138

Chapter 8
A Review of Artificial
Intelligence Technologies to
Achieve Machining Objectives
Deivanathan R.
VIT Chennai, India

ABSTRACT
Bridging the design, planning and manufacturing departments of a production enterprise is not a con-
clusive effort for the implementation of computer integrated manufacturing. Continuous interaction and
seamless exchange of information among these functions is needed and requires the maintenance of a
large database and user-friendly search and optimization techniques. Among several artificial intelligence
techniques capable of the above task, four important and popular ones are, expert systems, artificial
neural networks, fuzzy logic and genetic algorithms. In this chapter, these four techniques have been
conceptually studied in detail and exemplified by reviewing an application in the manufacturing domain.
Successful implementations of artificial intelligence that are recently reported in machining domain are
also reviewed, suggesting potential applications in the future.

INTRODUCTION

Decision support systems are in use in the technical domain for quite a long time. Their value would be
greatly improved if they could perform a decision-making step even in the absence of a human expert,
given the fact that the human thought processes can be modeled by computers. By tracking human
expertise and learning their preference strategies in a competitive or tie-breaking situation, decision
making could be imitated. But this exercise need not be scientific, rather it is humanistic. “The scientific
approach analyzes a phenomenon in quantitative terms, and has a tendency to analyze the behavior of
humanistic systems as if they were mechanistic systems…” (Zadeh, 1973). Therefore, a modern problem-
solving paradigm, known as Artificial Intelligence (AI) is being practiced when working out heuristic
and experiential solutions to industrial problems.

DOI: 10.4018/978-1-5225-7522-1.ch008

Copyright © 2019, IGI Global. Copying or distributing in print or electronic forms without written permission of IGI Global is prohibited.

A Review of Artificial Intelligence Technologies to Achieve Machining Objectives

The scope of this article is to assess the basic nature of problems faced in the production engineer-
ing domain which are sought to be solved by an intelligent manufacturing approach. This is followed
by a discussion on the various AI techniques in vogue to implement intelligent manufacturing systems.
Further, successful applications of AI recently reported in machining domain are reviewed.

BACKGROUND

The global research firm, Forrester says that 58% of business and technology professionals are researching
AI systems and 12% are actively using them (“Power of AI,” 2018). Accenture research on the impact
of AI in twelve developed economies extrapolated a significant boost in labour productivity of up to
38% in 2035 (Purdey, 2016). Using AI, manufacturers could make informed decisions at each stage in
the production process. There are several methods by which the human problem-solving abilities are
imitated; of which Expert System, Neural Network, Fuzzy Logic and Genetic Algorithm are popular.
Figure 1 displays the total number of published works, noted annually, featuring different AI techniques,
as observed from Microsoft Academic Search website. It is seen that over the last two decades, neural
network and genetic algorithm are used by researchers in large numbers, compared to knowledge-based
system and fuzzy logic technique.
Technologies based on AI are helping to ease and uplift the living standards of human life, in many
cases even we not being aware of it. Nowadays in the medical field, devices are produced with embed-
ded software incorporating expert systems or neural networks for the automatic interpretation of clinical
electrocardiograms (Celler, 1997). Banks and insurance firms could detect fraud and misappropriation
through image processing and neural networks that can analyse data pattern and its day to day devia-
tions. Genetic algorithms serve logistics planning functions in airports and factories, where they are
used to help solve the resource-allocation problems. Fuzzy logic schemes are inbuilt into television sets
to automatically adjust screen parameters based on ambient lighting and the time of the day.

Figure 1. Graphical presentation of total number of publications (annual) worldwide falling under the
keywords ‘knowledge-based system’, ‘artificial neural network’, ‘fuzzy logic’, and ‘genetic algorithm’
Source: Microsoft Academic Search website

139
 
A Review of Artificial Intelligence Technologies to Achieve Machining Objectives

THE MANUFACTURING DOMAIN

Manufacturing is defined broadly as the process by which, material, labour, energy and equipment are
brought together to produce a product having a greater value than the sum of materials put in. In fact,
manufacturing aims at production of goods from raw materials and also it aims to improve productivity
and quality of production. Several approaches have been taken in the industry to realize these objec-
tives and the implementation of intelligent manufacturing system is an epitome of the industrial efforts.
Intelligent Manufacturing Systems are said to have the ability to self-regulate a manufacturing process
within design specifications.

Knowledge Based Decision Support

Business persons may seek expert advice to make a best choice from alternative options, for example, a
product search, i.e., the search for a best product in a market, which is flooded with competing options.
A systematic approach would be to converge from vast pool, by need based pruning of alternatives while
matching user needs to available features of a product. The desirable process or product characteristics
should be listed by the customer to match with the product features of available options. This situation
can be automated by a program to first seek/collect the list of features desirable of the product and then
searching the database for a product, whose feature best meets the demand. Products may then be ranked
based on their suitability for intended usage. Product search is therefore a classical area of research for
intelligent decision support.
Another popular domain where intelligent systems are in demand is the production environment
involving machines, whose downtime due to repair could defer their production schedule. Hence, they
are unproductive until the domain expert corrects the irregularities that caused the breakdown. Fault
identification and correction at the shopfloor, in the absence of expert, is enticing. The remedy to a fault
could be as many as their causative factors. Intelligent systems mimicking human experts are known
to be helpful in drawing conclusions from a ‘situation: action’ or ‘stimuli: response’ paradigm. Expert
knowledge about a ‘system’ corresponding to various ‘subsystems’ and levels of functioning can be
captured and stored in a retractable database. Then, a fault observed by the operator can be traced back
to its root cause using the interactive database and finally one can arrive at a remedy. The input could
be abstract, but the advisory system would guide the respondent (customer) through, to search a stable
and safe condition. So, developing the Human-Machine Interface is an essential component of intelligent
system. Apart from this, safety monitoring is a special case of fault identification where intelligent deci-
sion making is needed when hazardous condition of a machine is detected through sensors.

Classification and Modeling

Engineering research is quite often dominated by data analysis involving classification or clustering of
data to extract information about the process represented by the data. The information extracted could help
in decision making about the state of the process. On a simple instance, data within known boundaries
may be categorized; a data point under consideration may be above / below, larger or smaller, within or
beyond a dividing criterion. Also, there can be more than one factor (criterion) influencing the state of
a process and so the values of all such factors are to be considered into a formula, in attempting ascribe
a data point to one group or other. Hence the criterion for dividing data into groups could be a function

140

A Review of Artificial Intelligence Technologies to Achieve Machining Objectives

of several factors; and may need to be repeatedly computed for the data to be classified. Quite often, the
parametric relationship is not explicit and not easily described by formulae. Learning and representing
the obscure relation among a set of data is major task for intelligent machines.
An extension of this concept, of learning the relation between variables in a set of data and relating
it to a state of a process, would supplement the efforts to develop mathematical models of a process.
The usefulness of a process model (mathematical or AI) are manifold, such as, describing relationship
between input and output, predicting the output for any change in input or other process variables. Fur-
ther, as the process model relates the control variable with the dependent variable, it helps to simulate
alternate trial runs of the process by adjusting the control variables in order to set an optimum output.

INTELLIGENT SYSTEMS

Human intelligence is known to be learnt, assimilated and practiced over years. The fundamental hy-
pothesis of Artificial Intelligence (AI) is that intelligent behavior is constituted by symbolic inference
and computers are capable of doing symbolic processing (Engelmore, 1993). Learning and assimila-
tion of knowledge by computers requires the representation of human knowledge and expertise in the
form of symbolic patterns. Therefore, knowledge representation and symbolic inference are the central
research issues for modelling intelligent systems. Knowledge can be factual, i.e., worldly truth or it can
be heuristic, i.e., experiential; both need to be stored in memory for knowledge processing by a computer
(Honavar, 1995).

Knowledge Based Systems

Knowledge Based systems, also known as Expert Systems, are computer programs that can partially
perform the role of an expert in his absence. An expert system has a user interface, a knowledge stor-
age subsystem, and an inferencing or reasoning subsystem to work over a problem at hand, as shown in
Figure 2. These components are comparable to user interface, database and algorithm in the sense of
conventional programming.
The representation of factual and heuristic knowledge of a domain in the form of structured database
(knowledge base) is an important stage in developing expert systems. One common method of knowl-
edge representation is the ‘production system’ – a rule-based system in the form of IF-THEN rules (IF
‘condition’ THEN ‘action’). Here, the IF part has a list of conditions, proclaiming a combination of
information available initially, to deduce a preliminary solution. Given a set of rules covering the entire
problem domain, a piece of knowledge in a rule is relevant during a user interaction, if the line of reason-
ing being developed (after user interaction) matches with ‘IF’ part of the rule. The ‘THEN’ part yields
the deduced information, thus taking the user closer to the solution or conclusion.
In cases where knowledge regarding the problem domain is hierarchical in nature, a pure rule based
system can result in an unnecessarily large number of rules and it is suggested to use an alternative knowl-
edge representation method along with rules (Sudhir,1992). Another form of knowledge representation,
invented by Minsky (1975), is the ‘frame’, also termed as the ‘unit’ or ‘list’ or ‘scheme’. A frame is an
assemblage of the properties of an entity being represented, synonymous to ‘facts’ in a rule-based sys-
tem. Each property defines a characteristic of an entity and may have a value assigned to it. Sometimes,

141
 
A Review of Artificial Intelligence Technologies to Achieve Machining Objectives

relations between entities may need to be represented. In a frame structure, the properties contain the
relations possessed by an entity; its value refers to other entity related by this property.
The ‘knowledge engineer’ is a person who compiles the real-world knowledge of a problem domain
constituting the knowledge base. He also develops a program module that can search through the knowl-
edge base by chaining the ‘IF_THEN’ rules, manipulating the symbolic phrases and thereby yielding
a logical conclusion. The search program can either start with a set of known conditions and narrow
down the search for a conclusion (forward chaining) or it can refer to a fixed conclusion and backtrack
the knowledge base to suggest possible conditions that are causative (backward chaining). LISP and
PROLOG are the earliest programming tools (languages) available for the ‘knowledge engineer’ to build
this inferencing mechanism and hence the Expert System. ‘Backtracking’, ‘recursion’ and ‘inheritance’
are the special capabilities of such programming languages. Working with uncertain knowledge and
explanation of the line of reasoning for a conclusion are other commendable features.

Neural Networks

An artificial neural network (ANN) is a network of neurons interconnected by weight vectors (synapse),
analogical to biological nervous system (brain cell) of neurons and the synapse. ANN topology consists of
an input layer of neurons used to present data to the network, Anandakumar and Umamaheswari (2018)
has an output layer of neurons to produce ANN’s response, and one or more hidden layer of neurons in
between them, as shown in Figure 3(a). The number of neurons in the input and output layer is decided
by the number of input and output variables respectively, with reference to the problem that is modelled.
In addition to their topology, ANNs are characterized by weight vectors (synapse), and activation
function used in the neurons of the network, as shown in Figure 3(b). The weight of a synapse, multiplied
by the strength of the data on that synapse, defines the contribution of that synapse to the activation of
a neuron for which it is an input. The total activation of a neuron is the sum of the activations of all its
inputs. That is, activation = Ʃwixi, for i = 1 to n, where, ‘i’ is the number of synaptic inputs to a neuron.
This activation determines the value of the output from that neuron, according to its transfer function.
Transfer functions are S-shaped curve with the output confined between limits of 0 to 1 at - ∞ to ∞.
Depending on the type of transfer function the limits may be -1 to 1, or 1/2, to -1/2 (Mishra, 2010). The

Figure 2. Structure of rule based expert system

142

A Review of Artificial Intelligence Technologies to Achieve Machining Objectives

Figure 3. (a) Neural network topology (b) Inputs and outputs at a neuron

S-curve or sigmoid curve is a special case of logistics function defined by S(t) = 1/(1+e-t). Hyperbolic
functions are also used to define the sigmoid, given by S(t) = tanh(t) = (1-e-2t)/(1+e-2t) = (et-e-t)/(et+e-t).
Neural networks can be designed to function in a feed forward mode or as a feedforward with back
propagation (FFBP) of errors. A multilayer perceptron (MLP) is a feedforward neural network with one
or more hidden layers. A feedforward neural network has sequence of layers consisting of a number of
neurons in each layer. The output of one layer becomes input to neurons in the succeeding layer (Chan-
drasekaran, 2010).
ANN learns the relation between a set of input and output variables corresponding to an experimental
data presented. A neural network is trained with a sample data of input-output relations and tested with
another set of data to assess the effectiveness of learning and arrive at an optimum topology and weights.
Actually, the topology and weights are the ANN parameters by which the given input is mapped on to
a given output, in the training phase. Once trained, the neural networks can be used for prediction of
output for a set of input variables. During the training process, the network adjusts its weights to mini-
mize the errors between the output of the network and actual output (Anandakumar, Umamaheswari &
Arulmurugan, 2018).
The Back Propagation algorithm is based on a steepest-descent approach for adjusting the weight
values to minimize the error, in an iterative fashion. An iteration of the training process consists of two
passes of computation: a forward pass and a backward pass through different layers of the network. To
start with, the synaptic weights are initialized at random between 0 and 1. In the forward pass, the input
data are applied to the neurons of the input layer. The activation function of each neuron is computed
and its weighted value (according to the synaptic weight) propagates through the network, layer by layer.
During the forward pass, synaptic weights are all fixed. The error which is the difference between the
output value from the network and the corresponding true value is propagated back during the backward
pass, to adapt the synaptic weights. The neural network is trained in an iterative manner until a stopping
criterion is reached in the form of allowable training error. After each iteration the synaptic weights are
updated in accordance with the learning rate constant (μ) and the momentum coefficient (η), initially
defined in the neural network (Hou, 1995).

Fuzzy Logic

When mathematical information available about a problem is highly complex or incomplete, the fuzzy
logic approach is useful instead of the classical approach to formalize information and systematically

143
 
A Review of Artificial Intelligence Technologies to Achieve Machining Objectives

trace the cause-effect relation. It is used when subjective knowledge and suggestion by the expert are
significant in defining objective function and decision variables. It differs from conventional computing
in that, it is tolerant of imprecision, uncertainty, partial truth and approximation.
Lofti Zadeh (1965) put forward the idea of fuzzy set theory, in which the elements i.e., classes of
objects can have partial membership and so the transition from membership to non-membership is
gradual rather than abrupt. Indeed, the logic behind human reasoning is (based on) fuzzy truths, fuzzy
connectives and fuzzy rules of inference. “The fuzzy logic approach has the ability to summarize the
information and extract from the collection of data impinging on the human brain, only those subcol-
lections which are relevant to the performance of the task at hand” (Zadeh, 1973).
Problem resolution by fuzzy logic involves three important stages, viz., fuzzification of inputs, fuzzy
inferencing with fuzzy rules and defuzzification of processed information into outputs, as shown in figure
4. A fuzzy set-based prediction system takes the crisp input data and carries out “fuzzification”. In the
fuzzification process, the real-world input data (input parameters like, feed, speed, depth of cut, cutting
force, etc.) undergo some translation in the form of linguistic terms such as “small feed”, “medium cut-
ting speed”, “large depth of cut”, “very large cutting force”, etc.
To affect this, the fuzzy set concept is used. The linguistic terms thus obtained are converted into
a fuzzy set with the help of subject knowledge of domain experts. For example, the “low feed” can be
represented by a fuzzy set in which the feed values more than an upper threshold value can be assigned
a membership grade 1 and those lower than a lower threshold value can be assigned a membership grade

Figure 4. Three important stages in fuzzy logic approach

Figure 5. Fuzzification and Defuzzification process in a fuzzy logic solution

144

A Review of Artificial Intelligence Technologies to Achieve Machining Objectives

0. Between lower and upper threshold, the feed values can have a gradual variation of membership grades
from 0 to 1. Thus, each input parameter is associated with a linguistic variable and a membership grade,
forming the fuzzified input.
The fuzzy inference system applies the fuzzy rules to the set of fuzzy input to produce an output.
The fuzzy rules is a set of predefined ‘IF: Condition-THEN: Action’ rules and a combination of which
leads to a solution, in the context of the problem. In the fuzzy inferencing process, the membership
value of the consequent of a rule is made equal to the minimum membership value of the antecedents.
The defuzzification process uses the weighted value of the memberships of the consequents of all the
relevant rules. The fuzzy output is defuzzified into crisp numeric data considering the output linguistic
variables and the membership grades.
Membership to a fuzzy set can be decided by a triangular or trapezoidal membership function, as
shown in figure, and conversion from fuzzy to crisp values could be established by well-known tech-
niques like the centroid theorem. Once the linguistic variables have been converted into fuzzy sets, set
theoretic operations on them can be carried out (Zadeh, 1973).

Genetic Algorithm

A Genetic Algorithm (GA) is a paradigm that tries to mimic the genetic evolution of a species over
generations in a population. GA can be construed as a population-based search methodology, where a
population is formed of innumerable feasible solutions, which are filtered by a fitness test in order to
yield a new population of feasible solutions, called the ‘next generation’. The optimality of a solution
is tested over generations by introducing ‘modified feasible solutions’ into each generation. Thus, a
feasible solution adapts itself to a fitness function. The adaptation process resembles the natural process
of genetic inheritance from parents to children and survival of the fittest.
In GA, a feasible solution is represented by binary or decimal numbers, known as a string or ‘chro-
mosome’ and the total population of feasible solutions forms the ‘search space’. The string of numbers,
forming the genes of a chromosome, encodes the various input/output variables corresponding to an
optimization problem. The feasible solution is evaluated against an objective or fitness function. Each
chromosome is then assigned a fitness value that indicates how closely it satisfies the desired objective
and its chances of survival into the next generation. Whenever a feasible solution violates the constraints
of the problem, a penalty is assigned to its fitness value.
The evolution of a feasible solution is ensured by three fundamental operations, viz., reproduction,
crossover and mutation of the set of chromosomes forming the population. Reproduction operation
replaces the population with large number of good strings having high fitness values, crossover opera-
tion produces new chromosomes by combining the various pairs of chromosomes in the population,
and mutation operation produces slight random modification of chromosomes. Crossover and mutation
are not applied to the entire population, but are taken as probabilistic events. A sequence of these three
operations constitute one generation. The process repeats till the system converges to a solution, decided
by a predefined number of iterations (generations). The finally obtained optimum solution combines the
best elements of previous generation of chromosomes (Chandrasekaran, 2010; Man, 1996). It is to be
noted that while the crossover enables the algorithm to extract the best genes from different individuals
and recombine them into potentially superior children, the mutation adds to the diversity of a popula-
tion and thereby increases the likelihood that the algorithm will generate individuals with better fitness
values. Both processes are essential to the GA.

145
 
A Review of Artificial Intelligence Technologies to Achieve Machining Objectives

To further illustrate the operational procedure, a one-point crossover mechanism is depicted on Figure
6. A crossover point can be randomly fixed, as shown by a dividing line across the parent strings. The
portion of the two chromosomes beyond this cut-off point to the right is to be exchanged to form the
offspring. An operation rate with a typical value of between 0.6-1.0 is normally used as the probability
of crossover. However, in mutation, the operation is applied to each offspring individually after the cross-
over exercise and it alters each bit randomly with a small probability of less than 0.1 (Namdari, 2014).

Manufacturing Applications of Artificial Intelligence

Manufacturing automation has evolved over ages. Initially hard automation was implemented followed
by programmed automation and flexible automation. Based on our assessment of the manufacturing
problems, we find that they are not purely logical or methodical to automate by a writing simple code
and executing. At several stages in the implementation of computer aided manufacturing, the skill and
knowledge of an expert are quintessential. Therefore, a new research topic named as ‘Soft Automation’
has arisen, that is utilizing computational resources and exhibiting intelligent behavior to fulfill the goals
of manufacturing automation.

Application of Knowledge Based System

The designing, process planning and manufacturing functions have been traditionally managed in sepa-
rate departments due to their colossal and unique functions. After computerization, several proponents
have attempted to integrate these functions. This section describes briefly, a case study by Prabhakar et
al. (2004) on the development of Knowledge Based System for integration of Computer Aided Design
and Manufacturing (CAD & CAM) for machining process using feature technology.
Process planning is the function that bridges CAD and CAM to implement computer integrated
manufacturing. It involves determination of appropriate machines, tools for machining the parts, cutting
fluid and machining parameters under cutting conditions for each operation on a given part. There are
two different methods of preparing the process plan. A variant system uses ‘group technology’ technique
to create process plans from a set of standard plans prepared earlier. A generative system creates process
plan automatically, based on logical procedures similar to that of a human planner.
While the user creates a feature-based model, the features and their attributes are stored in a database
in the same order. Using this data, process plan for its manufacture can be generated based on the se-
quence in which features on the model were created. The base features of the component (for example,
circular or square section) are used to select the workpiece. The inner features are used to identify the

Figure 6. Crossover and mutation phenomena in GA

146

A Review of Artificial Intelligence Technologies to Achieve Machining Objectives

operations, machines and cutting tools and the sequence of operations. The parameters required in the
process plan can also be derived from the design features or from the knowledge base. The technical
knowledge of process planners are captured and coded into a computer program. In this work around
fifty process rules (of the IF-THEN-ELSE type) were used for converting the design features of a CAD
model into manufacturing features.
Strength of the work: The authors proposed a hybrid of variant and generative approach for process
planning. Feature recognition is the strength of this work. Features are considered as information carriers
between CAD, CAPP and CAM. A feature library is maintained consisting of templates of all elements
of a part drawing with geometric and other parameters. The software was designed for generating process
plans by interacting with CATIA (part modeller) and a virtual factory environment offering the database
of available machines, cutting tools, workpieces, time and cost data. Their database was created and
updated using query-based user inputs.

Application of Artificial Neural Network

This section describes an intelligent system developed using ANN, to monitor and estimate the tool wear
in face milling, using Acoustic Emission (AE) and cutting force signals (Raimond, 2008; Kumudha,
1998). The ANN was trained using Back Propagation Algorithm (BPA) as it has the capability to solve
nonlinear problems, given the fact that tool wear is a nonlinear phenomenon. In order to relate the sen-
sor values to the actual wear, axial flank wear was measured off-line at regular intervals of machining.
Selecting the representative patterns of data to model the process is an important step. An AE moni-
toring system was used to sense and analyze the acoustic emission resulting from the machining process
and was translated into the AE parameters like, Ring down Count, Rise Time, Event, Event Duration, and
Energy. Machining force was measured using a three component (Fx, Fy, Fz) piezoelectric dynamometer
in-process. Data acquired through sensors during machining (5 AE parameters and 3 force components)
are normally highly variant due to the stochastic nature of the tool wear process. So, for proper training

Figure 7. (a) CAD model of a sample mechanical component (b) Representative design features and
manufacturing features of CAD model (Prabhakar, 2004)

147
 
A Review of Artificial Intelligence Technologies to Achieve Machining Objectives

of MLP and to achieve the desired performance, it is necessary to input the normalized data with higher
order variance from among those acquired. Usually, 50- 70% of the acquired data is used for training
the ANN. The internal process of updating the weights in the network is mostly dependent on η and μ
parameters and hence they were optimized to 0.1 and 0.7 respectively. In the present work, the number of
input nodes equals eight, with single output node (to estimate the flank wear). The network was trained
until its output converged to the point defined by the stopping criterion of a small error threshold of
value 6.6 × 10-5 along with minimum validation error.
Strength of the work: It is difficult to obtain analytical models based on cutting parameters to esti-
mate the tool wear since complex phenomena occurs during the machining process. Typical flank wear
characteristics of coated carbide when machining the steel workpiece indicates that the coated carbide
experienced rapid tool wear initially, followed by a slow propagating wear region and then there is a
further rise in the propagation rate. But, TiN coated carbide inserts undergo certain dynamic changes
such as deformation of coating material, formation of ledge over the rake face and continuously chang-
ing wedge geometry, which influence the nature of machining. Hence a simple mathematical model of
the wear curve falls short of accurate predictions under dynamic environment.
Sensor based data can be fused along with cutting parameters using ANN for the purpose of iden-
tification of tool status (Madhava Reddy, 2012; Nathan, 2001). But, a single sensor information often
remains unreliable and tool condition monitoring may fail to achieve the objectives. So, when more
number of input features are to be considered for modelling the machining process, ANN is the most
suitable technique. Also, we can include features measured in-process such as force and AE signals.

Application of Fuzzy Logic

This section describes certain aspects of a fuzzy logic-based model to predict the roughness after milling
a rectangular piece of soda lime glass plate (Sarhan, 2012). Soda-lime glass has been used as a mould
material with good dimensional accuracy even at elevated temperatures. The machining performance
characteristic of glass material is of great concern for silicone mould manufacturers in order to produce
good quality.
In glass milling operation, the lubricant pressure, spindle speed, feed rate and depth of cut are the
significant parameters affecting surface roughness. The conventional method to achieve low surface
roughness and cutting forces at different machining parameters is the “trial and error” approach. A reli-
able systematic approach to predict the surface roughness after milling is to capture the input-output
relation of a limited set of experimental data, followed by a fuzzy logic approach for learning of the said
relationship and subsequent tests. Based on L13 experimental array of operating parameters (lubrication
pressure, spindle speed, feed rate and depth of cut) experimental data on the glass milling process was
collected in a random sequence.
First, the input and output parameters in the numeric form were mapped to the linguistic variables.
While generating the fuzzy rules, the authors used the term ‘A’ for lubrication pressure, ‘B’ for cutting
speed, ‘C’ for feed rate and ‘D’ for depth of cut. For each input variable, four membership functions
were used which are Low, Medium, High, and Very High. The entire range of input variables would fall
into these four sets. Gauss shape of membership function was employed to map the fuzzy sets for the
four input variables. The output variable (roughness) also used four membership functions; Best, Good,
Average and Bad. In output fuzzy set, and triangular shape of membership function was used with its
single peak and linear variation. A set of 13 rules were constructed (based on the experimental array

148

A Review of Artificial Intelligence Technologies to Achieve Machining Objectives

of parameters and the surface roughness of a machined glass surface), to support the fuzzy inferencing
process of roughness prediction. Typical fuzzy rules are listed here.

1. IF (A is L) and (B is L) and (C is L) and (D is L) then (Roughness is Best)

Fuzzy logic modelling was carried out in the Matlab software on the basis of Mamdani Fuzzy Logic
concept. The centroid of area defuzzification method was used. In this method, the resultant member-
ship functions are developed by considering the union of the output of each rule. Results from the fuzzy
logic model indicate that, for better surface roughness, the recommended settings are the lowest values
of the depth of cut and feed rate and the highest values of the lubrication pressure and spindle speed.
Strength of the work: The fuzzy logic model could predict roughness values within 8.75% error. The
Gaussian membership function at the input side, well differentiates a nonlinear range of different input
data into fuzzy sets. Even a novice user gets benefited as he can work on the model using linguistic terms
for control. Moreover, the errors of numeric computation and optimization are avoided. However, profit-
able fuzzy applications are frequently noted in process control and optimization applications. Further,
Mamdani systems are intuitive and well suited to human input compared to Sugeno systems which are
well suited to mathematical analysis.

Application of Genetic Algorithm

A typical application of GA to machining parameter optimization is next described. The machining eco-
nomics problem concerns the determination of the right process parameters, usually cutting speed, feed
rate and depth of cut, in order to achieve the machining objectives. To find optimal cutting parameters
during a turning process, D’addona and Teti (2013) used the genetic algorithm as an optimal solution
finder.
The main objective is to determine the optimal machining parameters that minimize the production
time without violating any imposed cutting parameter and quality constraints.
Usually, the production time, Tp, is the time necessary for the fabrication of a product and is estimated as,

Tp = Ts + V(1 + Tc/T)/MRR + Ti

where, Ts, Tc, Ti, V and MRR are the tool set-up time, the tool change time, the tool idle time, the
volume of the removed material and the material removal rate. In some operations, the Ts, Tc, Ti and V
are constants so that Tp is the function of MRR and the tool life, T, given by Taylor’s formula.
The most important criterion for the assessment of the surface quality constraint is roughness, Ra, cal-
culated empirically from the machining parameters using exponents relevant to a specific tool-workpiece
combination. Another constraint related to machine is the maximum permissible value for cutting power.
Due to the limitations on the machine and cutting tool and due to the safety of machining, the cutting
parameters have a limited range. So, the cutting force, must be limited to a maximum value determined
by the strength and stability of the machine as well as the cutting tool.
It was desired to arrive at the optimal cutting parameters satisfying an objective function of minimum
production time, subject to constraints of roughness and power. This was performed using Matlab’s GA

149
 
A Review of Artificial Intelligence Technologies to Achieve Machining Objectives

toolbox. Priorly, machining data was collected from NC lathe while machining a cast steel blank by an
HSS tool. The basic GA program was setup with a population size = 20 -100 individuals; number of
generation = 10; probability for crossover = 0.8 and mutation rate = 0.1.
First, an initial population of 100 individuals, each comprising of cutting conditions (the combination
of speed, feed, depth of cut), were generated at random within the specified limits and encoded into a
string of numbers. Thus, each string (i.e) chromosome represents a set of cutting parameters, which is a
likely solution to the optimization problem. For each one of the initial population of chromosomes, the
constraints of power, force and roughness were calculated using formulae and checked for non-violation
of appropriate limits. Thus, a set of feasible solutions was obtained. The best among the feasible solu-
tions, in terms of fulfilling the objective function, will be the optimal solution.
The individuals that survive this selection step, undergo alteration by two genetic operators, crossover
and mutation, to produce the next generation of chromosomes. The crossover and mutation operations
were performed with a probability factor of 0.8, and 0.1 respectively. Subsequently, the new chromosomes
will be evaluated by the objective function. The cycle of activities were carried out up to ten generations.
Production time was found to be minimum in the 6th generation.
Strength/Justification of the work: GA is an optimization algorithm that does not require a strong
mathematical background, rather it is a bio inspired method. Compared to traditional optimization para-
digms, a GA is robust and may be applied exhaustively without recourse to domain-specific heuristics.
So, GAs are widely used for machine learning, function optimizing and system modeling. Since the
genetic algorithm-based approach can obtain a near optimal solution, it is preferable for machining pa-
rameter selection for parts with complicated detailing and bound by several machining constraints. GA
is capable of multi objective optimization and its integration into an intelligent manufacturing system
will lead to unimaginable improvement in quality and economic production.

RECENT RESEARCH IN AI APPLICATIONS

Knowledge Based Systems

An integrated manufacturing system needs automated conversion of design information into its manu-
facturing information. The information required at the manufacturing stage such as, the dimensional
tolerance specifications, operation sequence, part fixturing and orientation data, cutting tool and mate-
rial data, etc., are preferably extracted from CAD database in a CIM environment. Explicit programs
are developed to perform this task by a feature-based approach considering manufacturing features as
the common link between various design and manufacturing functions (Manafi, 2017; Elaakil, 2017).
Moreover, a multi agent system can be used for the capture of domain specific knowledge (Djapic, 2017).
The collective manufacturing information was also used by Ji et al. (2018), to assess both machinability
and machining cost.
Further, KBS has been extended to perform economic analysis of production process through ma-
chining parameter selection and operations scheduling (Agrawal, 2017). Deng, et al. (2018) constructed
an expert system for green cutting process using a knowledge base of part models and process models
of typical components.
Successful implementation of process monitoring and control has been reported through hybrid AI
techniques involving expert systems like the fuzzy expert system (Cica, 2017). Bio-inspired algorithms,

150

A Review of Artificial Intelligence Technologies to Achieve Machining Objectives

such as genetic algorithm (GA) and particle swarm optimization (PSO) are used for tool wear predic-
tion in hard turning of bearing steel. Zhu et al. (2018) proposed a smart monitoring system for CNC
machining based on Cyber Physical Production System framework, which involves the physical and
virtual modelling of the machine tool, process monitoring and big data analytics.

Artificial Neural Network

ANN has found widespread use for prediction of roughness, cutting force, tool wear and also for opti-
mization of machining parameters in various machining processes like turning, milling, drilling. ANN
models have been developed for predicting surface roughness due to milling of various novel materi-
als, using the machining parameters as the input factors in addition to humidity, vibration, white noise,
etc. (Bozdemir, 2018; Khorasani, 2017; Alharthi, 2017; Kilickap, 2017; Cirak, 2017; Ighravwe, 2015).
Similarly, tool wear in milling has been predicted through ANN models by several authors (Kilickap,
2017; Ighravwe, 2015; Pohokar, 2014), and cutting force was predicted by Kilickap, et al. (2017) and
Malghan et al. (2018). ANN process models have also been reported for roughness prediction in drilling
by Roy and Bhagat (2015), roughness prediction in turning by Fang, et al. (2016) and for cutting force
prediction in turning by Abbas, et al. (2018) and Makfi, et al. (2018).
By a better design of the ANN model, it has been shown that the machined surface roughness char-
acteristics such as Ra, Rz, Rt, Rq could also be predicted. For this purpose, a FFBP neural network
with 20 hidden layers was used by Al Hazza et al. (2013) while Fang, et al. (2016) applied both MLP
and Radial Basis Function. Lipinski, et al. (2017) argued that while developing the ANN for process
modeling, the determination of proper network architecture is a significant issue. The presented a neural
network with generalization ability, to predict the surface roughness, tangential force and normal force
in grinding of three different materials (Ti6Al4V, 100Cr6, Inconel) with a selected set of wheels under
various operating parameters. After extensive analysis the 11-8-9-3 ANN structure was found to produce
the lowest modeling error. Saric et al. (2016) used back propagation ANN to investigate the dependency
of plane parallelism error while CNC grinding of multilayered ceramics. They considered the machine,
operator, ceramic foil and the production line as the factors influencing the pane parallelism and tested
diverse network architectures with different learning rules (delta, delta bar delta) and transfer functions
(sigmoid, hyperbolic tangent) for getting the best performing network.
For roughness prediction in CNC milling, Cirak et al. (2017) showed after trials with MLFFBP
neural networks, that a 4-7-1 architecture trained with LM algorithm gives the best performance. Makfi
et al. (2018) proposed an improved design of ANN model with eleven hidden layer neurons, trained
with Bayesian regularisation in combination with LM algorithm. Roy and Bhagat (2015) reported an
improved performance of BP algorithm by allowing the learning rate to change during the training pro-
cess. Ighravwe et al. (2015) used a two hidden ANN layer model for prediction of tool wear and surface
roughness in milling and further optimization using fuzzy goal programming, big-bang big-crunch and
PSO as modelling tools. A novel Edgeworth–Pareto optimization of ANN was presented by Abbas et
al. (2018) and they suggested a 4-12-3 network for surface roughness prediction. Zhang et al. (2016)
developed a neuro fuzzy network for tool condition monitoring in milling using the vibration signals.
Kannan, et al. (2014) applied ANN model for parameter optimization in drilling after evaluating the
machining time, thrust and ovality. ANN technique was applied for machining parameter optimization
in milling by Harinath et al. (2014), Chaskar et al. (2017) and by Pohokar (2014) in turning. Machining
time, tool wear and cost factors were considered for optimization by Abbas et al. (2018).

151
 
A Review of Artificial Intelligence Technologies to Achieve Machining Objectives

Fuzzy Logic

Fuzzy logic model has been developed to aid in the selection of cutting parameters in drilling based
on material hardness, hole dia and federate (Pendokhare, 2012). Mahesh and Rajesh (2014), Al Sahib
(2014) used fuzzy logic for cutting parameter selection in end milling considering the roughness and
material removal rate.
Fuzzy logic-based machining parameter optimization to produce desirable surface roughness char-
acteristics, such as Ra, Rz, etc., has been studied by Das et al. (2014, 2016) when machining copper
alloys and composites.
Hu et al. describe the implementation of real-time machining process control in a new STEP-NC
controller, involving the transfer of information between CAPP and CNC, wherein a fuzzy control
algorithm for cutting force control is designed and embedded in the software kernel of the STEP-NC
controller (Hu, 2016).
Kalaichelvi et al. (2012) used fuzzy classification method to classify the tool wear states so as to
facilitate defective tool replacement at the proper time. Salimi et al. (2015) implemented a fuzzy logic-
model to predict the tool wear in turning operation using cutting parameters and forces as the input.
They suggested tool replacement based on ISO 3685 standard.
Different techniques have been employed to relate the input/output data to fuzzy sets such as the use
of Triangular membership function or Gaussian function (Barzani, 2015; Tanikic, 2017). Fuzzy logic
model was developed for predicting surface roughness while machining different engineering materials
(Rajasekhar, 2014; Naresh, 2014; Barzani, 2015; Babu, 2015).
Iqbal, et al. (2017) puts forward the application of fuzzy modeling for optimizing the machining
process in respect of various combinations of minimizing specific energy consumption and maximizing
productivity as well as tool life. Jogendra et al. (2018) applied grey-fuzzy hybrid analysis to optimize
milling parameters and to improve the quality. Machining parameter optimization of turning process
has been carried out using grey fuzzy logic approach by Senthilkumar et al. (2015) to obtain optimum
MRR, surface roughness and tool wear.

Genetic Algorithm

Tool path optimization in CNC milling has been successfully implemented by Barclay et al. (2015) using
evolutionary algorithm for optimizing the machining time and straightness achieved. GA approach is
also reported for achieving energy efficient work table movements (Raja, 2018). Guiotoko et al. (2018)
carried out GA based optimization of hole tolerance in 5 axis machining, in order to get low machining
time and high accuracy.
Response Surface Methodology (RSM) as a statistical technique, has been collectively used with
GA, to overcome the limitations of RSM and reach the global optimization process conditions. GA was
applied to obtain the proper choice of machining parameters to maximize MRR and minimize cutting
force and roughness, after an initial assessment of the parametric influence by RSM (Rishi, 2014; Vi-
jayakumar, 2014). Similarly, Shaik and Srinivas (2017) proposed a multi-objective optimization approach
based on genetic algorithms to simultaneously minimize the tool vibration amplitudes and work-piece
surface roughness. The optimum combination of process variable thus obtained was further verified by
the radial basis neural network model.

152

A Review of Artificial Intelligence Technologies to Achieve Machining Objectives

A hybrid genetic algorithm combines the exploration of GA with the exact convergence of a deter-
ministic algorithm. The hybrid GA has all of the steps of a regular GA (e.g., selection, mating, mutating,
elitism), however, after the mutation routine, each candidate is optimized locally (Tummuluru, 2016).
Nafeez et al. (2014) presented a hybrid approach to machining parameter optimization using ANN and
GA concepts. They considered cutting force, surface finish, machine power and speed to search for op-
timum machining parameters. Agrawal, et al. (2017) developed a hybrid genetic algorithm and expert
system for machining parameter selection and operations scheduling based on a precedence cost matrix
for features of cylindrical components.

SUMMARY

Problems in manufacturing engineering are knowledge intensive and unavoidably involve decision
making by experts at various levels. Considering the scarcity of expertise and the push for unattended
manufacturing, development and implementation of knowledge-based systems or intelligent systems
for process modeling and optimization is important. Further expert advisory systems are well suited in
planning and maintenance.
Soft computing techniques are preferred to physics-based methods for modeling and optimizing manu-
facturing systems which are characterized by complexity, uncertainty and lack of accuracy and precision.
Intelligent manufacturing can be achieved by utilizing the proven technologies of AI (fuzzy logic,
neural networks, etc.) to monitor and control the manufacturing functions such as production planning,
scheduling and shop floor operations. Four key technologies of AI domain, viz., expert system, neural
network, fuzzy logic and genetic algorithm, have been studied and are seen to be taking semantic and
evolutionary approach to model the machining process. Successful implementation examples stand proof
to the benefits of the intelligent systems.
Knowledge based systems, like those intended for cutting tool selection, tend to be a peripheral unused
system if they do not form part of regular information exchange activities of a manufacturing system.
Therefore, KBS should be preferably integrated with the manufacturing system to increase their usage.
ANN and fuzzy logic techniques have found widespread use for prediction of roughness, cutting force,
tool wear and also for optimization of machining parameters in various machining processes like turn-
ing, milling, drilling. GA has found widespread application in machining process optimization under
multiple constraints. Successful implementation of process monitoring and control has been reported
through hybrid AI techniques like fuzzy expert system, neuro fuzzy system and hybrid genetic algorithm.

REFERENCES

Abbas, A. T., Pimenov, D. Y., Erdakov, I. N., Taha, M. A., Soliman, M. S., & El Rayes, M. M. (2018).
ANN surface roughness optimization of AZ61 magnesium alloy finish turning: Minimum machining
times at prime machining costs. Materials (Basel), 11(5), 808. doi:10.3390/ma11050808 PMID:29772670
Agrawal, A., Rajput, R. S., & Shrivastava, N. (2017). Operation sequencing and machining parameter
selection in CAPP for cylindrical part using hybrid feature based genetic algorithm and expert system.
International Research Journal of Engineering and Technology, 4(6), 858–863.

153
 
A Review of Artificial Intelligence Technologies to Achieve Machining Objectives

Al Hazza, M. H., & Adesta, E. Y. (2013). Investigation of the effect of cutting speed on the surface rough-
ness parameters in CNC end milling using artificial neural network. IOP Conference Series. Materials
Science and Engineering, 53(1), 012089. doi:10.1088/1757-899X/53/1/012089
Al-Sahib, N. K. A., & Abdulrazzaq, H. F. (2014). Selection of optimum cutting speed in end milling
process using fuzzy logic. Innovative Systems Design and Engineering, 5(2), 14–30.
Alharthi, N. H., Bingol, S., Abbas, A. T., Ragab, A. E., El-Danaf, E. A., & Alharbi, H. F. (2017). Optimiz-
ing Cutting Conditions and Prediction of Surface Roughness in Face Milling of AZ61 Using Regression
Analysis and Artificial Neural Network. Advances in Materials Science and Engineering.
Anandakumar, H., & Umamaheswari, K. (2018). Cooperative Spectrum Handovers in Cognitive Radio
Networks. In M. H. Rehmani & R. Dhaou, (Eds.), Cognitive Radio, Mobile Communications and Wire-
less Networks, pp. 47-64. Springer. doi:10.1007/978-3-319-91002-4_3
Anandakumar, H., Umamaheswari, K., & Arulmurugan, R. (2019). A Study on Mobile IPv6 Handover
in Cognitive Radio Networks. In International Conference on Computer Networks and Communication
Technologies (pp. 399-408). Springer Singapore. doi:10.1007/978-981-10-8681-6_36
Babu, K. A., Kumar, G. V., & Venkataramaiah, P. (2015). Prediction of surface roughness in drilling of
Al 7075/10%-SiCp composite under MQL condition using fuzzy logic. Indian Journal of Science and
Technology, 8(12).
Barzani, M. M., Zalnezhad, E., Sarhan, A. A., Farahany, S., & Ramesh, S. (2015). Fuzzy logic based
model for predicting surface roughness of machined Al–Si–Cu–Fe die casting alloy using different
additives-turning. Measurement, 61, 150–161. doi:10.1016/j.measurement.2014.10.003
Bozdemir, M. (2018). Prediction of surface roughness considering cutting parameters and humidity
condition in end milling of polyamide materials. Computational Intelligence and Neuroscience, 1–7.
doi:10.1155/2018/5850432 PMID:30050565
Celler, B., De Chazal, P., & Lovell, N. (1997). A comparison of expert systems for the automated in-
terpretation of the ECG: regulatory implications of the use of neural networks. In APAMI-HIC 1997:
Managing Information for Better Health Outcomes in Australia and the Asia Pacific Region: 11 to 13
August 1997, Asia Pacific Association of Medical Informatics, HISA: Conference Proceedings (p. 492).
Health Informatics Society of Australia.
Chandrasekaran, M., Muralidhar, M., Krishna, C. M., & Dixit, U. S. (2010). Application of soft comput-
ing techniques in machining performance prediction and optimization: A literature review. International
Journal of Advanced Manufacturing Technology, 46(5-8), 445–464. doi:10.100700170-009-2104-x
Chaskar, P. R., Lad, A. S., Mane, J. K., Sangre, A., & Kirkire, M. S. (2017). Optimization of CNC Mill-
ing Parameters Using “Artificial Neural Network.” Imperial Journal of Interdisciplinary Research, 3(2).
Cica, D., Sredanovic, B., Borojevic, S., & Kramar, D. (2017). An Integration of Bio-inspired Algorithms
and Fuzzy Logic for Tool Wear Estimation in Hard Turning. In International Conference on Advanced
Manufacturing Engineering and Technologies (pp. 1-12). Cham: Springer.

154

A Review of Artificial Intelligence Technologies to Achieve Machining Objectives

Cirak, B. (2017). Mathematically Modeling and Optimization by Artificial Neural Network of Surface
Roughness in CNC Milling–A Case Study. World Wide Journal of Multidisciplinary Research and
Development, 3(8), 299–307.
D’addona, D. M., & Teti, R. (2013). Genetic algorithm-based optimization of cutting parameters in
turning processes. Procedia CIRP, 7, 323–328. doi:10.1016/j.procir.2013.05.055
Das, B., Roy, S., Rai, R. N., & Saha, S. C. (2014). Surface Roughness of Al-5Cu Alloy using a Taguchi-
Fuzzy Based Approach. Journal of Engineering Science & Technology Review, 7(2).
Das, B., Roy, S., Rai, R. N., & Saha, S. C. (2016). Application of grey fuzzy logic for the optimization
of CNC milling parameters for Al–4.5% Cu–TiC MMCs with multi-performance characteristics. Engi-
neering Science and Technology, an International Journal, 19(2), 857-865.
Deng, Z., Zhang, H., Fu, Y., Wan, L., & Lv, L. (2018). Research on intelligent expert system of green
cutting process and its application. Journal of Cleaner Production, 185, 904–911. doi:10.1016/j.
jclepro.2018.02.246
Djapic, M., Lukic, L., Fragassa, C., Pavlovic, A., & Petrovic, A. (2017). Multi-agent team for engineer-
ing: A machining plan in intelligent manufacturing systems. International Journal of Machining and
Machinability of Materials, 19(6), 505–521. doi:10.1504/IJMMM.2017.088893
Elaakil, R., Ahmed, R., El Mesbahi, A., & Jaider, O. (2017). Technical Data Extraction and Representa-
tion in Expert CAPP System. Transactions on Machine Learning and Artificial Intelligence, 5(4).
Engelmore, R. S., & Feigenbaum, E. (1993). Expert systems and artificial intelligence. Expert Systems:
International Journal of Knowledge Engineering and Neural Networks, 100, 2.
Fang, N., Fang, N., Pai, P. S., & Edwards, N. (2016). Neural Network Modeling and Prediction of Sur-
face Roughness in Machining Aluminum Alloys. Journal of Computer and Communications, 4(5), 1–9.
doi:10.4236/jcc.2016.45001
Guiotoko, E. H., Aoyama, H., & Sano, N. (2017). Optimization of hole making processes consider-
ing machining time and machining accuracy. Journal of Advanced Mechanical Design, Systems, and
Manufacturing, 11(4).
Harinath Gowd, G., Theja, K. D., Rayudu, P., Goud, M. V., & Roa, M. S. (2014). Modeling & Analysis
of End Milling Process Parameters Using Artificial Neural Networks. Applied Mechanics and Materials,
592, 2733–2737. doi:10.4028/www.scientific.net/AMM.592-594.2733
Honavar, V. (1995). Symbolic artificial intelligence and numeric artificial neural networks: towards a
resolution of the dichotomy. In Computational Architectures integrating Neural and Symbolic Processes
(pp. 351–388). Boston, MA: Springer.
Hou, T. H., & Lin, L. (1995). Using neural networks for the automatic monitoring and recognition of
signals in manufacturing processes. In Design and implementation of intelligent manufacturing systems
(pp. 141–160). Prentice-Hall, Inc.
Hu, P., Han, Z., Fu, Y., & Fu, H. (2016). Implementation of Real-Time Machining Process Control Based
on Fuzzy Logic in a New STEP-NC Compatible System. Mathematical Problems in Engineering, 2016.

155
 
A Review of Artificial Intelligence Technologies to Achieve Machining Objectives

Ighravwe, D. E., & Oke, S. A. (2015). Machining performance analysis in end milling: Predicting using
ANN and a comparative optimisation study of ANN/BB-BC and ANN/PSO. Engineering Journal (New
York), 19(5), 121–137.
Iqbal, A., & Al-Ghamdi, K. A. (2017). Incorporating Energy Efficiency in Performance Measures of
Machining: Experimental Investigation and Optimization. In Sustainable Machining (pp. 47–65). Cham:
Springer. doi:10.1007/978-3-319-51961-6_3
Ji, W., Wang, L., Haghighi, A., Givehchi, M., & Liu, X. (2018). An enriched machining feature based
approach to cutting tool selection. International Journal of Computer Integrated Manufacturing, 31(1),
1–10.
Jogendra, J., Amit, K. V., & Sanjay, V. (2018). Experimental Investigation of Machining Parameters on
Milling Machine using Grey-Fuzzy Logic Method. International Journal of Mechanical and Production
Engineering, 6(2).
Kalaichelvi, V., Karthikeyan, R., Sivakumar, D., & Srinivasan, V. (2012). Tool wear classification using
fuzzy logic for machining of al/sic composite material. Modeling and Numerical Simulation of Material
Science, 2(02), 28–36. doi:10.4236/mnsms.2012.22003
Kannan, T. D. B., Kumar, B. S., & Baskar, N. (2014). Application of artificial neural network modeling
for machining parameters optimization in drilling operation. Procedia Materials Science, 5, 2242–2249.
doi:10.1016/j.mspro.2014.07.433
Khorasani, A., & Yazdi, M. R. S. (2017). Development of a dynamic surface roughness monitoring sys-
tem based on artificial neural networks (ANN) in milling operation. International Journal of Advanced
Manufacturing Technology, 93(1-4), 141–151. doi:10.100700170-015-7922-4
Kilickap, E., Yardimeden, A., & Celik, Y. H. (2017). Mathematical modelling and optimization of cutting
force, tool wear and surface roughness by using artificial neural network and response surface methodol-
ogy in milling of Ti-6242S. Applied Sciences, 7(10), 1064. doi:10.3390/app7101064
Kumudha, S., Srinivasa, Y. G., & Krishnamurthy, R. (1998). Estimation of Tool Status Using Artificial
Neural Network in Face Milling Operation. Journal for Manufacturing Science and Production, 1(3),
189–198. doi:10.1515/IJMSP.1998.1.3.189
Lipiński, D., Bałasz, B., & Rypina, Ł. (2018). Modelling of surface roughness and grinding forces using
artificial neural networks with assessment of the ability to data generalisation. International Journal of
Advanced Manufacturing Technology, 94(1-4), 1335–1347. doi:10.100700170-017-0949-y
Madhava Reddy, S., Chennakesava Reddy, A., & Sudhakar Reddy, K. (2012). Latest Developments in
Condition Monitoring of Machining Operations. Journal of Applied Sciences (Faisalabad), 12(10),
938–946. doi:10.3923/jas.2012.938.946
Mahesh, T. P., & Rajesh, R. (2014). Optimal selection of process parameters in CNC end milling of Al
7075-T6 aluminium alloy using a Taguchi-fuzzy approach. Procedia Materials Science, 5, 2493–2502.
doi:10.1016/j.mspro.2014.07.501

156

A Review of Artificial Intelligence Technologies to Achieve Machining Objectives

Makhfi, S., Haddouche, K., Bourdim, A., & Habak, M. (2018). Modeling of machining force in hard
turning process. Mechanics, 24(3), 367–375. doi:10.5755/j01.mech.24.3.19146
Malghan, R. L., Rao, K., Shettigar, A. K., Rao, S. S., & D’Souza, R. J. (2018). Forward and reverse
mapping for milling process using artificial neural networks. Data in Brief, 16, 114–121. doi:10.1016/j.
dib.2017.10.069 PMID:29188231
Man, K. F., Tang, K. S., & Kwong, S. (1996). Genetic algorithms: Concepts and applications. IEEE
Transactions on Industrial Electronics, 43(5), 519–534. doi:10.1109/41.538609
Manafi, D., Nategh, M. J., & Parvaz, H. (2017). Extracting the manufacturing information of machining
features for computer-aided process planning systems. Proceedings of the Institution of Mechanical Engi-
neers. Part B, Journal of Engineering Manufacture, 231(12), 2072–2083. doi:10.1177/0954405415623487
Minsky, M. (1974). A framework for representing knowledge.
Mishra, R., Malik, J., Singh, I., & Davim, J. P. (2010). Neural network approach for estimating the re-
sidual tensile strength after drilling in uni-directional glass fiber reinforced plastic laminates. Materials
& Design, 31(6), 2790–2795. doi:10.1016/j.matdes.2010.01.011
Nafeez, A., Tomohisha, T., & Yoshio, S. (2014). Machining parameter optimisation by genetic algorithm
and artificial neural network. International Journal of Data Analysis Techniques and Strategies, 6(3),
261–274. doi:10.1504/IJDATS.2014.063061
Namdari, M., Jazayeri-Rad, H., & Hashemi, S. J. (2014). Process fault diagnosis using support vector
machines with a genetic algorithm based parameter tuning. Journal of Automation and Control, 2(1), 1–7.
Naresh, N., (2014). Modeling and Analysis of Machining GFRP Composites Using Fuzzy Logic and
ANOVA. IUP Journal of Mechanical Engineering, 7(4).
Nassehi, A., Essink, W., & Barclay, J. (2015). Evolutionary algorithms for generation and optimization
of tool paths. CIRP Annals, 64(1), 455–458. doi:10.1016/j.cirp.2015.04.125
Nathan, R. D., Vijayaraghavan, L., & Krishnamurthy, R. (2001). Intelligent estimation of burning lim-
its to aid in cylindrical grinding cycle planning. International Journal of Heavy Vehicle Systems, 8(1),
48–59. doi:10.1504/IJHVS.2001.001154
Pendokhare, D. G., & Quazi, T. Z. (2012). Fuzzy Logic Based Drilling Control Process. Int. J. Scientific
and Engg. Research, 5(12), 61–65.
Pohokar, N., & Bhuyar, L. (2014). Neural Networks Based Approach for Machining and Geometric
Parameters optimization of a CNC End Milling. Neural Networks, 3(2).
Prabhakar, A., Kingsley, D., Singh, J., & Jebaraj, C. (2004). Creating process plan sheet from a feature
based model. In Proceedings of the National Conference on Advanced Manufacturing and Robotics
(pp. 247-254).
Purdy, M., & Daugherty, P. (2016). Why Artificial Intelligence is the Future of Growth.
Raimond, K. (2008). Effective tool wear estimation through multisensory information fusion using
Artificial Neural Network. Journal of EEA, 25, 33–42.

157
 
A Review of Artificial Intelligence Technologies to Achieve Machining Objectives

Raja, C., & Saravanan, M. (2018). Tool path optimization by genetic algorithm for energy efficient
machining. TAGA Journal of Graphic Technology, 14, 1670–1679.
Rajasekhar, K., & Naresh, N. (2014). Modeling and analysis of process parameters in machining Aisi
304 stainless steel using fuzzy logic. i-Manager’s Journal on Mechanical Engineering, 4(2), 18.
Rishi, K., Pradhan, M. K., & Rajesh, K. 2014. Modeling and optimization of milling parameters on Al-
6061 alloy using multi- objective genetic algorithm. In Proc. of 26th AIMTDR Conference, IIT Guwahati,
India, December 12-14.
Roy, A., & Bhagat, K. (2015). An application of Artificial Neural Network to predict surface rough-
ness during drilling of AISI1020 steel. International Journal of Scientific Research Engineering and
Technology, 4(8).
Salimi, A., Özdemir, A., & Erdem, A. (2015). Simulation and monitoring of the machining process via
fuzzy logic and cutting forces. Iranian Journal of Materials Science and Engineering, 12(3), 14–26.
Sarhan, A. A., Sayuti, M., & Hamdi, M. (2012). A Fuzzy Logic Based Model to Predict Surface roughness
of a machined surface in glass milling operation using CBN grinding tool. World Academy of Science,
Engineering and Technology, 6, 564–570.
Šarić, T., Šimunović, G., Lujić, R., Šimunović, K., & Antić, A. (2016). Use of soft computing technique
for modelling and prediction of CNC grinding process. Technical Gazette, 23(4), 1123–1130.
Senthilkumar, N., Sudha, J., & Muthukumar, V. (2015). A grey-fuzzy approach for optimizing machining
parameters and the approach angle in turning AISI 1045 steel. Advances in Production Engineering &
Management, 10(4), 195–208. doi:10.14743/apem2015.4.202
Shaik, J. H., & Srinivas, J. (2017). Optimal selection of operating parameters in end milling of Al-6061
work materials using multi-objective approach. Mechanics of Advanced Materials and Modern Processes,
3(1), 5. doi:10.118640759-017-0020-6
Sudhir, K., & Rajagopalan, R. (1992). An artificial intelligence approach to precedence network generation
for assembly line balancing. Computers in Industry, 18(2), 177–191. doi:10.1016/0166-3615(92)90112-Z
Tanikić, D., Marinković, V., Manić, M., Devedžić, G., & Ranđelović, S. (2016). Application of response
surface methodology and fuzzy logic based system for determining metal cutting temperature. Bulletin
of the Polish Academy of Sciences. Technical Sciences, 64(2), 435–445. doi:10.1515/bpasts-2016-0049
The Manufacturer. (2018, February 26). Power of artificial intelligence. Retrieved from https://www.
themanufacturer.com/articles/power-artificial-intelligence-manufacturing/
Tumuluru, J. S., & McCulloch, R. (2016). Application of hybrid genetic algorithm routine in optimiz-
ing food and bioengineering processes. Foods, 5(4), 76. doi:10.3390/foods5040076 PMID:28231171
Vijaykumar, K., Panneerselvam, K., & Sait, A. N. (2014). Machining Parameter Optimization of
Bidirectional CFRP Composite Pipe by Genetic Algorithm. Materials Testing, 56(9), 728–736.
doi:10.3139/120.110623

158

A Review of Artificial Intelligence Technologies to Achieve Machining Objectives

Zadeh, L. (1965). Fuzzy sets. Information and Control, 8(3), 338–353. doi:10.1016/S0019-9958(65)90241-
X
Zadeh, L. A. (1973). Outline of a new approach to the analysis of complex systems and decision
processes. IEEE Transactions on Systems, Man, and Cybernetics, SMC-3(1), 28–44. doi:10.1109/
TSMC.1973.5408575
Zhang, C., Yao, X., Zhang, J., & Jin, H. (2016). Tool condition monitoring and remaining useful
life prognostic based on a wireless sensor in dry milling operations. Sensors (Basel), 16(6), 795.
doi:10.339016060795 PMID:27258277
Zhu, K., & Zhang, Y. (2018). A Cyber-Physical Production System Framework of Smart CNC Machining
Monitoring System. IEEE/ASME Transactions on Mechatronics, 1. doi:10.1109/TMECH.2018.2834622

159