JQME

4,3 Optimization of maintenance

organization and manpower in
process industries
168
Gerald M. Knapp
Louisiana State University, Baton Rouge, Louisiana, USA, and
Milind Mahajan
Mastercard International, St Louis, Missouri, USA

Introduction
A maintenance system is an element of a complete production system and is
directly related to profitability through equipment output and equipment
running cost. The maintenance organization forms the foundation on which
maintenance control rests. The organization must clearly define the
responsibility, accountability, and function of the workforce. Unfortunately, in
many companies maintenance organization has been the result of random
growth and unplanned reactions to the management needs. Such maintenance
organizations are often operating far below their peak efficiency.
The process industries are particularly vulnerable to plant, process, and
product failures. Among the main reasons are the complexities of the
technology involved, the range of products, and the often-subtle nature of
compositional changes, process shifts and hidden product flaws (Witherell,
1991).
In manpower planning, one is concerned with the description and prediction
of the behavior of large groups of people. Manpower planning is an attempt to
match the supply of people with the jobs available for them and is concerned
with having the right number of workers in the right places at the right time
(McClean, 1991). The model developed in this work is unique in that it examined
the simultaneous decision of both manpower levels and the structuring of that
manpower between area and centralized departments, a problem often
encountered within the process industries.

Literature review
The use of mathematical models for manpower planning is well established and
can be traced back to 1779 when Rowe used a model to plan careers in the Royal
Marines (McClean, 1991). Mathematicians and statisticians have done much
work on the development of models of manpower systems in the years since the
Journal of Quality in Maintenance
1950s. Many organizations have made successful use of such models, but
Engineering, Vol. 4 No. 3, 1998,
pp. 168-183, © MCB University
despite these successes, manpower planning models are only gradually coming
Press, 1355-2511 into widespread use.
 Today, manpower models are required for effective use of the available Optimization of
manpower to reduce the cost of the manpower and have become an essential maintenance
tool for the modern manager, especially in a competitive business environment. organization
Manpower planning models are usually divided into two main categories:
(1) Supply models, which analyze current manpower stocks and their
projected flows to predict future stocks.
169
(2) Demand models, which determine current or future manpower needs for
given workload demands.
It can be further viewed at two levels as given below (Zanakis and Maret, 1981):
(1) Macro level – considers aggregate numbers of people per category, used
for annual or long-range planning.
(2) Micro level – mainly used for job type and personnel matching
assignment.
Leeson (1982), Bartholomew and Forbes (1979), and Vassiliou (1978) have
developed manpower supply models dealing only with the internal flows such
as promotions and wastage of the manpower, and do not consider demand
requirements or organizational structure. Gass (1991) developed a military
manpower supply model, which uses transition rate models to forecast
personnel inventory. Though the model considers different grades, skills and
experience among the workforce, it deals only with internal flow and wastage
of the personnel (workforce) inventory. Stewart et al. (1994) developed a
mathematical model for optimizing the cost of training the workforce
depending on the production demand.
Singhal (1992) used discrete optimal control theory to develop a non-iterative
algorithm for workforce planning in a multi-product environment. He uses a
quadratic cost function which allows for a realistic cost structure in the
planning process. The model does not consider the organizational structure of
the manpower.
While doing maintenance planning, not only are the number of craftsmen
important, but craft-type and training levels are also equally important. Chu et
al. (1991) and Chu and Lin (1993) discuss a case study of a Hong Kong-based
electronic company. The study did not consider the cost of the manpower, but
rather considered the least possible travel time as the criterion. Here the focus
was to provide the fastest service to the customer and on applying the model to
the available manpower and suggesting the best structure; however, it did not
consider optimizing manpower levels. Curtis (1987) and Ntuen (1989) use simple
forecasting techniques such as exponential smoothing and regression analysis
to estimate aggregate manpower requirements. Although the method is very
simple, it requires extensive data to be precise. Dijkstra et al. (1991; 1994)
developed a decision support system (DSS) for capacity planning of aircraft
maintenance personnel. The DSS is used to guarantee that sufficient engineers
with appropriate skills (qualifications) are always available at the airport for
carrying out the maintenance work. Here the main focus is on the effective
JQME scheduling of the maintenance jobs. Though it considers different craft-type, it
4,3 does not consider different maintenance areas as they are not required in the
case under study. The main shortcoming is that it considers the organization of
the maintenance workforce in the identical and smallest possible teams to
match the required workload. This would not be an efficient use of the resources
in many plants. Duffuaa and Raouf (1992) conclude that in a plant, the size of
170 the maintenance staff must be related to the level of output of the plant. The goal
of the developed simulation model is to estimate the optimal size of a
maintenance crew in a plant.
Hambleton (1982) developed a manpower planning model for maintenance
personnel considering different maintenance areas for forecasting the
maintenance crew. He measures service level as the response time to a
breakdown call.

Maintenance manpower model overview

In a maintenance organization, manpower can be further classified into
categories, such as areas (technical and/or geographical), craft-type (fitter,
welder, etc.), training levels (experience), and sub-contracted and in-house
employees. The following points were considered:
• centralized versus decentralized departments;
• the level and mix of the in-house and sub-contracted employees in each
area;
• depending on the workload in different areas, whether employees should
be assigned to a particular area or to the float;
• need to retain expertise in a particular area.
The objective of the developed model is to reduce the cost of the resources used
in the maintenance department. Maintenance cost includes cost of the spare
parts, material, manpower and the cost of down time. Ideally, maintenance
should be optimized relative to the true costs of maintenance – including lost
production, costs of storages required due to breakdowns, insurance
costs/liability expenses as a result of maintenance failures, etc. In reality, these
costs are extremely difficult to track or even estimate in most plants. Few plants
track this data, and fewer track it accurately. In this work, only the cost of
manpower (wages) is directly considered. Wages depend on craft-type, training
level, and whether the employee is in-house or sub-contracted. For simplicity,
we have utilized average salaries for each worker depending on his category,
craft-type and training level. In reality, workers in the same category, craft-type
and training level will have a range of salaries.
Though size and type of plant, level of operation of the plant, age of plant,
and organizational structure should be considered while estimating the
workload demand, we have considered only organizational structure as the
main variable affecting the maintenance workload demand. Data for estimating
workload should be available if the plant has good accounting practices.
 The workload demand is segregated in the categories, such as maintenance Optimization of
area, type of maintenance (emergency, preventive or routine), and craft-type. maintenance
This workload segregation is essential because the manpower of required type organization
should be available at the required place to do the job when required. From the
past work-orders this information should be readily available.
Maintenance work of a particular type can be performed only by craftsmen
with the required capabilities. The number of workers of a particular category 171
required in any maintenance area will depend on the productivity, salary and
efficiency of the craftsman, and the organizational structure (these variables are
discussed in the following sections). These effects are considered in this model.
Productivity can be defined as the percentage of the total time a worker
spends on productive work. Sometimes workers are required to do paper,
arrange resources, travel, etc.; these are considered non-productive jobs. The
time the worker actually spends on productive (maintenance) work will give us
the productivity of that worker. The productivity factor can be calculated by
doing job or work sampling. Mann (1983) discusses at length the work sampling
and analysis techniques. In our model, the productivity factor is considered the
same for all areas, craft and training levels, but is different for in-house
employees and sub-contracted employees.
Different employees of a required craft will take different times for doing a
given job. The time required to complete a job will depend on the level of
training of a craftsman. This can be seen as an “efficiency” of a particular
craftsman to do the job. For simplicity we have considered efficiency of a
particular craft-type and expense level to be the same across all maintenance
areas. In a maintenance area, depending on the level of emergency work and
safety requirements, some restrictions may be imposed on the level of training
of the craftsman allowed to work in that area. This constraint is accounted for
in the model by assigning the required level of craftsman to the respective
maintenance areas.
There are many different possibilities for allocating available manpower.
Organizational structures depend on the following main factors:
• Workload – this helps to estimate the number of employees required.
• Level of sub-contracted employees – this depends on the level of in-house
employees and the peaks in the workload demand.
• Centralized or decentralized – this depends on the size of the plant and
the management, each type has its own advantages and disadvantages.
If a worker is required to work on a floating basis, then he will be required to
travel from a centralized office. This will reduce the productivity of that worker
compared to other workers of similar category who are fixed in that area. In
turn, the reduction in the productivity will depend on the traveling time. In the
model, a decentralization factor is considered the same for a given area because
it largely depends on the distance to be traveled. The decentralization factor is
considered as a penalty for not assigning the worker to the area.
JQME The main assumptions of the model are as follow:
4,3 • the demand distribution is stationary for a given period of interest;
• all the hired employees (in-house and sub-contracted) for a given period
work full time;
• workload demand in the maintenance areas are considered to be
172 independent.
Model variations not considered are given below:
• ability to use overtime;
• workforce mobility: internal promotion or demotion due to training or
growth in seniority, retirement, transfers, etc.;
• no absenteeism is considered, except possibly as a static percentage
reduction in the workforce.

Notation
Subscript i and s stand for in-house and sub-contracted employees respectively.

Indices
A = Maintenance area.
C = Craft classification.
T = Training level of the craft-type “c”.
L = Priority levels such as emergency, preventive and routine
maintenance work.

Parameters
r[l,a,c] = Value of workload demand for each priority level, area and
craft level at a given time (man-hours/month).
f(r)[l,a,c] = PDF of workload demand RVs for each priority level, area and
craft level.
C[c,t] = Average salary per employee, same for all areas ($/month).
E[a,c] = Emergency maintenance workload demand, (man-hours/
month).
H = Current number of in-house employees (optional).
P[a,c] = Preventive maintenance workload demand, (man-hours/
month).
R[a,c] = Routine maintenance workload demand, (man-hours/month).
W[a,c] = Total maintenance workload demand, (man-hours/month).
Wwl[a] = Maintenance workload demand satisfied by the workers
assigned to the area “a”, a = 1..A (man-hours/month).
Pi/s = Productivity of employees, same for all areas 0 ≤ Pi/s ≤ 1.
Yi/s[t] = Efficiency of employees, same for all areas, 0 ≤ Yi/s ≤ 1.
X[a] = Decentralization factor, assigned to area “a”, where 0 ≤ X[a] ≤
1.
 VT[a,c,t] = Minimum number of craftsmen of type “c” required in the Optimization of
maintenance area “a” of the training level “t” such that t ≥ T. maintenance
organization
Decision variables
Nia[a,c,t] = Total number of in-house employees assigned to area “a”,
a = 1..A, c = 1..C and t = 1..T.
Nif[c,t] = Total number of in-house employees assigned to float.
173
Nsa[a,c,t] = Total number of sub-contracted employees assigned to area
“a”, a = 1..A, c = 1..C and t = 1..T.
Nsf[c,t] = Total number of sub-contracted employees assigned to float.

Manpower model
The objective is to minimize the total cost of the overall system.

(1)

Subject to:

(2)

(3)

(4)

(5)

where,

(man-hours/month) (6)

*160 (man-hours/month) (7)

JQME An optional constraint may also be added to force the sum of all in-house
4,3 employees allocated to equal H, the current number of in-house employees. The
model is then forced to allocate all in-house employees before considering
contract employees.
In this model, the cost of material and other resources are not considered
because it will be the same irrespective of the organizational structure of the
174 maintenance department. The total cost is the cost of the manpower; that is, the
wages of all the in-house and the sub-contracted employees.
It is assumed that in-house employees are utilized first to satisfy workload
demand. If the workload demand is more than can be satisfied by in-house
employees, then sub-contracted employees are hired. The workload demand
should be satisfied by all the hired employees (in-house and contracted). The
productivity, efficiency and the decentralization factors are used to estimate the
workload that can be satisfied by all the hired employees. We utilize a time
period of one month, with workload demand data given in man-hours/month. In
one month, any full-time employee will work 160 hours, so Equation 2 checks
the viability of the manpower hired to satisfy the workload demand.
If the total workload in a given area exceeds the workload that can be
performed by the employees assigned to that area then the department has to
assign either employees from the float or hire sub-contracted employees to do
the jobs (since we are not considering any overtime). The number of sub-
contracted employees required will depend on the organizational structure of
the department and the workload. The employees assigned to the area are
utilized first and if the workload demand is more, then the employees from the
float are utilized; this depends on the level of employees already assigned to that
area. Equation 3 estimates the employees required from the float.
Sometimes there is a requirement that, in the particular area, only a certain
level of trained craftsman can work. The minimum level of training required in
the area is specified and is satisfied by the Equation 4. Equation 5 checks
whether the estimated values are positive or not.
The workload demand is expected to have different distribution for each of
the priority levels so the total workload demand will be the sum of workload
demand for all priority levels. Equation 6 calculates the workload demand for
each area, craft and the priority levels. Equation 7 calculates the workload
demand satisfied by the workers assigned to the area, so that the number of
workers assigned to the float can be estimated.

Model implementation
The optimization model was implemented in LINGO, an optimization modeling
package.
The data required for the model are expected to be readily available, as
computerized maintenance information systems are now used by most
maintenance organizations. Past records of work-orders should provide the
data required, such as total craftsmen required and man-hours required, for
every month.
 SIMAN, a simulation software package, was used to generate test data for Optimization of
demonstrating the model (while representative of a typical process plant, actual maintenance
plant data were not used). The following input data are required; this data can organization
be collected from the past historical database of work-orders, experience of the
supervisors, and by observations:
• Total maintenance man-hours required for every priority level, area,
craft, and training level (man-hours/month). Data for workload demand 175
in every maintenance area were generated by SIMAN for every craft-
type and priority level and the total workload is given by:
Total workload = Emergency maintenance work + Preventive
maintenance work + Routine maintenance work
• The productivity and efficiency (in-house and sub-contracted employees)
and decentralization factor was assumed.
• Wages of all employees were assumed for both in-house and sub-
contracted employees (for every craft and training level).
After running the model, it gives the hierarchical manpower required for the
given maintenance workload. The data for the required parameters are stored in
a separate file so it can be changed as requirements change.
Table I shows sample data for workload demand in man-hours/month for
each area and craft-type.
The workload demand for emergency, preventive and routine maintenance
was assumed to have a lognormal distribution; however, any distribution could
be utilized. SIMAN was used to fit the distribution of the dataset and was found
to be lognormal. Using the lognormal distribution function and the parameters
the value of the workload in man-hours for one month is as given below:

(man-hours/month)(8)

where, µ[l,a,c], σ[l,a,c] – mean and standard deviation for dataset of maintenance
type l, area 1..a, and craft-type 1..c.
The emergency, preventive and routine maintenance workload demand in
man-hours/month were combined, which represents the overall workload

Craft-type
Maintenance area Craft 1 Craft 2 Craft 3 Craft 4

Area 1 44,404 2,040 1,360 1,020

Area 2 17,298 21,369 4,930 2,236 Table I.
Area 3 1,335 3,332 15,853 15,300 Workload demand in
Area 4 697 38,480 32,147 553 man-hours/month
JQME demand in a given area and for a given craft-type. Since the distribution of all
4,3 the workloads are the same, parameters (mean and standard deviation) of PDF
for the workload for a given area are calculated as given below:

(9)
176
(10)

where, µ[a], σ[a] – mean and standard deviation for dataset of total maintenance
workload for area 1..a.
Table II shows data for salary, efficiency, and productivity. Salary is given in
$/month for each craft-type, training level, and type of workers. Salary is
assumed to be higher for the higher training levels.

In-house employees Sub-contracted employees

Training level Training level
Identifier Craft-type Trng-1 Trng-2 Trng-3 Trng-1 Trng-2 Trng-3

Salary Craft 1 1800 1600 1480 1656 1472 1362

($/month) Craft 2 1952 1920 1600 1796 1766 1472
Table II. Craft 3 1856 1792 1600 1708 1649 1280
Input data for Craft 4 2400 2024 1760 2208 1863 1620
salary, efficiency Efficiency All crafts 0.94 0.89 0.83 0.92 0.87 0.8
and productivity Productivity All crafts 0.83 0.74

Productivity is stated as a fraction and is given for each training level and craft-
type. Efficiency is also stated as a fraction and is given for each craft-type. The
decentralization factor is also assumed and can be easily changed if the actual
data are available. Since it is a penalty imposed due to not assigning the
workers to the area so it is assumed to be less than one.

Demonstration
Table III demonstrates the output given by the manpower model for the sample
input given in Tables I and II. The optimization model gives the hierarchical
manpower allocation for each of the following categories:
• maintenance area;
• craft-type;
• training level;
• type of employees – sub-contracted or in-house;
• whether floating (centralized) or assigned to the maintenance areas.
 Employees assigned to areas
Optimization of
Maint. In-house employees Sub-contracted employees maintenance
area Craft-type Trng-1 Trng-2 Trng-3 Trng-1 Trng-2 Trng-3 organization
Area 1 Craft 1 5 2 234
Craft 2 4 1
Craft 3 3 2 177
Craft 4 1 1
Area 2 Craft 1 1 2 1
Craft 2 5 2 1
Craft 3 1 2 2
Craft 4 6 3 3
Area 3 Craft 1 4 2 3
Craft 2 2 3 2
Craft 3 6 4 6
Craft 4 5 4 116
Area 4 Craft 1 2 2
Craft 2 1 2 155
Craft 3 2 2 154
Craft 4 2

Floating employees
In-house employees Sub-contracted employees
Craft-type Trng-1 Trng-2 Trng-3 Trng-1 Trng-2 Trng-3

Craft 1 167
Craft 2 203
Craft 3 141
Craft 4 10 Table III.
Output of the
Cost of manpower ($) 2,031,567 optimization model

Model validation
This research was motivated by a real-world problem brought to us by a large
industrial chemical processing corporation. The company’s local processing
facility was addressing the issue of whether to centralize any of its maintenance
crafts. Unfortunately, as is often the case with industrial partners, when it came
time to get actual data, the company could not release specifics due to the
confidential and competitive nature of the data. Therefore, to validate the
optimization model, a simulation model in SIMAN representing a maintenance
department of a chemical process industry was created. The advantage of the
simulation model is that it gives us statistical information, such as utilization of
the workers and the queue length of the failed machines in each area and for
each craft-types. This information can be used to determine if the “optimized”
model is performing as well or better than the original allocation.
In the simulation model, machines were assumed to have exponential failure-
time distributions. After failure, the equipment requires a certain number of
workers of different craft-types for the repair job, determined by certain
JQME probability. On failure, equipment is sent to a queue and waits for the workers
4,3 to become available. If all the craftsmen are busy, then the required number of
craftsmen from float are requested to do the job. Repair times are considered to
have an exponential distribution with different means for the different
equipment items. Furthermore, the generated repair time is modified to account
for non-productive time by dividing it by the “worst” (lowest) craft productivity
178 factor for the required crew.
Queue length is the number of jobs in a queue waiting for a craftsman to
become available. Worker utilization and queue length should be considered
together as a performance measure because as the worker utilization goes on
increasing there is a possibility that the queue length may also go on increasing,
which will increase the down time of the equipment.
The data generated by the simulation model were analyzed as described
above, and an ASCII input file for LINGO was prepared. The hierarchical
manpower allocation for this workload demand was then generated using the
optimization model. The ASCII output file was then retyped in a spreadsheet
for analysis.
Output of the optimization model gives the hierarchical manpower allocation
for the given workload demand. This manpower allocation is used as an input
to the simulation model for the same workload demand that was used for the
optimization model. The simulation was then run to determine the performance
of the new manpower allocation.
The output of the simulation model provides the resource utilization and
queue lengths. The performance measures for the optimized manpower
allocation were compared with the resource utilization given by the earlier
model.
The simulation model was run for 20 replications, which gave 20 data points
for one set of workload demand for each area and craft-type. Using SIMAN the
distribution of the datasets and the parameters were generated. The distribution
of a particular dataset was found to be a normal distribution. Using the given
parameters and the type of distribution (in this case normal), workload demand
in man-hours/month can be calculated. This information about the workload is
used as an input to the optimization model. The output of the decision variables
from the optimization model is used again as input to the simulation model.

Results
A simulation model was run for 20 replications using original manpower
allocation and the manpower allocation done by the optimization model.
Tables IV and V show the means of the various performance measures
(length of the queue and the worker utilization) for both allocations. The
identifiers shown in the table are in the standard format used by SIMAN. The
identifier NQ(A1C11) represents the average queue length of the repair jobs in
the maintenance area A1 waiting for one worker of craft-type C1. The identifier
NR(A1C1) represents the worker utilization of craft-type C1 in the maintenance
area A1.
 Statistical analysis of the dataset of 20 replications
Original (simulation) model Optimization model “t-test” for inference of two means
Mean Std. Dev. Confidence Mean Std. Dev. Confidence at 95% confidence levels
Identifier ×1 Pop. Int. (95%) ×2 Pop. Int. (95) (+T) (–T) Remark

NQ(A1C11) 1.39 0.17 0.081 1.25 0.16 0.073 0.230 0.054 Improved
NQ(A1C12) 2.31 0.29 0.135 1.77 0.22 0.104 0.680 0.406 Improved
NQ(A1C21) 0.92 0.12 0.054 0.97 0.12 0.057 0.016 –0.110 No difference
NQ(A1C22) 0.52 0.06 0.030 0.45 0.06 0.026 0.103 0.039 Improved
NQ(A2C11) 2.03 0.25 0.119 1.94 0.24 0.113 0.226 –0.038 No difference
NQ(A2C12) 5.52 0.69 0.323 2.50 0.31 0.146 3.306 2.735 Improved
NQ(A2C21) 0.47 0.06 0.028 0.45 0.06 0.026 0.054 –0.007 No difference
NQ(A2C22) 0.00 0.00 0.000 0.00 0.00 0.000 0.000 0.000 No difference
NQ(FC11) 9.77 1.22 0.571 8.94 1.12 0.523 1.454 0.208 Improved
NQ(FC12) 13.79 1.72 0.807 9.84 1.23 0.575 4.750 3.156 Improved
NQ(FC21) 18.48 2.31 1.081 15.12 1.89 0.884 4.487 2.240 Improved
NQ(FC22) 1.37 0.17 0.080 1.23 0.15 0.072 0.221 0.047 Improved
NR(A1C1) 24.99 3.12 1.462 26.04 3.25 1.523 0.653 –2.744 No difference
NR(A1C2) 23.97 3.00 1.402 25.69 3.21 1.503 –0.061 –3.369 Improved
NR(A2C1) 35.30 4.41 2.065 37.65 4.71 2.203 0.071 –4.788 No difference
NR(A2C2) 26.25 3.28 1.536 29.73 3.72 1.740 –1.617 –5.351 Improved
NR(FC1) 16.64 2.08 0.974 19.70 2.46 1.152 –1.841 –4.269 Improved
NR(FC2) 1.89 0.24 0.111 7.57 0.95 0.443 –5.314 –6.049 Improved

Cost ($) 736,909 723,777 Improved

(Case I)
optimization models
simulation and the
Comparison of
179
maintenance
Optimization of

organization

Table IV.
 4,3

180

(Case II)
JQME

Table V.
Comparison of
simulation and the
optimization models
Statistical analysis of the dataset of 20 replications
Original (simulation) model Optimization model “t-test” for inference of two means
Mean Std. Dev. Confidence Mean Std. Dev. Confidence at 95% confidence levels
Identifier ×1 Pop. Int. (95%) ×2 Pop. Int. (95) (+T) (–T) Remark

NQ(A1C11) 0.59 0.089 0.0414 0.53 0.077 0.0360 0.104 0.016 Improved
NQ(A1C12) 0.98 0.147 0.0688 0.75 0.109 0.0509 0.299 0.161 Improved
NQ(A1C21) 0.20 0.030 0.0140 0.42 0.061 0.0285 –0.194 –0.246 Deteriorated
NQ(A1C22) 0.22 0.033 0.0154 0.19 0.028 0.0129 0.046 0.014 Improved
NQ(A2C11) 0.86 0.129 0.0604 0.82 0.119 0.0557 0.106 –0.026 No difference
NQ(A2C12) 2.34 0.351 0.1643 1.06 0.154 0.0720 1.424 1.136 Improved
NQ(A2C21) 0.00 0.000 0.0000 0.19 0.028 0.0129 –0.180 –0.200 Deteriorated
NQ(A2C22) 0.00 0.000 0.0000 0.00 0.000 0.0000 0.000 0.000 No difference
NQ(FC11) 7.29 1.094 0.5118 6.67 0.967 0.4529 1.170 0.070 Improved
NQ(FC12) 10.29 1.544 0.7224 7.34 1.064 0.4983 3.656 2.244 Improved
NQ(FC21) 13.79 2.069 0.9681 11.28 1.636 0.7658 3.503 1.517 Improved
NQ(FC22) 0.00 0.000 0.0000 1.02 0.148 0.0693 –0.964 –1.076 Deteriorated
NR(A1C1) 18.65 2.798 1.3093 19.43 2.817 1.3192 0.715 –2.275 No difference
NR(A1C2) 17.89 2.684 1.2559 19.17 2.780 1.3015 0.175 –2.735 No difference
NR(A2C1) 26.34 3.951 1.8492 28.10 4.075 1.9078 0.377 –3.897 No difference
NR(A2C2) 19.59 2.939 1.3753 22.19 3.218 1.5066 –0.959 –4.241 Improved
NR(FC1) 12.42 1.863 0.8719 14.70 2.132 0.9980 –1.214 –3.346 Improved
NR(FC2) 1.41 0.212 0.0990 5.65 0.819 0.3836 –3.921 –4.559 Improved

Cost ($) 549,932 540,132 Improved

The mean and standard deviation shown in the Tables IV and V are for the Optimization of
population of 20 data points. Population-1 is for original (simulation) model and maintenance
population-2 is for optimization model. A t-test was carried out to determine organization
whether a statistical significance exists between the two means. The interval
estimate of the difference between the means of two populations is given by:
181
(11)

Where,
n1 is size of the population-1 and is 20.
n2 is size of the population-2 and is 20.
The interval from –T to +T of the difference of the means is calculated at 95 per
cent confidence level. If the difference in the means (x1-x2) for the two
populations is positive for the identifier “queue length” and zero is not included
in the interval of –T to +T, then we can statistically conclude that the
performance of the optimization model has improved over the original model.
For identifier NQ(A1C11) the interval for the difference of two means is from
0.0540 to 0.230, which means it is positive and zero is not included in the
interval so we can conclude with 95 per cent confidence that the performance of
the optimized system is improved.
If the difference in the means (x1-x2) for the populations is negative for the
identifier worker utilization and zero is not included in the interval of –T to +T,
then we can statistically conclude that the performance of the optimization
model has improved over the original model. For identifier NR(A2C2) the
interval for the difference of two means is from –5.351 to –1.617, which means it
is negative and zero is not included in the interval so we can conclude with 95
per cent confidence that the performance of the optimized state is improved.

Conclusions
The objective function (cost of manpower) is improved for the optimized
allocation (see Tables IV and V). The decrease in manpower cost is due to the
use of fewer craftsmen to satisfy the workload demands. Test results indicate
that the performance of the optimization model improved overall as a result of
the reduction / re-allocation of staff.
In Case-I (Table IV) performance measures of all craft-types have either
improved or remained the same for the optimized model over the non-optimized
model. This indicates that the overall performance of the model has improved
even with fewer craftsmen. This is due to allocation of the workers where they
are needed most and effective sharing of the workers by allocating to the float.
Table V shows that by effective allocation of fewer workers, worker utilization
and queue lengths can be improved at the same time.
JQME In Case-II (Table V) the optimized model uses fewer craftsmen, but not all the
4,3 performance measures improved. Performance measures of craft-type C1 for the
optimized model are improved over the non-optimized model. Only for craft-type
C2 is there a deterioration in one of the performance measures. For craft-type C2
all the three performance measures are never deteriorated in any given area; at
least two performance measures are either improved or there is no change in
182 them. The developed model has the objective of reducing cost, so the model tries
to allocate not only fewer workers, but also workers with the lowest salary. Table
II shows the salaries used for all craft-types. Salary for craft-type C1 is lower
than the salary for the craft-type C2 (for both in-house and sub-contracted) and
remaining variables (productivity, efficiency, and decentralization factor) are the
same for both craft-types. When the optimization model allocates manpower, it
tries to lower the cost by allocating fewer workers of craft-type C2 to reduce the
cost. The improved worker utilization and increased queue length of the craft-
type C2 can be due to allocation of fewer workers of craft-type C2. But even with
fewer workers of craft-type C2, not all the queue lengths are deteriorated and
those deteriorated are not large.
Considering both cases and all the performance measures (queue length and
utilization) together for any craft-type, the overall performance of the optimized
model has improved over the non-optimized model (Tables IV and V) for both
examples.
Future research on this model will consider the inclusion of internal flows
(retirement, promotion) and of hiring/firing costs.

Contribution
Failures cannot be avoided completely but the availability of the machines can
be increased and maintenance costs reduced by proper maintenance planning.
Effective use of manpower planning and the organizational structure will help
reduce the cost of maintenance system and increase its efficiency. Due to the
high cost of down time in the chemical process industry, it is vital to estimate
the required resources properly and efficiently. The developed model will assist
in estimating manpower requirements at minimum cost.
References
Bartholomew, D.J. and Forbes, A.F., (1979), Statistical Techniques for Manpower Planning, John
Wiley and Sons.
Chu, S.C.K. and Lin, C.K.Y. (1993), “A manpower allocation model of job specialization”, Journal of
Operational Research Society, Vol. 44 No. 11, pp. 983-89.
Chu, S.C.K., Lin, C.K.Y. and Ng, K.Y.K. (1991), “Centralized versus decentralized manpower
resource planning: the case of Hong Kong company”, Journal of Operational Research Society,
Vol. 42 No. 7, pp. 525-36.
Curtis, R.G. (1987), “Top management systems for forecasting and monitoring pay costs and staff
numbers”, Journal of Operational Research Society, Vol. 38 No. 10, pp. 935-40.
Dijkstra, M.C., Kroon, L.G., van Nunen, J.A. and Salomon, M. (1991), “A DSS for capacity planning
of aircraft maintenance personnel”, International Journal of Production Economics, Vol. 23, pp.
69-78.
Dijkstra, M.C., Kroon, L.G., Salomon, M., van Nunen, J.A. and van Wassenhove, L.N. (1994), Optimization of
“Planning the size and organization of KLM’s aircraft maintenance personnel”, Interfaces, Vol.
24 No. 6, pp. 47-58. maintenance
Duffuaa, S.O. and Raouf, A. (1992), “A simulation model for determining maintenance staffing in organization
an industrial environment”, SIMULATION, Vol. 59 No. 2, pp. 93-9.
Gass, S.I. (1991), “Military manpower planning models”, Computers in Operational Research, Vol.
18 No. 1, pp. 65-73.
Hambleton, R.S. (1982), “A manpower planning model for mobile repairmen”, Journal of 183
Operational Research Society, Vol. 33, pp. 621-7.
Leeson, G.W. (1982), “Manpower planning procedures for hierarchical manpower systems”,
European Journal of Operational Research, Vol. 10, pp. 260-9.
Mann, L. (1983), Maintenance Management, Lexington Books.
McClean, S. (1991), “Manpower planning models and their estimation”, European Journal of
Operational Research, Vol. 51, pp. 179-87.
Ntuen, C.A. (1989), “Forecasting maintenance crew size requirement based on periodic
maintenance records”, Production and Inventory Management Journal, 2nd Quarter,
pp. 41-3.
Singhal, K. (1992), “A non-iterative algorithm for the multi-product production and work force
planning problem”, Operations Research, Vol. 40 No. 3, pp. 621-5.
Stewart, B.D., Webster, D.B., Ahmad, S. and Matson, J.O. (1994), “Mathematical models for
developing a flexible workforce”, International Journal of Production Economics, Vol. 36, pp.
243-54.
Vassiliou, P.C.G. (1978), “A high order nonlinear Markovian model for promotion in manpower
system”, Journal of Royal Statistical Society, Vol. 141A No. 1, pp. 86-94.
Witherell, P.E. (1991), “Forestalling failures in the plants”, Chemical Engineering, Vol. 98 No. 10,
pp. 126-32.
Zanakis, S.H. and Maret, M.W. (1981), “A Markovian goal programming approach to aggregate
manpower planning”, Journal of Operational Research Society, Vol. 32, pp. 55-63.