Intelligent integrated maintenance of manufacturing systems 297

15
X

Intelligent integrated maintenance

of manufacturing systems
Roubi A. Zaied1, Kazem Abhary2 and Attia H. Gomaa1
1University of Benha (Egypt)
2University of South Australia (Australia)

1. Introduction
The management of maintenance activities extremely affects the useful life of the
equipments, product quality, direct costs of maintenance and consequently production
costs. Thus, a reliable maintenance system is critical to maintain an acceptable level of profit
and competition. Studies over 20 years have indicated that around Europe the direct cost of
maintenance is four to eight percent of total sales turnover (Muller 2007). The indirect cost of
maintenance is likely to be a similar amount. Thus, the potential savings from modern
maintenance would be massive. Neural Management Maintenance System (NMMS) is a
new technique yet to be further developed to reduce the involvement of analysts/engineers
in data processing and thus adding quality in decision-making process. NMMS is based on
Artificial Neural Networks (ANNs). The attractiveness of ANNs comes from their
remarkable information processing characteristics mainly to nonlinearity, high parallelism,
fault and noise tolerance, and learning and generalization capabilities. They have the ability
to extract patterns and detect trends that are too complex to be noticed by either humans or
other computer techniques. A trained neural network can be thought of as an "expert" in the
category of information it has been assigned to analyze. The NMMS would permanently
monitor the system and suggest the most appropriate actions and strategies. This chapter
explains a NMMS that integrates Corrective Maintenance (CM), adaptive Preventive
Maintenance (PM) and Condition Based Maintenance (CBM) with suitable maintenance
strategy addressed for each component/subsystem. The NMMS monitors the system and
suggests the most appropriate maintenance actions. The main characteristics of the system
includes; integration of expert opinion in a knowledge base, storing maintenance history
and tracking components, alarming predetermined maintenance activities, alerting for spare
parts and materials, updating schedules, considering limitation of resources, and measuring
the effectiveness of the maintenance system. The easiness and intelligence of the proposed
NMMS depends on keeping the maintenance data in MS-EXCEL spreadsheets and linking it
to MATLAB® which in turn updates the models and makes the decisions.

www.intechopen.com
298 Engineering the Future

2. Background
2.1 Maintenance Integration
Integration of maintenance into manufacturing organization is partitioned into "hard
integration" and "soft integration" variables. The "hard" issues deal with integration
supported by technology and computers. "Soft" integration, on the other hand, deals with
human and work organizational integration issues. The two integration variables are closely
related to the prevention variable, and are considered important enablers for effective
realization of preventive policies (Jonsson, 2000). Integration must facilitate the bi-
directional flow of data and information into the decision-making and planning process at
all levels. This reaches from business systems right down to sensor level.
Hard maintenance integration issues deal with CMMS (Computerised Maintenance
Management System) of the maintenance, repair and operating supplies store and
scheduling of maintenance work, condition monitoring technologies, built-in test
equipment, databases with reliability data on electronic and mechanical components, and
decision support. On the other hand, soft integration issues of maintenance deal with the
structure and the actors in the organization. New technology allows plants to have fewer
humans directly participating in the physical manufacturing processes.

To integrate maintenance policies and study their impact on complex production systems, a
powerful modelling tool is essential. One (or more) maintenance policies may be associated
with each machine. Thus, an elementary cell is defined as a set made up of a machine,
including associated maintenance policies, as well as its input/output stocks (Abazi and
Sassine, 2001).

Maintenance Integration is necessary to increase availability and reliability of

manufacturing systems to reduce unnecessary investment in maintenance without great
increasing of investment. The integration is achieved through combining optimal
maintenance types to have the benefits and to avoid the shortage of individual maintenance
types. Thus, the proper maintenance program must define different maintenance plans for
different machines.

The literature survey of the previous works indicates that a major interest of researchers has
been the maintenance optimization not causing a measurable response from the engineering
world due to two reasons:
1) Applicability: The works were mostly very theoretical, used difficult mathematics
impractical to apply and required data were not then generally available.
2) Accessibility: The papers were published in journals of applied mathematics and
operations research (OR), which most maintenance engineers do not read and few
would understand if they did.

Thus, maintenance methods applied at present should be combined together within a

comprehensive management maintenance system, which would permanently monitor the
system and suggest the most appropriate actions. Thus, the scheme proposed herein serves
this purpose, i. e. it combines maintenance integration and neural management maintenance
system.

www.intechopen.com
Intelligent integrated maintenance of manufacturing systems 299

2.2 Artificial Neural Networks (ANNs)

ANNs are applicable to multivariable systems; they naturally process many inputs and
produce many outputs. They are used as a black-box approach (no prior knowledge about a
system) and implemented on compact processors for space and power constrained
applications (Magali et al, 2003). ANNs; have ability to extract patterns and detect trends
that are too complex to be noticed by either humans or other computer techniques. A
trained neural network can be thought of as an "expert" that can then be used to provide
projections given new situations of interest and answer "what if" questions. ANNs learn by
example and cannot be programmed to perform a specific task. The examples must be
selected carefully otherwise the network might be functioning incorrectly. The disadvantage
is that because the network finds out how to solve the problem by itself, its operation can be
unpredictable.
The initial weights of an ANN play a significant role in the convergence of the training
method. Without a priori information about the final weights, it is a common practice to
initialize all weights randomly with small absolute values. In linear vector quantization and
derived techniques it is usually required to renormalize the weights at every training epoch.
A critical parameter is the speed of convergence, which is determined by the learning
coefficient. In general, it is desirable to have fast learning, but not so fast as to cause
instability of learning iterations. Starting with a large learning coefficient and reducing it as
the learning process precedes results in both fast learning and stable iterations.

2.3 Types and applications of ANNs

There are different types of ANNs according to how data is processed through the network.
Some of the more popular ANNs include multilayer Perceptron (MLP), learning vector
quantization, radial basis function (RBF), Hopfield and Kohonen networks. Some ANN are
classified as feed forward while others are recurrent (i.e. implement feedback) depending on
the direction of data flow. Another way of classifying ANNs is by their learning method, as
some ANN employ supervised training, while others are referred to as unsupervised or self-
organizing.

Magali et al. (2003) presented a comprehensive review of the industrial applications of

ANNs in 12 years prior to 2003. The study found that the approximate percentage of
network utilization was: MLP, 81.2%; Hopfield, 5.4%; Kohonen, 8.3%; and the others, 5.1%.

2.4 Modelling of maintenance systems

A manufacturing system of one failure mode can have one of two transition states either in
operation mode or in failure mode. Figure 1 shows the transition diagram of a system that
can either be in up (operating) or down (failed) state. A good maintenance management
system is to decrease the failure rate and increase the repair rate.

In practical maintenance application, the measured variables and/or checked attributes are
used to determine the next estimated time for replacement, repair or checking of the system.
The steps for maintenance decision making are shown in Figure 2. Figure 3 shows more
details of these steps and fine tools that can be used; it illustrates the system of data flow
from the input phase to the output phase.

www.intechopen.com
300 Engineering the Future

λ
System Failure System
Operating failed
Normally µC
Repair
Fig. 1. System operating states

Data gaining through Data Decision making

inspection or processing for
measurement maintenance

Fig. 2. Steps of maintenance decision

Fig. 3. Data flow from the input phase to the output phase

3. Design of the NMMS

3.1 Conceptual design of the NMMS
NMMS should model the manufacturing system in all its details, process the data and take
the maintenance decisions. The system is designed to integrate maintenance in the
manufacturing system and contribute to achieve high performance. It eases the bi-
directional flow of data and information into the decision-making and planning process at
all levels. Figure 4 shows the system of the maintenance optimization process in NMMS and
Figure 5 explain the learning mechanism of the ANN-based system.

Fig. 4. The maintenance optimization process

www.intechopen.com
Intelligent integrated maintenance of manufacturing systems 301

Fig. 5. The learning mechanism of the ANN-based system

An intelligent NMMS has to:

a. Integrate expert opinion into a knowledge base.
b. Store maintenance history and track components.
c. Alarm predetermined maintenance activities.
d. Alert for spare parts and materials, update schedules with occurrence of events, and
consider limitation of resources.
e. Measure the effectiveness of maintenance system.
A NMMS can be characterized by the following:
a. It integrates CM, PM and CBM (Condition Based Maintenance) with suitable
maintenance strategy addressed for each system.
b. CM is to be carried out directly when failure occurs without waiting.
c. PM timing will be automatically determined for subsystems and/or systems when
condition monitoring is not particularly appropriate.

3.2 Flowchart of NMMS

A NMMS is designed to run online and the plan update is triggered by signals of failures or
other events. However, update can be triggered by manual data input. The system runs in a
cyclic manner and the frequency is adapted according to the production rate and limitations
of the computer system capacity. Each cycle is executed in five phases; Initial-input phase,
Running-input phase, Evaluation phase, Outputs and decisions phase, and Feedback-input
phase. The NMMS flowchart is detailed in Figure 6.

3.3 Structural design of NMMS

The system design simulates the brain action and consists of sorted layers of networks. A
NMMS is a modular one, which consists of sorted layers of modules. Basically, each module
in the scheme will be represented by a neural network. The networks have interconnections
as well as external inputs for interfacing with maintenance administrator. The purpose is to
manipulate the maintenance strategy according to the goals of the manufacturing
corporation. Figure 7 shows the overall structure of a NNMS system.

www.intechopen.com
302 Engineering the Future

Fig. 6. Flowchart of the NMMS

www.intechopen.com
Intelligent integrated maintenance of manufacturing systems 303

Fig. 7. The NMMS framework

The main monitoring and maintenance module is on the top level of the architecture (the
brain). The inputs of this module come from the sublevels and the administrator. Each
component/subsystem is represented as a module. Maintenance status outputs shown on
the right side of Figure 7 are for control purposes and the outputs shown at the top of the
figure are to monitor the entire health and cumulated maintenance cost of the
manufacturing system. The system can operate automatically but it enables the
administrator with some interpose.

The most significant modules in this design are the subsystems modules (basic monitoring
units) in the ground level. Each subsystem has its own monitoring unit, each system
(equipment or machine) is assembled in the next level and each group of machines or
systems is clustered in the level preceding the main monitoring and maintenance unit. The
number of modules in each level depends on the size of each system, machine or group (e.g.
production line).

Since each machine consists of several functional parts, it is necessary to perform an analysis
for each functional part and then, based on the results, select the most favourable schedule
of maintenance activities. The design of each subsystem module differs according to the
assigned schedule for this subsystem. It can be CBM-system module, PM-system module or
CM-system module.

The desired modules are assigned on the basis of maintenance expert’s view according to
the applications and herein some rules can be employed:
a. Utilizing the whole history of available information for decision making.
b. Maintenance strategies are based on the failure rate characteristics, i.e. constant or
variable, failure impact and failure rate trend.
c. Equipment does not have to be checked repeatedly if it has not been used.

www.intechopen.com
304 Engineering the Future

d. There are certain pieces of equipment that require visual inspection when they have not
been run.
e. When the machine is down, the opportunity is used to repair or replace other items,
which are found to be faulty or in need of maintenance within a short period of time.
f. A maintenance action is considered if the deterioration level of the system/components
fall in specified trigger zones.
g. It is suitable to apply Run-to-failure for small, non-critical, low price component.

Figure 8 shows the details of a typical subsystem module. The module outputs to its linked
system (machine) module at the higher level.

Fig. 8. The details of a typical subsystem module

It was affirmed by Zeng (1997) that the manufacturing systems are moving towards high
automation, flexibility and integration. Intelligent sensors/actuators, intelligent distributed
I/O modules, and field buses will provide a significant opportunity for maintenance
systems to minimize system downtime and to further increase efficiency. That will create a
new generation computer integrated maintenance system which combines computer aided
fault detection, fault diagnosis, condition monitoring systems, and focus on necessary
highly efficient corrections. Unfortunately, with the present technology, not all failures are
detectable. Detectable failures develop rapidly or instantly and can be detected after the
failure has occurred. The monitoring system cannot identify non-detectable failures.

The best case for the CBM is the predictable failures, which result from gradual degradation
of the subsystem showing measurable changes with time. CBM individual module is
employed when it is applicable and cost effective for monitoring a subsystem. With the pace
of today’s technology, many sophisticated products have sensor-enabled monitoring units

www.intechopen.com
Intelligent integrated maintenance of manufacturing systems 305

embedded in the products. The monitoring units provide real-time condition information
for the system. Thus, CBM is widely implemented in practice (Gupta and Lawsirirat, 2003).
To catch this goal, the integrated system designed herein allocates a particular monitoring
unit associated with each subsystem. Firstly, all subsystems must be evaluated to determine
whether regular monitoring is cost-effective or not. Figure 8 shows a typical CBM module
enclosing five built-in units. Each unit has its external inputs from the monitored system
and there are interconnections among the five units. The module outputs to its linked
system (machine) module at the higher level.

4. Application Circumstance
In practical application of the proposed NMMS, maintenance data were obtained from an
Egyptian factory of florescent lamps. The proprietor is a leading company in the field of
white goods in Egypt. There are three production lines in the factory; each of them consists
of 16 machines forming 6 groups. It works 6 days weekly, 24 hours daily on three shift basis.
The production stops on Fridays and public holidays however, it works overtime on Fridays
and public holidays when more production is demanded. The company’s target for 2009
compared to the realized values is listed in Table 1 and Table 2 explains the formulas.

Criterion Target Realized

s
Total downtime as a percentage from working time 3.5% 5.96%
Total downtime ratio related to maintenance issues 2.25% 2.86%
Total downtime ratio related to production issues 1.0% 2.86%
Total downtime ratio related to supplies and other issues 0.25% 0.24%
Spareparts cost per lamp/cost of lamp production 2.3% 2.02%
Table 1. Maintenance targets set by factory management for 2009

Criterion Rule
Overall Unavailability ( Total downtime as a Total downtime/working time
percentage from working time) U = MTTR/(MTTR+MTTF)
Unavailability due to breakdown ( Total Repair downtime/working time
downtime ratio related to maintenance issues ) U = MTTR/(MTTR+MTTF)
Unavailability due to low production quality Restore downtime/working time
(Total downtime ratio related to production issues) U = MTTR/(MTTR+MTTF)
Unavailability due to shortage of supplies and
Restore downtime/working time
other issues ( Total downtime ratio related to
U = MTTR/(MTTR+MTTF)
supplies and other issues)
Spare parts Cost ratio ( Spareparts cost per Cost of spare parts per lamp/total
lamp/cost of lamp production) cost of lamp production
Table 2. Maintenance targets and their formulas

www.intechopen.com
306 Engineering the Future

4.1 Changes in maintenance plan and staff in the factory

The maintenance plan had been yielded several changes in the last years mainly because of
labour shortage. The short term PM was carried out daily for each production line from 2004
to 2006, then twice weekly in 2008 and now it is performed only once weekly. It was found
that this change in the PM plan reduced the maintenance cost. The staff distribution and
maintenance effort per single production line in the last two years in are shown in Table 3
and Table 4 respectively. Table 3 demonstrates, for the three lines, the monthly average
number of short term PMs executed in the last two years.

Maintenance Maintenance effort (Man-hour)

level Year 2008 Year 2009
Yearly 280 256
6 month 179 164
4 month 179 0
1.5 month 0 164
Monthly 179 0
2weeks 179 164
Weekly 0 42
3 Days 31 0
Table 3. Changes in maintenance effort in two years

Average No. of monthly short term PMs per

Months single production line
Year 2008 Year 2009
January 7.3 6.3
February 8.0 6.3
March 8.0 6.7
April 7.0 5.0
May 7.3 4.3
June 7.7 4.3
July 7.7 4.0
August 7.7 4.3
September 9.3 3.7
October 6.7 4.3
November 6.0 3.7
December 5.7 4.0
Table 4. Average number of monthly short term PMs per single production line in two years

www.intechopen.com
Intelligent integrated maintenance of manufacturing systems 307

4.2 Factors affecting production system failure rate

Apparently there are three dependant variables greatly affecting the production system
performance. The first one is the weather and the second one is the amount of regular PM
performed on the system. The third one is the age of line which represents the effect of
production line deterioration along its age and in the same time it simulates the effect of the
accumulated experience of the staff.

4.2.1 Effect of weather

For the purpose of studying the weather effect, the average quarterly downtime ratios
(Unavailability) were calculated for three successive years and the result is shown in Figure
9 from which it is clear that the downtime ratio is high in winter and summer (13% over the
average in winter) and low in spring and fall. After reviewing the factory staff it was
affirmed that the processing is highly sensitive to the weather as natural gas welding is used
in sealing and assembling of the lamps.

Fig. 9. Average downtimes ratios in three year quarters

4.2.2 Effect of the maintenance effort

The second factor affecting the failure rate of the production lines is the PM plan and staff
whose changes was presented in section 4.1. A full analysis was performed on the failure
data of the three lines to find the effect of the maintenance effort on the availability and cost.

For studying the effect of the frequency of short term PM on the availability and spareparts
cost, averages of monthly availabilities were calculated from data of three lines in two years
2008 and 2009. The change of PM plan caused average time saving in PM duration by
22.61%. Trendline were created to find the correlation between the numbers of short term
monthly performed PMs and the resulted availability, failures downtime and spareparts
cost ratio. Linear trendlines were drawn in Figure 10 with its mathematical equation. It is
clear from the charts that decreasing the frequency of short term PMs had positive effect on
the availability (increased by 0.5 %) and the cost of consumed spareparts (decreased by
10.71%) but it had negative effect on the total downtime caused by failures (failures
downtime increased by 10.1 %). However the availability increased because the downtime
caused by more failures is less than the time saved from the regular PM durations.

www.intechopen.com
308 Engineering the Future

(a) Effect of No. of short term PMs per line on the availability

(b) Effect of No. of short term PMs per line on the cost ratio

(c) Effect of No. of short term PMs per line on the downtime
Fig. 10. Effect of No. of short term PMs on the performance

www.intechopen.com
Intelligent integrated maintenance of manufacturing systems 309

4.3 Modelling of the production lines

The first step of the model building is to determine the model inputs and outputs. The
independent variables affecting the system’s performance are used as model inputs to study
the system’s behaviour under different conditions. However, an inverse model can be used
e.g. the number of PMs per month is an independent input but it can be used as an output
for the purpose of optimization.

The proposed inputs of the production line neural model are the line age and the
maintenance effort as illustrated in Figure 11. Line age is expressed in two individual inputs;
the number of years and the month. The number of years simulates the effect of line
deterioration along its age and it simulates the effect of the accumulated experience of the
maintenance staff as well. A number between 1 and 12 for the month represents the weather
effect. Five output variables are assigned to the neural model; Availability, Unavailability
due to breakdowns, MTBF, MTTR and Cost of spareparts as a percent from the production
cost.

Fig. 11. The neural model of a production line

4.4 Training of the production line-neural model

Vectors of real inputs and their associated real outputs are used to train the ANNs. The
purpose of the training process is to fix the coefficients or internal weights of ANNs. A
trained ANN model is capable of expecting the line behavior under different PM policies.
Furthermore, it can be used to expect the next adaptive PM timing. In practical applications,
the NMMS updates the Neuro-models of the lines using the most recent computed data
through MATLAB-EXCEL link. The average/total monthly values of the five model outputs
were computed from failures data of lines from Jan 2008 to Dec 2009. The purpose of this
accumulation of data is to build line models using refined data of relatively complete
periods. The extracted data were used as training data for the neural models of each of the
three production lines. Table 5 shows the training data for the first line.

www.intechopen.com
310 Engineering the Future

May-08

Aug-08

Nov-08
Mar-08

Dec-08
Apr-08

Sep-08
Feb-08

Oct-08
Jun-08
Line 1 - Maintenance

Jan-08

Jul-08
effort (man.hr)-2008

Year 5 5 5 5 5 5 5 5 5 5 5 5
Month 1 2 3 4 5 6 7 8 9 10 11 12
Input Vectors

Yearly 280 280 280 280 280 280 280 280 280 280 280 280
6 month 179 179 179 179 179 179 179 179 179 179 179 179
1.5 month 0 0 0 0 0 0 0 0 179 179 179 179
0.5 month 179 179 179 179 179 179 179 179 179 179 179 179
Weekly (per month) 0 0 0 0 0 0 0 0 0 0 0 0
0.5 week (per month) 191 182 219 235 226 266 197 141 205 167 141 153
Availability* 91 94 95 92 96 95 93 97 94 97 94 95
Output Vectors

Breakdown maintenance
6.5 3.7 2.1 4.9 2 2.7 4.9 1.3 4.2 1.3 4.2 3.2
downtime ratio %

MTBF (day) 1.6 1.4 1 1.4 0.8 0.9 1.8 0.7 1.9 0.8 2.2 1.1

MTTR (min) 94 54 30 71 29 39 70 19 61 18 60 47
Cost of spareparts/ Total
1.8 2 2.7 2.7 1.8 1.1 2.3 2.5 1.6 1.8 1.2 1.5
production cost %
May-09

Aug-09

Nov-09
Mar-09

Dec-09
Apr-09

Sep-09
Feb-09

Oct-09
Jun-09

Line 1 - Maintenance
Jan-09

Jul-09

effort (man.hr)-2009

Year 6 6 6 6 6 6 6 6 6 6 6 6
Month 1 2 3 4 5 6 7 8 9 10 11 12
Input Vectors

Yearly 256 256 256 256 256 256 256 256 256 256 256 256
6 month 164 164 164 164 164 164 164 164 164 164 164 164
1.5 month 164 164 164 164 164 164 164 164 164 164 164 164
0.5 month 164 164 164 164 164 164 164 164 164 164 164 164
Weekly (per month) 0 0 0 0 0 0 0 157 142 124 128 126
0.5 week (per month) 141 209 141 159 153 211 133 0 0 0 0 0
Availability* 90.3 94.9 95.8 96.1 92.9 93.7 94.2 94.6 96.4 95.2 94.4 94.3
Output Vectors

Breakdown maintenance
8.07 2.7 2.74 2.11 5.37 4.06 4.42 3.63 1.94 3.28 3.86 3.71
downtime ratio %
MTBF (day) 2.79 1.35 1.48 0.96 1.95 1.73 1.92 1.61 0.92 1.58 1.78 2.24
MTTR (min) 116 38.9 39.5 30.4 77.3 58.5 63.6 52.2 27.9 47.3 55.5 53.4
Cost of spareparts/ Total
2.41 2.85 3.17 1.22 2.64 0.89 3.72 1.15 1.93 3.48 2.29 2.45
production cost %
Table 5. Training data for line 1

www.intechopen.com
Intelligent integrated maintenance of manufacturing systems 311

Neural fitting tool of Neural Network toolbox of MATLAB® was used successfully to train
the ANN model of the production lines. For each production line, 22 training vector pairs
are used besides one vector pair for validation and one vector pair for testing. Each input
vector is of size of 8x1 and each output vector is of size 5x1. The default value of number of
hidden neurons by the toolbox is 20 neurons but, that best number for this case was found
by trial to be 25 hidden neurons. The mean square error (MSE) between the actual and the
target outputs is used as an accuracy measure of the trained ANN. For line 1 e.g., the MSE
reached 2.7x10-21 for the first line and the training process terminated after 111 epochs.
Figures 12 and 13 show the results of the training process.

Fig. 12. Training error convergence of ANN model for line 1

4.5 The Whole Production System Model

In the same analogy of modelling the individual lines, an ANN was trained to model the
whole production system. Data used for training are based on the averages and summation
of the variables of the three individual lines.

4.6 Performance evaluation of the production system

Six performance indicators are used to evaluate the health of the production lines, the
maintenance policy and the maintenance staff efficiency. Table 6 summarizes the computed
performance indicators for the three lines in the two years period.

www.intechopen.com
312 Engineering the Future

Fig. 13. Regression plot of ANN model training for line 1

Line 2 outperforms the other two lines generally regarding the availability and cost of
spareparts. Analysis of the concluded data revealed different behaviour of each production
line. The performance of the three lines was compared in terms of some criteria. The data
used for this comparison are those of Table 6. The availability and MTBF indicator shows a
general increasing trend for the three lines and total downtime shows a general decreasing
trend; this is a good indicator of the policy and staff efficiency. Regarding the cost data, the
effect of the international financial crises in 2008 must be taken into consideration when
evaluating the cost trend along the time. Table 7 summarizes the computed global
performance variables. Figure 14 show the trend of the whole system performance.
Generally it reveals a good performance in terms of cost and MTTR but other criteria yields
oscillating performance. The results of the application of the indicators on the three lines are
illustrated in Figure 15.

www.intechopen.com
Intelligent integrated maintenance of manufacturing systems 313

Total
Spareparts MTTR MTBF MTTM
Time Period Downtime Availability
cost % (min) (hr) (hour)
(hr)
First half year-2008 2.0 53 28 35 2 93.8
Second half year-2008 1.8 46 33.9 32.1 2.0 95.2
Line1

First half year-2009 2.2 60 41.1 35.6 2.4 93.9

Second half year-2009 2.5 50 40.2 29.3 2.3 94.8
First half year-2008 3.5 37 17 27 2 95.3
Second half year-2008 1.6 39 25.9 30.9 2.1 95.0
Line2

First half year-2009 1.6 53 37.1 32.2 2.0 94.3

Second half year-2009 1.7 41 30.0 25.4 2.2 95.7
First half year-2008 1.9 45 30 31 1.8 94.4
Second half year-2008 2.1 40 30.9 30.7 2.1 94.9
Line3

First half year-2009 1.5 45 28.4 29.7 2.6 95.0

Second half year-2009 2.1 42 33.4 27.9 2.5 95.6
Table 6. Summary of the computed lines performance variables

Average Average Average Total Average Average

Time Period spareparts MTTR MTBF Downtime MTTM Availabilit
cost ratio (min) (hr) (hour) (hour) y
First half year-2008 2.5 45 25 31 4 94.5
Second half year-2008 1.8 42 30.2 31.2 3.9 95.0
First half year-2009 1.8 53 35.5 32.5 3.8 94.4
Second half year-2009 2.1 44 34.5 27.5 3.1 95.4
Table 7. The computed whole system performance variables

Fig. 14. Performance Evaluation of the whole system

www.intechopen.com
314 Engineering the Future

Fig. 15. A comparison of the three lines performance in 2 years

www.intechopen.com
Intelligent integrated maintenance of manufacturing systems 315

5. Conclusions
This chapter proposes a framework to design an NMMS. The structure of the system is
designed to simulate the brain action. A comprehensive design of an NMMS is achieved:
1. For application purposes, data were obtained from a fluorescent lamps factory in Egypt
to simulate of the proposed NMMS. The data yielded itself to analysis and simulation.
2. Failure data analysis showed that lowering of the frequency of short term PMs had a
positive effect on the availability (increased by 0.5 %) and the cost of consumed
spareparts (decreased by 10.7%). But the total downtime caused by failures increased by
10.1 %.
3. The availability increased because the downtime caused by more failures is less than the
time saved from the regular PM durations.
4. The maintenance staff experience helped in decreasing the MTTR. Thus compensated the
resulted increased downtime when PM assigned time decreased.
5. It was found that few numbers of repeated faults cause the major downtime; this should
attract the management attention to find the roots of these faults and find the suitable
solutions. The solutions might be a modification of the machine design and/or the
maintenance policy of these subsystems. The factory management already modified
some designs and the maintenance policy of some subsystems.
6. An individual ANN was assigned to each production line model as they showed
different behaviors.
7. The extracted data were used as training data for the neural models of each of the three
production lines. The ANN could realize the production system maintenance model and
it is capable of expecting the line behavior under different PM policies.
8. The MSE of ANN training reached 2.7x10-21 for the first line.
9. The availability and MTBF indicator showed general increasing trend for the three lines
and total downtime showed general decreasing trend, i.e. a good indicator of the policy
and staff efficiency.

6. Acknowledgement
We are grateful and thankful to the administration of the Egyptian factory of fluorescent
lamps for kind help, encouragement and providing us useful data for the case study
especially the maintenance staff.

7. References
Aparna Gupta and Chaipal Lawsirirat (2006). Strategically optimum maintenance of
monitoring-enabled multi-component systems using continuous-time jump
deterioration models. Journal of Quality in Maintenance Engineering, Vol. 12, No. 3,
pp. 306-329.
Basheer I. A., Hajmeer M. (2000). Artificial neural networks: fundamentals, computing,
design, and application, Journal of Microbiological Methods, Vol. 43, pp 3–31.
Garg Amik and Deshmukh S.G. (2006). Maintenance management: literature review and
directions. Journal of Quality in Maintenance Engineering, Vol. 12, No. 3, pp. 205-238.

www.intechopen.com
316 Engineering the Future

Magali R. G. Meireles, Paulo E. M. Almeida, and Marcelo Godoy Simões. (2003). A

Comprehensive Review for Industrial Applicability of Artificial neural networks.
IEEE transactions on industrial electronics, Vol. 50, No. 3, June 2003, pp. 585-601.
Mobley R Keith. (2004). Maintenance Fundamentals. Second Edition Elsevier Inc., ISBN: 978-
0-7506-7798-1.
Muller A, Suhner M and Iung B. (2007). Maintenance alternative integration to prognosis
process engineering. Journal Of Quality in Maintenance Engineering, Vol. 13, No. 2,
pp. 198-211.
Patrik Jonsson. (2000). Towards a holistic understanding of disruptions in Operations
Management. Journal of Operations Management, Vol. 18, pp. 701–718.
Polimac V, Polimac J, 2001, “Assessment of present maintenance practices and future
trends”, Proceedings Annual Reliability and Maintainability Symposium IEEE, pp 891-
894.
Polimac V, Polimac J. (2002). Conceptual development of neural management maintenance.
Power Systems Management and control, IEE 2002, No. 488, pp. 457-460.
Shuo Wei Zeng. (1997). Discussion on maintenance strategy, policy and corresponding
maintenance systems in manufacturing, Reliability Engineering and system Safety,
Vol. 55, pp. 15 l- 162.
Zaied Roubi and Abhary Kazem (2009). A Design of an Intelligent Maintenance Integrated
System into Manufacturing Systems. Proceedings of International Conference on
Industrial Technology, pp. 1296-1301, IEEE, 978-1-4244-3507-4, 2009, GIPPSLAND,
Australia.
Zaied Roubi A, Nawara Gamal Abdel-Salam Mohammad and Abhary Kazem. (2010). A
Neural Management Maintenance System for Manufacturing Systems. Proceedings
of 19th international conference for the international association of management of
technology (IAMOT2010), pp. 1-18, 1569263581, March 8-11, Cairo, Egypt.
Zineb Simeu-Abazi and Chadi Sassine (2001). Maintenance Integration in Manufacturing
Systems: From the Modeling Tool to Evaluation”, the International Journal of Flexible
Manufacturing Systems, pp. 267–285.

www.intechopen.com
 Engineering the Future
Edited by Laszlo Dudas

ISBN 978-953-307-210-4
Hard cover, 414 pages
Publisher Sciyo
Published online 02, November, 2010
Published in print edition November, 2010

This book pilots the reader into the future. The first three chapters introduce new materials and material
processing methods. Then five chapters present innovative new design directions and solutions. The main
section of the book contains ten chapters organized around problems and methods of manufacturing and
technology, from cutting process optimisation through maintenance and control to the Digital Factory. The last
two chapters deal with information and energy, as the foundations of a prospering economy.

How to reference
In order to correctly reference this scholarly work, feel free to copy and paste the following:

Roubi A. Zaied, Kazem Abhary and Attia H. Gomaa (2010). Intelligent Integrated Maintenance of
Manufacturing Systems, Engineering the Future, Laszlo Dudas (Ed.), ISBN: 978-953-307-210-4, InTech,
Available from: http://www.intechopen.com/books/engineering-the-future/intelligent-integrated-maintenance-of-
manufacturing-systems

InTech Europe InTech China

University Campus STeP Ri Unit 405, Office Block, Hotel Equatorial Shanghai
Slavka Krautzeka 83/A No.65, Yan An Road (West), Shanghai, 200040, China
51000 Rijeka, Croatia
Phone: +385 (51) 770 447 Phone: +86-21-62489820
Fax: +385 (51) 686 166 Fax: +86-21-62489821
www.intechopen.com