the production of new products

provided by the R&D department. This flexibility is guaranteed by a job shop layout.
 148 Chapter 7. Model Application

After a careful analysis of the layout of the company, we identified which

machines we believe would be appropriate to implement predictive maintenance to

avoid excessive downtime due to breakages that could cause delays for many

products. Our analysis underlined that the production of many products depends on

some critical machines. This is the case of the decanter, a machine once used typically

in the field of oil production for the separation of oil from water. Today no longer in

use in this sector, it is employed by ConBio to process soybeans to make tofu (a

vegetable protein) which is then reused in many products, as well as being sold in

bulk.

The Flow on the Machine

The production process for the preparation of tofu in ConBio starts from the

soybean. It is left to soak for a long time in containers in a special area (A) to soften

and rehydrate the dried bean. Once ready, an operator picks the beans up and takes

them to the work area B where they are loaded into a cutter, which chops the bean and

begins to heat it (machine 29). Once sufficiently crushed, they are taken and loaded

into a heat exchanger, which brings the liquid pulp obtained to 103/104 degrees

(machine 26). Once heated, the product is placed in the decanter (machine 38 and 39)

which separates the filtrate, then curdled to obtain the tofu, from the pulp that is

Figure 7.3: Layout of ConBio

 7.3 Real Cases Application: ConBio 149

reused in the production of some dishes. The curd filtrate that comes out of the

machine goes back into the production process with a tube and is transported on a

press (machine 27) to make the shapes. Once the shape has been obtained, the tofu is

brought to department C, where it is placed in a blast-chilling cell. Once the tofu path

has cooled down it is divided into two: on one side the tofu, which is sold without

further processing, is brought to department C, where it is cut by a portioning machine

(machine 6) and proceeds towards packaging, on the other, the tofu shape enters other

production cycles for the preparation of other products. The tofu line works 220 days

a year on three shifts of 6 hours each.

Maintenance Strategy – As-Is Situation

Among all the machines used in the tofu production cycle, we chose the decanter

as it was considered one of the most critical machines in terms of breakage. Two are

the most common types of breakage:

a. Shaft breakage: The machinery shaft break about once every two years. No

preventive intervention is implemented to avoid the breakdown to occur

therefore it perfectly fit the case of corrective maintenance;

b. Breakage of bearings and seals: this breakdown occurs much more frequently

but has much less impact in terms of downtime. Also in this case, even if the

Figure 7.4: Decanter

 150 Chapter 7. Model Application

seals is occasionally oiled, there is not a real maintenance schedule and

therefore this second cluster is considered a case of corrective maintenance.

As the entire tofu line, this machine works three shifts of 6 hours each. However,

setups and other inefficiencies reduce its operation to 16 hours.

Cluster 1 – Shaft

The breakdown of the shaft is the most serious break that can happen to the

machinery and stop the production for about seven working days. This long stop is

due to the fact that the process for restoring the machinery is very long and complex

and involves more people during its course. Let's see it in detail. When the piece breaks

an external, non-specialized technician with whom the company has to deal with

various maintenance interventions is called to intervene urgently. The technician,

assisted by an internal staff member, disassembled the machinery and tried to

understand if the fault could be adjusted on-site or by the company of which the

maintenance technician is part. If the repair could be carried out directly by them,

generally, the repair lasted a day, otherwise, it is necessary to call the Italian distributor

of the vendor, situated in Bologna, which sent a technician, if available, the following

day. The specialized technician finishes to dismantling the machine and make his

diagnosis. He disassembles the piece completely, takes it to Bologna34and contacts the

34 The manager we spoke with stated that the technician is used to take the piece
to the distributor in Bologna to try to fix it, but even if they managed to repair or they
require a new product the time is always the same and the price is similar. Therefore,
we assume that the piece is ordered every time.
 7.3 Real Cases Application: ConBio 151

manufacturer in Germany to have a new piece sent. After the piece arrives, the

technician returns to the ConBio factory and assembles the new part. Since most of the

time the repair required the entire cycle described above, we put ourselves in the

worst-case and assume the maximum time of seven days. completely different is the

case of a scheduled intervention that would only take one day to restore the

machinery.

The produced tofu is used for various products including natural tofu (bulk), tofu

cutlet, grilled tofu and soy burgers35. Other products use this raw material but have

not been considered in the calculation because they generally employ tofu in such a

low quantity that production, in case of decanter breakdown, is not stopped because

stocks are used.

Figure 7.5: An example of the data insert in the model from Cluster 1

Cluster 2 – Bearings and Seals

The second cluster that we have formed belongs to the machine's bearings and

seals, which break twice a year. We put these two elements together because the costs

of breakdown linked to them are very similar. This time a possible unscheduled

breakage of one of these two pieces stops the machine for only about six hours.

Particularly critical is the case of bearings in which the bearing balls spread throughout

the machine and, if not stopped in time, can cause breakage of other pieces including

the shaft. After stopping the machine to avoid further damage, the aforementioned

external not specialized maintenance company is called, who repairs the machinery.

35 The unitary gross profit values are not shown for privacy reasons.
 152 Chapter 7. Model Application

Gross profit follows the same logic previously described and therefore the overall

value is the same.

Benefits Evaluation

By evaluating the transition from corrective to predictive maintenance applying

our model, it appears that predictive maintenance has a significant economic

advantage, expressed in annual savings, compared to a corrective.

Table 7.3: Cost of the initial solution and savings passing to a predictive maintenance
solution.

The percentages of improvement for the cluster 1 and 2 are respectively of 49,3%

and 56,3%. Assuming to activate a subscription with the OEM based in Bologna with

a value of 50 cents per hour of machine uptime, the total annual benefit is € 24,388.93.

It was assessed together with the company manager that the investment necessary to

install the sensors and the IoT system for remote monitoring should be around € 5,000.

Assuming a useful life of the machinery of 30 years (the machinery has never been

changed in 20 years of production and has an almost infinite useful life) and the

company's WACC to date of 4.75%, the corresponding NPV is € 380,843 € and the

investment would payback from the first year.

Table 7.4: The final output of the model for the ConBio example
7.3 Real Cases Application: ConBio 153
 Chapter 8

8 Conclusions and Discussions

A cost-benefit analysis of an Industrial IoT application that enables the predictive

maintenance service in a pay-per-performance contract was discussed in this

dissertation. The main variables of the model were reported, underlining the topic of

collaboration between the manufacturer and vendor, and reporting possible final

outputs.

Interests in the topic of predictive have grown exponentially in recent years,

intertwined with the availability of data now made accessible by Industrial IoT. The

literature has grown from about 200 to almost 700 published paper per year in the last

five years. The literature is still highly focused on technical aspects (hardware or

software development), and the benefits are discussed qualitatively; when they are

considered quantitatively, only little depth analyses are conducted. There is a need for

a cost sounding model that starts to bring clarity on overall benefits. The survey on

small and medium enterprises highlighted a need for understanding the Industrial IoT

theme at all levels. The other survey on large enterprises brings positive results in

some respects, showing several active projects in the predictive field. Nevertheless, at

the moment of a direct interview, the prevalence of predictive application is re-

evaluated: predictive maintenance is still minimal, in some circumstances not well

understood. Excellent cases are however present, among which we have reported

Fabio Perini's case study. After an initial period of offering services related to the
 156 Conclusions and Discussions

monitoring of machinery in pay-per-performance formulas, the company is beginning

to offer a predictive maintenance service. The interest in the subject is relevant from

both the scientific and the practitioner side. Significant changes are imminent for

maintenance, which is not bound only to the cost of the maintainer but keeps in mind

the ability to reorganise with production and warehouse processes. Integrating

maintenance is key to achieving high levels of efficiency and productivity.

Our attempt is aimed at bridging the gap in the literature regarding benefit

evaluation by proposing an analytical model with a flexible structure. One hypothesis

of the model is that the predictive maintenance analysis, which is in many ways

complex, is carried out by the vendor who can potentially have access to all the

different machines being sold (once the manufacturers have granted authorisation for

data processing). We wanted to link the maintenance service to a pay-per-performance

agreement to highlight where the vendor can have its economic return, as well as

having obtained a competitive advantage over those who do not offer this service. The

theme of a closer relationship is becoming more and more present. The use of the

model is therefore addressed both to the manufacturer (to estimate the investment)

and to the vendor (to adapt its offer better).

The tool used to conduct the analysis is the NPV. To calculate the cash flows, a

differential between the costs of the previous maintenance and those of the predictive

was considered. These costs have a simple structure: a mutual part between all types

of maintenance and another part specific. In the first one, the cost per intervention is

modelled, where the cost variables of downtime, materials used and lack of customer

service are present. A focus on the cost of downtime is appropriate; it is a function of

the Mean Time To Repair (a key concept in maintenance), the gross profit of the

machined products, and the percentage of slow down/stop of the production flow due

to breakdown (the presence of buffers and alternative work cycles can significantly

limit this effects). The cost of the intervention will be calculated in two cases:

unscheduled and scheduled. The unscheduled one is by its nature more expensive.

However, the ability to react quickly leads to a decrease in the distance between the

two costs. This concept has to be combined with the number of interventions made for

each maintenance (Figure 8.1)

 Conclusions and Discussions 157

Figure 8.1: The number of interventions is shown on the y-axis. The colours blue and
orange represent when the intervention is unscheduled or scheduled respectively.

The number of annual interventions undertaken in corrective maintenance

corresponds to the number of faults per year (represented by the failure rate), which

is the minimum number of interventions to be carried out. The preventive time based,

to deal with the same number of faults expected, implements several interventions at

fixed intervals; nevertheless, a separate quota of unscheduled interventions can be

considered to take into account the result of an optimisation method or the possible

inaccuracy of a fix intervention. The predictive maintenance is in the middle, trying to

anticipate faults exactly. Fundamental to describe this capability are the variables

Precision and Recall. For the proposed model, it is necessary to have an estimation of

these variables, which is based on the number of failures predicted with a time interval

T in advance, large enough to be able to organise a scheduled intervention. The

difficulties in the evaluation of the algorithm performance can be severe, and they

could not always be overcome. These considerations lead to the need to analyse

different scenarios for the different values of Precision and Recall, performed in the

form of sensitivity analysis. Ultimately, in the specific part for the type of maintenance

approach, the cost of labour/maintenance service and the cost of the warehouse have

been considered.

The topic of predictive maintenance is vast, and some aspects have not been

considered in the model. The increase in the useful life of the equipment is an example,

which in some cases, could represent an important factor in the implementation of a

 158 Conclusions and Discussions

predictive service. Other evaluations not considered, concern redundant

parts/machines: an improvement in availability resulting from a predictive

maintenance approach can change the choice of whether or not to include redundancy

in the system, thus bringing as savings the cost delta between the two situations.

Lastly, it may be useful to address a possible risk (and opportunity) assessment, such

as the financial risk assessment caused by the breakdown. These further considerations

could represent some interesting follow-up to be explored.

In the model, what could constitute the main future development is an extended

calculation to include all maintenance strategies. Next, other types of intervention cost

could be added, in addition to scheduled and unscheduled types, as can be

represented by a quick intervention; it could be arranged to be more precise in the

evaluation of the benefits of the predictive (or possible condition-based) approach that

quickly notice out-of-normal parameters to have to be corrected immediately. The goal

is to create a general model that evaluates the optimal maintenance solution. In this

dissertation, we have chosen to evaluate a to-be represented only by predictive

maintenance. This choice was taken to remain focused on the object of the thesis, an

Industrial IoT application that enables the predictive maintenance service.

The model's hypothesis of assuming predictive maintenance as a service reduces

its usability so far considering this offer still little present. A maturity of this offer may

still require a certain amount of time. However, it should be pointed out that many

resistances of data sharing by the manufacturer are disappearing (reported in the case

of Fabio Perini). This aspect represents the correct starting point for the development

of the predictive maintenance service.

In addition to the analysis that guided the structure of the model, a conceptual

scenario and a real case (represented by the company ConBio, a vegan ready meal

producer part of the Granarolo Spa group) have been considered. We are now ready

to answer the research questions.

 Conclusions and Discussions 159

Q1: Does an investment in Industrial IoT for predictive maintenance generate value
for the company?

Each case must be evaluated; the benefits will be calculated thanks to the

presented model, flexible enough to be adapted to different situations. From the tests

conducted, a shift from a corrective to a predictive approach leads to a positive NPV,

considering even deficient prediction performance. A payback time in the short term

(less than two years) is obtained. Regarding a shift from preventive maintenance, a

positive result is not sure, and further assessments of benefits are to be taken into

account.

The validation of these results comes from both the literature and the meeting

with managers. In the former, although analytical models are almost absent, some

estimates are present: savings provided by the shift from corrective and preventive to

predictive (40%-50%, 8-12% respectively), return on investment (10 times) and

downtime reduction (up to 35%) (U.S Department of Energy, 2010). These values are

comparable to those obtained in the two applications conducted. As far as managers

are concerned, a positive opinion was received regarding the evaluation of the cost

structure and the final NPV obtainable. This investment can be considered part of the

path of digitalisation of the company, and it can be envisioned as part of that overall

project.

Sub. Q1: How can the vendor benefit from the value generated to the manufacturer?

The vendor, which takes on a primary role, has as a direct return the pay-per-

performance subscription. Secondly, being on the market with such an offer can lead

to considerable competitive advantages; he has the potentiality to scale the solution to

all his clients. In a final perspective, the large amount of data collected, in addition to

being used for maintenance service purposes, can have applications in the context of

smart lifecycle.
 160 Conclusions and Discussions

Significant changes are imminent for maintenance sector, where over the total

maintenance resource and activities of an average facility, the percentage regarding

the predictive is expected to grow from 12% to 45%/55% (U.S Department of Energy,

2010). These changes will impact the industrial sector in the following years, and, in

this dissertation, we aimed to be promoters of knowledge related the future

developments allowed by the Industrial IoT, and more specifically, to foster the

interest in understanding and considering the development of a predictive

maintenance solution.
Appendix A

An Introduction to the

Internet of Things
The appendix examines the new technological paradigm of IoT, a paradigm with

the potential to radically change the way we live. The main characteristics of the

interconnected intelligent objects and the IT architecture necessary for their correct

operation are therefore explained. In the end, the main applications areas of the

Internet of Things are summarized.

Brief introduction to the new paradigm

An increasing number of devices will be able to access the Internet and actively

interact with the web. The new reference scenario, which is revolutionizing the

consumer and business world, has been called the Internet of Things (IoT). The

potential benefits of the new paradigm are undoubtedly extraordinary, and its

applications are radically changing work and private life habits, saving time and

resources, and opening new opportunities for growth, innovation and knowledge

development.

As an essential tool to reach and connect millions of objects and acting as a

generic enabler of a new hyper-connected society, the IoT has all the potential to help

the Western society that grows older, to improve the energy efficiency of cities and to

optimize mobility and transport. The complementarity with "cyber-Physical Systems",

cloud technologies, Big Data and new network technologies like 5G, is then evident.
 164 Appendix A

IoT definition

To provide an adequate definition of the IoT paradigm, three specific and

different IoT definitions of three of the most accredited expert bodies on the subject in

the Italian, European and international fields were taken into consideration. The

‘Osservatorio Internet of Things’ of the Politecnico di Milano: "The expression

“Internet of Things” indicates a path in technological development based on which,

through the internet, potentially every object of our daily experience acquires its own

identity in the digital world. It is a structured route, characterized by countless fields

of application and from different - for variety and dynamism - enabling technologies."

(Osservatorio IoT - Politecnico di Milano, 2015).

The European Research Cluster on the Internet of Things (IERC), supported by

the Commission European which facilitates the sharing of knowledge on the IoT and

supports and supports best practices and new business models in this regard: "A

dynamic global network infrastructure with self-configuring capabilities based on

standard and interoperable communication protocols where physical and virtual

’things’ have identities, physical attributes and virtual personalities, and are

seamlessly integrated into the information network." (IERC, 2014)

The Internet Engineering Task Force (IETF), an international organization

founded in 1986 and composed of an open community of network designers,

operators, vendors and interested researchers to the technical and technological

evolution of the Internet. "[...] IoT will connect objects around us (electronic, electrical,

non-electrical) to provide seamless communication and contextual services provided

by them." (IETF, 2010)

 Appendix A 165

The motivations that led to the pervasiveness

of the IoT paradigm

The convergence in recent years of a series of innovations within the broad

technological scenario has pushed the diffusion of the IoT, not leaving it confined

within market niches. Among these innovations must certainly be mentioned: major

improvements in performance, miniaturization and energy efficiency of sensors and

batteries; the new systems for processing and storing extraordinarily compact data

economically accessible, allowing to turn individual objects into small computers;

antennas and low-cost wireless ubiquitous connectivity; the tools that enable rapid

and "agile" prototyping of the software; the so-called Big Data analytics; the new IPv6

Internet registration system that allows activating, in addition to those already

existing, 340 trillion addresses that can be associated with a single device; the new

protocols designed to guarantee greater security of information, which make easier to

pass a device between different networks and allow servers to automatically delegate

and address without the need for IT support.

All these factors have made connected smart objects cheaper (technically and

economically) for both enterprises and final consumers (Porter & Heppelmann, 2014).

A brief historical background

To realize a synthetic framework of the Internet of Things, it is necessary to start

from RFID (Radio Frequency Identification) technology, the basis of the subsequent

development of the IoT paradigm. RFID, the simplest technology with which an object

can enter the IoT, is based on the ability of small tags (or transponders), constituted of

a chip and an antenna, to store transmitted information from the outside that can then

be read by other objects, called readers, able to read them at a distance using radio

waves. In the ’50s and ’60s, several studies were carried out on the remote

identification of objects through radiofrequency energy. Between the 60s and 70s of

the last century, the cold war and the nuclear race between the United States and the
 166 Appendix A

Soviet Union have pushed the development of RFID technology even further and laid

the foundations for the birth of the internet. Another important period for the

development of this technology was the early ‘90s when the World Wide Web was

born thanks to the publication of the first site by Tim Berners-Lee. The IBM engineers

developed for the first time a UHF (ultra-high frequency) RFID system with one better

reading skills and faster data transfer. The development of UHF-type RFIDs dated

back to 1999, when the Uniform Code Council, the European Article Number (EAN),

Procter & Gamble and Gillette have decided to finance and establish the Auto-ID

Center in Massachusetts Institute of Technology (MIT) to study how to exploit low-

cost RFID tags to track any produced throughout the Supply Chain. It was Kevin

Ashton (director for many years of that research centre) to indicate himself as the one

who had coined the "Internet of Things" neologism in 1997 within a publication of the

International Telecommunication Union (ITU).

Intelligent interconnected objects and their

characteristics

Devices that are part of an interconnected network are called “smart objects” or

“smart things” because, in addition to being constituted by traditional physical

components, that materially constitute the object, they also present "intelligent"

components and connectivity components. The connectivity amplifies the potential of

intelligent components, thus creating a virtuous circle of incremental generation of the

value of the object.

Intelligent components and enabled properties

Intelligent components are all sensors, actuators, microprocessors, control

systems data storage, software and controls for the user interface on the device. These
 Appendix A 167

"smart" components enable three particular properties of the product (Osservatorio

IoT - Politecnico di Milano, 2015), not necessarily co-present:

1. "Self-awareness". This characteristic can be summarized in the ability of the

object of being able to be uniquely identified (the digital code of a product

must be unique) and universally in the digital domain (understanding of the

ID worldwide), in the ability of being able to be located in real-time (for

example through GPS technology) and the possibility to check the operating

status (with detail levels of the internal systems more or less accentuated).

2. The smart objects have three different capabilities in relation to the type of

information received: metering capability, that is the automatic measurement

and recording of a flow of information on a particular physical quantity such

as cubic meters of water or gas used, kilowatts hour of energy consumed,

miles travelled for example; capability of sensing or receipting certain

information from outside that are not a continuous flow over time (for

example the perception of movement, fumes in the environment, brightness);

ability to implement or perform a specific action that changes the status of an

element (for example, switch off the light, close the shutters, activate the

alarm).

3. The third is instead the possibility to process information obtained from the

outside through predetermined algorithms, implementable at a time after the

acquisition. Data, for example, can be aggregated, filtered, crossed, correlated

etc. to improve and increase knowledge of the surrounding environment by

the device (and therefore of all those who can access it such information).

Connectivity components

For connectivity components, reference is made instead to the antennas, to the

doors and to the protocols that make the connection with the device possible.

Communication may occur individually or simultaneously and generally takes three

different forms (Porter & Heppelmann, 2014):

 168 Appendix A

 "One-to-one": a single object connects with another single entity (the user,

the manufacturer or another object).

 "One-to-many": according to a central-star configuration, a central access

point communicates at the same time with two or more devices in

continuous or intermittently.

 "Many-to-many": more products are connected both to each other and

possibly also to sources external data. Connectivity, in its three different

forms, must be implemented together to reach highs product functionality

levels. Firstly, it allows the exchange of data and information between the

reference object and the external environment (such as users, producer, other

intelligent products, operating systems). Secondly, it makes possible the

existence of some features external to the product in the so-called "Product

Cloud" (definition provided by Michael Porter), that is software that runs

remotely, on the company’s server or an external partner.

The four main capabilities of intelligent objects

The components of smartness and connectivity have enabled a completely new

range of functionality and capabilities of the products. It is possible to classify them

into four main groups (Porter & Heppelmann, 2014):

1. Monitoring: connected intelligent objects can be remotely monitored in a

complete and historicized way. Final users or companies are able in this way

to know their working criteria, operations, as well as the conditions of the

external environment. Furthermore, the product can send reports or

warnings in case of particular changes in the parameters sensed.

2. Control: remote commands or software integrated into the product or in the

cloud, can remotely control the various functions of the smart objects. Thanks

to the embedded algorithms, the object (and its actuators) react in case of

specific changes. This feature enables personalization of the performance, in

ways that in the past were not economically viable or possible.

 Appendix A 169

3. Optimization: the previous two capabilities allow the creation of algorithms

that optimize the operation and use of the object. In this way, it is possible to

improve the performance of the product itself and to programme in advance

maintenance services, assistance and timely repairs (through software).

4. Autonomy: monitoring, control and optimization, if combined, can allow

interconnected intelligent objects to reach levels of autonomy that in the past

were hardly achievable except through unsustainable economic efforts.

Through sensors and embedded software, the product can "become aware"

of the external environment in which it is located and make decisions without

human intervention. This refined feature allows the smart devices also to

detect problems and to adapt, thanks to the understanding of the data and

information received, to the needs of the client. Each capacity integrates and

enriches the one that precedes it. Theoretically, an intelligent, interconnected

object could have all four.

A difficult convergence on the standards

Even today, the world of the Internet of Things is very fragmented and not

standardized in terms of:

 Hardware (Mica, Motes, Sunspot, Jennic, etc.);

 Software (Tiny OS, SOS, Mantis, Contiki, FreeRTOS, etc.);

 Middleware (Tiny DB, GNS, DNS, SWORD, etc.); Communication under the

point of view of the nature of technology (radio, wired, mobile phone) and of

protocols (eg ZigBee, Bluetooth LE, Wi-Fi, etc.). (Osservatorio IoT -

Politecnico di Milano, 2015)). Often, due to the variety characterizing the

Internet of Things ecosystem, the development for the generation of IoT

solutions followed mainly a "vertical" approach: starting from one specific

application problem, an ad hoc design of hardware, operating system,

application software, middleware, and communication standards for the first

two levels of network architecture, is conduced. The aim is substantially to

optimize the specific application (for example in terms of performance or

 170 Appendix A

energy savings) instead of abstracting the design from the single problem,

allowing the implementation and use of a wider range of applications. The

IoT architecture of the near future makes of the interoperability at single

product level its foundation. In this way, the access and exchange of

information between devices becomes free and direct switching to

application level, without the presence of a hub control centre is therefore

possible. By exploiting this logic, each application can communicate and

command the objects, consequently abstracting from the specific application

problem.

IoT Application Fields

The various fields of application of the Internet of Things, defined within the

research carried out in (Osservatorio IoT - Politecnico di Milano, 2015), are the

following:

 Smart City & Smart Environment: with the aim of improving liveability,

sustainability and competitiveness of a city, IoT enables the monitoring and

management of the elements such as vehicles for public transport, public

lighting, parking lots, and the surrounding environment (for example rivers,

woods, mountains).

 Smart Metering & Smart Grid: the so-called smart meters for measuring

consumption (electricity, gas, water, heat) exploit the IoT paradigm in order

to improve billing and remote management; "Intelligent" electricity grid

(Smart Grid) is instead an interesting field in order to optimize distribution

and to manage distributed production and electric mobility.

 Smart Home & Building: for energy saving, comfort, safety purposes, IoT

objects give the users the possibility to automatically manage systems and

building equipment (for example those for lighting, air conditioning and

appliances).

 eHealth: thanks to real-time of remote monitoring of vital parameters, it is

possible to reduce the recourse to hospitalization, diagnosis and care.

Appendix A 171

 Smart Car: by connecting vehicles (among them or to the surrounding

infrastructure), it is possible to detect and prevent accidents and also to open

new opportunities for innovative insurance model and for georeferenced

information on the road traffic.

 Smart Logistics: IoT has an interesting potential in improving supply chain

traceability, brand protection, cold chain monitoring and security in complex

logistics poles.

 Smart Asset Management: IoT enables the remote control to manage

breakdowns, tampering, localization, traceability and inventory remote

management of valuable assets.

 Smart Factory: new production management, supply chain planning logic

and product life cycle management are some of the industrial application

fields to which IoT (through Cyber-Physical Systems, connection of

machinery, operators and products) gave the birth.

 Smart Agriculture: lastly, IoT can also support agriculture. By monitoring the

micro-climatic parameters, it is possible to improve the quality of products,

to reduce both the resources used and the environmental impact.

Appendix B

Survey Analysis
The results of the two surveys conducted inside the ‘Osservatorio Internet of

Things’ of the Politecnico di Milano and presented in the thesis are grouped. They both

concern the Italian manufacturing sector, one looking to Large Enterprises (LEs) and

the other to Small-Medium Enterprises (SMEs).

Regarding the LEs survey, in October 2019, 714 questionnaires were sent to

companies operating in the Italian industrial context to understand their point of view

on the Industrial IoT theme. The research was carried out using the CAWI

methodology (Computer Assisted Web Interview).

Regarding the SMEs survey, in December 2019, a sample of 525 small and

medium enterprises, representative of the Italian scenario by sector, geographical area,

number of employees and turnover, was carried out. The research was conducted

using the CATI methodology (Computer Assisted Telephone Interview).

The aim of both the researches was to collect data about

1. Level of knowledge regarding IoT field;

2. The area of completed/underway projects, and related services added;

3. Objectives, benefits, and use of data;

4. Barriers and how to manage them;

5. Economic dimension of the investments done in Industrial IoT, and the area

of interest in future projects.

The results have been compared between the two. In caption, the reference

sample.
 174 Appendix B

Q 1.1: Have you ever heard of Internet of Things (IoT) solutions for Industry 4.0?

Figure A.1: Results of Q1.1 - Sample size LEs, 100. Sample size LEs, 525.

Q 1.2: What is your level of knowledge of IoT solutions for Industry 4.0?

Figure A.2: Results of Q1.2. - Sample size LEs, 97. Sample size LEs, 218. The mean in
the survey LEs and SMEs corresponds to 6.4 and 6.2.
 Appendix B 175

Q 2.1: Has your company initiated IoT projects for Industry 4.0 in the past?

Figure A.3: Results of Q2.1. - Sample size LEs, 97. Sample size LEs, 218.

Q 2.2: Indicate, for each IoT project for Industry 4.0 launched by your company,

the state of progress (proof of concept, pilot project, executive project).

Figure A.4: Results of Q2.2 related to Smart Factory - Sample size LEs, 90. Sample
SMEs, 31 – Indicate the category (and sub-classes) for each IoT project for Industry 4.0
launched by your company (pilot project and executive project). The graph indicates how the
applications inside the Smart Factory category are distributed between the sub-classes.
 176 Appendix B

Figure A.5: Results of Q2.2 related to Smart Supply Chain - Sample size LEs, 41. Sample
SMEs, 23. - Indicate the category (and sub-classes) for each IoT project for Industry 4.0
launched by your company (pilot project and executive project). The graph indicates how the
applications inside the Smart Supply Chain category are distributed between the sub-classes.

Figure A.6: Results of Q2.2 related to Smart Lifecycle - Sample size LEs, 24. Sample
SMEs, 15 - Indicate the category (and sub-classes) for each IoT project for Industry 4.0
launched by your company (pilot project and executive project). The graph indicates how the
applications inside the Smart Lifecycle category are distributed between the sub-classes.
 Appendix B 177

Q 3.1: What were the main objectives that led the company to launch IoT projects

for Industry 4.0? Up to three options can be entered.

Figure A.7: Results of Q3.1 - Sample size LEs, 61. Sample SMEs, 47.
 178 Appendix B

Q 4.1: What are the barriers (internal and external) that in your opinion can slow

down or prevent the start of IoT projects for Industry 4.0? Up to three options can be

entered.

Figure A.8: Results of Q4.1 - Sample size LEs, 88. Sample size SMEs, 525.
 Appendix B 179

Q 5.1: What additional services enabled by the Internet of Things technologies is

the company you are part of interested in activating inside the factory? Up to three

options can be entered. (Have the IoT applications for Industry 4.0 that you have

launched included additional services?).

Figure A.9: Results of Q 5.1 - Sample base regarding past project, 59. Sample base
regarding future project, 83 - The answers are scaled for the answers received regarding the
future.
Bibliography
2030 Vision for Industrie 4.0. (2019, October). Retrieved from Platform-i40:

https://www.plattform-

i40.de/PI40/Redaktion/EN/Downloads/Publikation/Vision-2030-for-

Industrie-4.0.pdf?__blob=publicationFile&v=8

Adu-Amankwa, K., Attia, A. K., & Janardhanan, M. N. (2019). A predictive

maintenance cost model for CNC SMEs in the era of industry 4.0. Int J Adv

Manuf Technol 104, 3567–3587.

Agrawal, M. (2016, May 3). Business Models in Internet of Things. Retrieved from

Linkedin: https://www.linkedin.com/pulse/internet-things-business-models-

mohit-agrawal/

AIDA. (2019, December). Report Fabio Perini S.p.A. Retrieved from aida:

https://aida.bvdinfo.com/version-

2020512/Report.serv?_CID=271&context=28HMI0CXDK057WV&SeqNr=0

Alzghoul, A. (2014). Comparing a knowledge-based and a data-driven method in

querying data streams for system fault detection: A hydraulic drive system

application. Elsevier.

AmazonRobotics. (n.d.). Retrieved from Amazon Robotics:

https://www.amazonrobotics.com/#/

Andrews, E. (2019, October 28). Who invented the Internet? Retrieved from History:

https://www.history.com/news/who-invented-the-internet

Antràs, P., & Voth, H.-J. (2003). Factor prices and productivity growth during the

British industrial revolution. In Explorations in Economic History. Vol. 40 (pp.

52-77). Elsevier Science (USA).

Atzori, L., Iera, A., & Morabito, G. (2010, December 10). The Internet of Things: A

Survey. Computer Networks, pp. 2787–2805.

 182 Bibliography

Basson, M. (2017, March 17). P-F Curve Explained! Retrieved from

https://www.linkedin.com/pulse/p-f-curve-explained-marius-basson/

Baur, M., Albertelli, P., & Monno, M. (2019). A review of prognostics and health

management of machine tools. The International Journal of Advanced

Manufacturing Technology.

Baur, M., Albertelli, P., & Monno, M. (2020). A review of prognostics and health

management of machine tools. Int J Adv Manuf Technol 107, 2843–2863.

Bengtsson. (2004). Användandet av tillståndsbaserat underhåll i svensk industri – en.

Sweden.

Bevilacqua, M., & Braglia, M. (2000). The analytic hierarchy process applied to

maintenance strategy selection. Reliability Engineering & System Safety Vol. 70,

71-83.

Bloch, H., & Geitner, F. (2012). Machinery Failure Analysis and Troubleshooting: Practical

Machinery Management for Process Plants. Butterworth-Heinemann.

Bokrantz, J., Skoogh, A., Berlin, C., Wuest, T., & Stahre, J. (2020). Smart Maintenance:

an empirically grounded conceptualization. Int. J. Production Economics 223.

Bousdekis, A., Apostolou, D., & Mentzas, G. (2020). Predictive Maintenance in the 4th

Industrial Revolution: Benefits, Business Opportunities, and Managerial

Implications. IEEE Engineering Management Review.

Boyes, H., Hallaq, B., Cunningham, J., & Watson, T. (2018). The industrial internet of

things (IIoT): An analysis framework. Computers in Industry, 1-12.

Brynjolfsson, E., & McAfee, A. (2014). The Second Machine Age: Work, Progress, and

Prosperity in a Time of Brilliant Technologies. W.W. Norton & Co.

Calenda, C. (2017). Piano Nazionale Industria 4.0. Retrieved from Ministero dello

Sviluppo Economico (MISE):

https://www.mise.gov.it/images/stories/documenti/guida_industria_40.pdf
 Bibliography 183

Capone, A., Pitic, R., Tumino, A., & Salvadori, G. (2018). L'evolutione delle tecnologie IoT:

Assistenti vocali e sensoristica in primo piano.

CERN. (n.d.). The birth of the Web. Retrieved from CERN:

https://home.cern/science/computing/birth-web

Cognizant. (2014, June). Designing for Manufacturing’s ‘Internet of Things’. Retrieved

from https://www.cognizant.com/InsightsWhitepapers/Designing-for-

Manufacturings-Internet-of-Things.pdf

Columbus, L. (2019, February 3). Forbes. Retrieved from Top 25 IoT Startups To Watch

In 2019: https://www.forbes.com/sites/louiscolumbus/2019/02/03/top-25-iot-

startups-to-watch-in-2019/#6e467c673cc0

Compare, M., Baraldi, P., & Zio, E. (2019). Challenges to IoT-enabled Predictive

Maintenance for Industry 4.0. IEEE Internet of Things Journal.

Computer History Museum. (n.d.). Timeline of Computer History. Retrieved from

Computer History Museum, CHM:

https://www.computerhistory.org/timeline/1964/

Connect Advisory Forum, European Commission. (2014). Internet of Things the next

revolution: A strategic reflection about a European approach to the Internet of

Things.

Desjardins, J. (2017, November 28). Cybersecurity: Fighting a Threat That Causes $450B of

Damage Each Year. Retrieved April 2020, from Visual Capitalist:

https://www.visualcapitalist.com/cybersecurity-fighting-450b-damage/

Diez-Olivan, A., Del Ser, J., Galar, D., & Sierra, B. (2019). Data fusion and machine

learning for industrial prognosis: Trends and perspectives towards Industry

4.0. Information Fusion, Vol. 50, 92-111.

Digital4. (2017, June 5). Alfa Romeo, lo stabilimento di Cassino verso l'industria 4.0.

Retrieved from NetworkDigital360: https://www.digital4.biz/supply-

chain/operations-e-plm/alfa-romeo-lo-stabilimento-di-cassino-verso-l-

industria-40/
 184 Bibliography

Durán, O., Afonso, P. S., & Durán, P. A. (2019). Spare Parts Cost Management for Long-

Term Economic Sustainability: Using Fuzzy Activity Based LCC.

Sustainability - MDPI.

Evans, P., & Annunziata, M. (2012, November 26). Industrial Internet: Pushing the

Boundaries of Minds and Machines. General Electric Report.

Freeman, C., & Louçã, F. (2010). As Time Goes By: From the Industrial Revolutions to the

Information Revolution. Oxford University Press: New York, 2010.

Gartner. (n.d.). Gartner Glossary: Advanced Analytics. Retrieved from Gartner:

https://www.gartner.com/en/information-technology/glossary/advanced-

analytics

Geitner, F., & Heinz, B. (2006). Maximizing Machinery Uptime. Elsevier.

Ghosh, I. (2019, November 07). This is the crippling cost of cybercrime on corporations.

Retrieved from The World Economic Forum:

https://www.weforum.org/agenda/2019/11/cost-cybercrime-cybersecurity/

Gordon, R. J. (2000). Interpreting the "One Big Wave" in U. S. Long-term Productivity

Growth. In B. van Ark, S. K. Kuipers, & G. H. Kuper, Productivity, Technology

and Economic Growth (pp. 19-65). Kluwer Publishers.

Gordon, R. J. (2012, September 11). Is US economic growth over? Faltering innovation

confronts the six headwinds. Retrieved from VOX CEPR Policy Portal:

https://voxeu.org/epubs/cepr-reports/us-economic-growth-over-faltering-

innovation-confronts-six-headwinds

Gupta, V., & Simmons, D. G. (2010). Building the Web of Things with Sun SPOTs.

Retrieved from Sun Labs, Oracle:

https://sunspotdev.org/S314730_Sun_SPOTs_Web_Of_Things/S314730-

BuildingWebOfThingsWithSunSPOTs.pdf

Hankel, M., & Rexroth, B. (2015). Industrie 4.0: The Reference Architectural Model Industrie

4.0 (RAMI 4.0), ZVEI: Die Elektroidustrie.

 Bibliography 185

Hecht, H. (2006). Prognostics for Electronic Equipment: An Economic Perspective.

Annual Reliability and Maintainability Symposium (pp. 165-168). IEEE.

Helmiö, P. (2017). Open source in Industrial Internet of Things: A systematic literature

review. Master's Thesis, School of Business and Management, Lappeenranta

University of Technology.

Höller, J. (2018, November 18). Untangling Industrial IoT networking. Retrieved from

Ericsson: https://www.ericsson.com/en/blog/2018/10/untangling-industrial-

iot-networking

IBM. (2016). As much as half of every dollar you spend on preventive maintenance is wasted.

Retrieved from IBM blog: https://www.ibm.com/blogs/internet-of-things/as-

much-as-half-of-every-dollar-you-spend-on-preventive-maintenance-is-

wasted/

ICT4Executive. (2014, September 18). Harvard Review: così l’IoT cambia la creazione e la

cattura del valore. Retrieved from ICT4Executive:

http://www.ict4executive.it/executive/approfondimenti/harvard-review-

cosi-l-iot-cambia-la-creazione-e-la-cattura-del-valore_43672153606.htm

IEA. (2009). Energy Technology Transitions for Industry: Strategies for the Next Industrial

Revolution.

IERC. (2014). 2014 Conference. Retrieved from IERC - International Energy Research

Centre: ierc.ie/events-listing/2014-conference/

IETF. (2010). Conference 2010. Retrieved from IETF: https://ietf.org/

Immerman, G. (2018, May 8). The Real Cost of Downtime in Manufacturing. Retrieved

from Machinemetrics: https://www.machinemetrics.com/blog/the-real-cost-

of-downtime-in-manufacturing

Industrial Internet Consortium. (2014, June 17). Industrial Internet Consortium.

Retrieved from Overview of the Industrial Internet Consortium:

https://www.iiconsortium.org/ma-

14/Industrial_Internet_Consortium_Information_Day_June_17_2014.pdf
 186 Bibliography

Industrial Internet Consortium. (2019). The Industrial Internet of Things. Volume G1:

Reference Architecture.

Industrial Internet Consortium. (n.d.). Industrial Internet Consortium & OpenFog FAQ.

Retrieved from industrial internet CONSORTIUM:

https://www.iiconsortium.org/IIC-OF-faq.htm

Jardine, A., Lin, D., & Banjevic, D. (2006). A review on machinery diagnostics and

prognostics implementing condition-based maintenance. Elsevier, 1483-1510.

Jia, X., Feng, Q., Fan, T., & Lei, Q. (2012). RFID Technology and Its Applications in

Internet of Things (IoT). pp. 1282-1285.

Kaur, K., Selway, M., Grossmann, G., Stumptner, M., & Johnston, A. (2018). Towards

an open-standards based framework for achieving condition-based

predictive maintenance. IOT '18: Proceedings of the 8th International Conference

on the Internet of Things, No. 16. Santa Barbara.

Komonen, K. (2002). A cost model of industrial maintenance for profitability analysis

and benchmarking. International Journal of Production Economics 79, 15-31.

Körber Tissue. (2019, December 6). Digital Tissue - I-IoT workshop presentation.

Milan.

Lamarre, E., & May, B. (2017). Making sense of Internet of Things platforms. Retrieved

from McKinsey: https://www.mckinsey.com/business-functions/mckinsey-

digital/our-insights/making-sense-of-internet-of-things-platforms

Laney, D. (2001). 3D Data Management: Controlling Data Volume, Velocity and

Variety.

Lueth, K. L. (2019, December 23). IoT Platform Companies Landscape 2019/2020: 620 IoT

Platforms globally. Retrieved from IoT-Analytics: https://iot-analytics.com/iot-

platform-companies-landscape-2020/

Manyika, J., Chui, M., Miremadi, M., Bughin, J., George, K., Willmott, P., & Dewhurst,

M. (2017, January). Harnessing automation for a future that works. Retrieved

 Bibliography 187

from McKinsey & Company: https://www.mckinsey.com/featured-

insights/digital-disruption/harnessing-automation-for-a-future-that-works

McGagh, J. (2014). Internet of Things World Forum 2014. Retrieved from Internet of

Things World Forum: https://www.iotwf.com/iotwf2014/sponsor_keynotes

McKinsey&Company. (2015). The Internet of Things: mapping the value beyond the hype.

Miragliotta. (2019). Advanced Planning - SCP 3.0 and 4.0 - Part 1. Politecnico di Milano.

Miragliotta, G. (2020, Marzo 26). Industrial Internet of Things: definizione, applicazioni e

diffusione. Retrieved from Osservatori.net digital innovation:

https://blog.osservatori.net/it_it/industrial-iot-definizione-applicazioni

Miragliotta, G. (n.d.). Supply Chain Planning: Information Technologies 3.0 and 4.0. In

Advanced Supply Chain Planning Lab (pp. 18-31).

Miragliotta, G. (n.d.). Supply Chain Planning: Information Technologies 3.0 and 4.0

(Part 2). Advanced Supply Chain Planning Lab.

Miragliotta, G., Macchi, M., & Terzi, S. (2019). Industria 4.0: la rivoluzione si fa con le

persone! Politecnico di Milano.

Müller, J. M., & Däschle, S. (2018). Business Model Innovation of Industry 4.0 Solution

Providers Towards Customer Process Innovation. Processes, 6(12), 260.

Nakagawa, T. (2006). Maintenance Theory of Reliability. Springer Science & Business

Media.

Nikolic, B., Ignjatic, J., Suzic, N., Stevanov, B., & Rikalovic, A. (2017). Predictive

manufacturing systems in industry 4.0: Trends, benefits and challenges. Vienna:

DAAAM.

Noureddine, R., Solvang, W. D., Johannessen, E., & Yu, H. (2020). Proactive Learning

for Intelligent Maintenance in Industry 4.0. Advanced Manufacturing and

Automation IX, 250-257.

 188 Bibliography

O’Donovan, P., Leahy, K., & Bruton, K. e. (2015). A data pipeline for PHM data-driven

analytics in large-scale smart manufacturing facilities. Journal of Big Data 2,

25.

Omar, Y. M., Minoufekr, M., & Plapper, P. (2019). Business analytics in manufacturing:

Current trends, challenges and pathway to market leadership. Operations

Research Perspectives, Vol. 6.

Osako, L. F., Matsubayashi, M. O., Takey, S. M., Cauchick-Miguel, P. A., & Zancul, E.

(2019). Cost evaluation challenges for Internet of Things (IoT) based

Product/Service-Systems (PSS). Procedia CIRP, Vol. 84, 302-306.

Osservatori Digital Innovation della School of Management del Politecnico di Milano.

(n.d.). Osservatori Digital Innovation della School of Management del Politecnico di

Milano. Retrieved June 2020, from Osservatori Digital Innovation:

https://www.osservatori.net/it_it/osservatori/osservatori

Osservatorio Industria 4.0 - Politecnico di Milano. (2019, June). Industra 4.0: la

rivoluzione si fa con le persone!

Osservatorio IoT - Politecnico di Milano. (2015). Retrieved from Osservatori.net Digital

Innovation: https://www.osservatori.net/ww_en/observatories/internet-of-

things

Osservatorio IoT - Politecnico di Milano. (2017, July). La valorizzazione dei dati

generati dall'Internet of Things.

Osservatorio IoT Politecnico di Milano. (2018). Attenda, la stiamo collegando all’oggetto

desiderato.

Oxford Economics. (n.d.). Global Industry Databanks. Retrieved from Oxford

Economics: https://www.oxfordeconomics.com/forecastsand-

models/industries/data-and-forecasts/global-industry-databank/benefits-

and-uses
 Bibliography 189

Paré, G., & Kitsiou, S. (2017). Chapter 9: Methods for Literature Reviews. In F. Lau, &

C. Kuziemsky, Handbook of eHealth Evaluation: An Evidence-based Approach.

Victoria: University of Victoria.

Polanin, J. R., Pigott, T. D., Espelage, D. L., & Grotpeter, J. K. (2019). Best Practice

Guidelines for Abstract Screening Large‐Evidence Systematic Reviews and

Meta‐Analyses. In Research Synthesis Methods.

Porter, M., & Heppelmann, J. (2014). How Smart, Connected Products Are

Transforming Competition. Harvard Business Review.

Prajapati, A., Bechtel, J., & Ganesan, S. (2012). Condition based maintenance: a survey.

Journal of Quality in Maintenance Engineering, 384-400.

Real Time Innovations Inc. (2015). Retrieved from Industrial Internet of Things, RTI

FAQ: https://info.rti.com/hubfs/docs/Industrial_IoT_FAQ.pdf

Reply. (n.d.). IoT trend research. The evolution of consumer IoT. Retrieved from Reply:

https://www.reply.com/en/topics/internet-of-things/the-evolution-of-the-

consumer-internet-of-things

Roda, I., Macchi, M., & Fumagalli, L. (2018). The Future of Maintenance Within

Industry 4.0: An Empirical Research in Manufacturing. Advances in Production

Management Systems. Smart Manufacturing for Industry 4.0, 39-46.

Rolls Royce. (n.d.). Power by the hour. Retrieved from Rolls Royce: https://www.rolls-

royce.com/media/our-stories/discover/2017/totalcare.aspx

Romagnoli, A. (2008). Cambiamento tecnologico e mutamenti sociali nell’evoluzione

del capitalismo.

Schallehn, M., Schorling, C., Bowen, P., & Straehle, O. (2019). Beyond Proofs of Concept:

Scaling the Industrial IoT, Bain & Company.

Selcuk, S. (2015). Predictive maintenance, its implementation and latest trends. Journal

of Engineering Manifacture, 1670-1679.

Semeraro. (2012). TM1 Introduzione.

 190 Bibliography

Seuring, S., & Gold, S. (2012). Conducting content‐analysis based literature reviews in

supply chain management. In R. W. Wilding, Supply Chain Management, Vol.

17 No. 5 (pp. 544-555). Emerald Group Publishing Limited.

Sisinni, E., Saifullah, A., Han, S., Jennehag, U., & Gidlund, M. (2018). Industrial

Internet of Things: Challenges, Opportunities, and Directions. IEEE

Transactions on Industrial Informatics, Vol. 14, n° 11, 4724-4734.

Strauß, P., Schmitz, M., Wöstmann, R., & Deuse, J. (2018). Enabling of Predictive

Maintenance in the Brownfield through Low-Cost Sensors, an IIoT-

Architecture and Machine Learning. IEEE.

Thames, L., & Schaefer, D. (2016). Procedia CIRP. Software-Defined Cloud Manufacturing

for Industry 4.0. Changeable, Agile, Reconfigurable & Virtual Production, 12-17.

The World Bank. (2017). Internet of Things, The New Government Business Platform: a

Review of Opportunities, Practices, and Challenges. Washington, DC 20433.

The World Bank. (2018). GDP (current US$). Retrieved April 2020, from The World

Bank: http://wdi.worldbank.org/table/4.2

T-Mobile. (n.d.). What is 5G? Retrieved from T-Mobile: https://www.t-mobile.com/5g

Trump, D. J. (2018, October). President proclamation - National Manufacturing day 2018.

Retrieved from White House: https://www.whitehouse.gov/presidential-

actions/presidential-proclamation-national-manufacturing-day-2018/

Tumino, A. (2020). Internet of Things: l'innovazione parte da qui., (p. Industrial IoT:

come guidare l'innovazione?).

U.S Department of Energy. (2010). Operations & Maintenance Best Practices: A Guide to

Achieving Operational Efficiency.

United Nations. (2008). International Standard Industrial Classification of All Economic

Activities (ISIC), Rev.4. New York.

van Rijn, C. (2007). Maintenance modelling and applications; lessons learned.

 Bibliography 191

Van Zanden, J. L. (2009). The Long Road to the Industrial Revolution: the European Economy

in a Global Perspective, 1000-1800. The Netherlands: Koninklijke Brill.

Webster, J., & Watson, R. T. (2002). Analyzing the Past to Prepare for the Future:

Writing a Literature Review. MIS Quarterly, Vol. 26, No. 2, xiii-xxii.

Weyrich, M., & Ebert, C. (2016, January). Reference Architectures for the Internet of

Things. IEEE Software, Vol. 33, pp. 112-116.

World Bank. (2018). Indicator % of GDP. Retrieved from Data World Bank:

https://data.worldbank.org/indicator/NV.IND.MANF.ZS?locations=IT&nam

e_desc=false

World Economic Forum. (2015, January). Industrial Internet of Things: Unleashing the

Potential of Connected Products and Services.

World Wide Web Foundation. (n.d.). History of the Web. Retrieved from World Wide

Web Foundation: https://webfoundation.org/about/vision/history-of-the-

web/

Wu. D., Rosen D. W., Wang L., Schaefer D. (2014). Cloud-Based Manufacturing: Old

Wine in New Bottles? Elsevier, 94-99.

Xu, L. D., He, W., & Li, S. (2014, November 4). Internet of Things in Industries: A

Survey. IEEE Transactions on Industrial Informatics, pp. 2233-2234.

Xu, X. (2012). From cloud computing to cloud manufacturing. Elsevier, 75-86.

Yu, W., Dillon, T., Mostafa, F., Rahayu, W., & Liu, Y. (2020). A Global Manufacturing

Big Data Ecosystem for Fault Detection in Predictive Maintenance. IEEE

Transactions on Industrial Informatics, vol. 16, no. 1, 183-192.

Zio, E., & Compare, M. (2012, April). A snapshot on maintenance modeling and

applications. Retrieved from ResearchGate:

https://www.researchgate.net/publication/281923126_A_snapshot_on_maint

enance_modeling_and_applications
 192 Bibliography

Zio, E., & Compare, M. (2012). Evaluating maintenance policies by quantitative

modeling and analysis. Elsevier.