1Sakarya
ev
iew
Safiye Turgay1, Samet Koç2
ed
Big Data-Driven Preventive Maintenance Decision Support
with Monte Carlo Simulation
University/Department of Industrial Engineering, Sakarya, Turkey
1safiyeturgay2000@yahoo.com
2smtkoc58@gmail.com
KEYWORDS- Stochastic Modelling, Preventive Maintenance Planning, Monte Carlo
rr
Simulation, Big Data
ABSTRACT
Production activities without interruption, growing the products on time, and producing the desired
ee
quality are essential for the sustainability of the enterprises. Today, with the developing technology,
sustainability in the production process and analysis of production data has to integrate and the
emerge of the big data and Internet of Things (IoT) concept. Continuously obtaining and analyzing
tp
the production data related to maintenance planning and revealing the rules to be used in the
enterprise to analyze these data also support maintenance planning activities. They will make a
significant contribution to the production process. However, most operating and maintenance costs
no
are wasted due to incorrect, unsystematic, and unplanned maintenance methods. Maintenance costs
are known to account for between 15% and 60% of the total operating costs.
This study aims to standardize the types of breakdowns and faults to estimate them during the
int
operation of the existing system by considering the frequency values of the breakdowns and faults
obtained from dynamic structures from big data and encountered. It is one of the most important
objectives to carry out the production process without interruption without increasing its
pr
performance with preventive measures and without deteriorating the production quality with a
Pr
e
decision support system.
1
INTRODUCTION
Preventing and predicting machine stoppages, disruptions, and breakdowns in production is very
important in terms of sustainability of production and efficient and productive operation of the
production process. Thanks to the sensors in the machines involved in the production, a continuous
This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=4308323
flow of information during production and the minor changes in the system can instantly be noticed
ed
and analyzed, and relevant decision making in maintenance situations. This information is also
possibly used with the concepts of big data and IoT[1,2,5]. The occurrence of breakdowns and
faults are a necessity to maintenance activities which are causing a temporal loss in the production
ev
iew
system for the reason of results in additional costs due to disruption. This study focuses on the
negative impact on the enterprise's competitiveness and improves the production system efficiency.
It is necessary to implement the appropriate periodic maintenance process and develop maintenance
policies to prevent such damages. We perform proper periodical maintenance, determining which
machine, period, and frequency of maintenance or proper maintenance policy is among the most
critical goals in the suggested framework.
The Monte Carlo method uses random variables and then moves them to find the probability of
rr
distribution. The Monte Carlo sample uses probability simulation models that are the stochastic
model of an actual situation that is preparing sampling experiments for the model. Such simulations
are used to study system outputs with a large number of interrelated variables in the stochastic
ee
structure.
The remainder of the paper is structured as follows: the next section reviews some concerning
tp
notions of maintenance planning. The Monte Carlo simulation method and big data-driven
maintenance model are discussed in Section 3. Section 4 includes the actual case application in the
2
no
radiator factory. Then the last section covers the discussion of results and some concluding remarks.
RELATED WORK
Since the invention of machinery and tools, maintenance and repair activities have to be
int
developed at the same rate as technological developments, increasing needs, fast access of
consumers, and sensitive production planning. Development of technology, gathering of sensors
and information, and IoT simultaneously brought the big data event [3,4,9]. The decision support
pr
system proposed in this study also operates in a support system that enables big data processing
from the sensors. It produces decisions for situations from historical data with the Monte Carlo
technique.
Pr
e
Big data structures and IoT structures are considered a whole. This type of work has been
adopted to feed this structure in a dynamic environment and support a continuous decision support
structure—machine stops and deteriorations terrible affect product quality and sustainability. The
Monte Carlo simulation method is mainly preferred by accident risk situations in the literature [6].
Integer programming and linear programming approaches are not suitable for big data status since
the system cannot perform the analysis process continuously, as it considers the only find
This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=4308323
momentary values. The Monte Carlo simulation structure also ensures the system operates with a
ed
continuous structure and information flow within the decision support system. Mendes and others
preferred the Monte Carlo simulation technique to maintain and repair cold-by stand systems [7].
The technique can also be used in the process of estimating unknown situations. For example,
ev
iew
Gayrav and Sharma gave an example of the analysis of the deteriorating conditions of metals
depending on environmental factors and material structure over time [8].
Inputs
Manufacturing
Machines Sensor Big Data
rr
Vibration
Acoustic signals
Power consumption
Images
Maintenance
Demand
ee
Manufacturing
Capacity
Product
Maintenance
tp
Fig. 1. Big Data, Maintenance and Manufacturing Relation
The suggested framework provides the planning and scheduling, the ability to monitor the desired
no
levels of system performance. It is one of the fastest and most effective investments an organization
can make to increase efficiency and availability, maintenance planning activities, especially in
production planning, in balance or integration with other possible sub-activities.
Maintenance Concept
int
2.1
All components in a production system are subject to wear; faults and breakdowns may occur when
performing the production tasks. The rate at which wear occurs and how often errors may force
pr
employees, equipment, and perhaps the entire system to be empty depends on the design and
operating conditions of the process. Poor maintenance can lead to incorrect output, downtime, and
increases in production costs due to repairs [10, 11].
Pr
e
Today, competitiveness is compulsory for companies around the world. New technologies and the
continuous emergence of new markets create undeniable global competition. Then there are forced
to complement the production plans quickly with resulting from these competition conditions that
cause the short life of the used machines and put the system under stress. It is necessary to consider
the cost and capacity of the equipment used in production and the technological trends. Proper
This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=4308323
maintenance helps reduce the life cycle cost and allows the correction or reduction of internal
ed
system problems that provides a comprehensive and easily accessible technical solution with an
appropriate maintenance policy. The suggested approach implies preventive maintenance. Also, the
following sub-section gives the information briefly about maintenance activities and effects.
Classification of Maintenance Activities
Maintenance activities are classified as below:
ev
iew
2.2
2.2.2.1 Preventive (Periodic) Maintenance Activities: It significantly reduces capacity
records and breakdowns resulting from unexpected failures. Before any failure occurs, the
determined times are analyzed, and planning is done. It can be determined periodically and
according to the working hours or working principles of the equipment.
Fault measurement and analysis are essential to decide whether to choose preventive
rr
maintenance and incidental maintenance alternatives to the system. The failure rate and the average
time between failures should be determined according to the appropriate.distribution type. The
ee
causes of breakdowns and faults may depend on each other and another component that can cause
different faults. The conditional probabilities are vital for the production and maintenance planning
to act as a result of the union of data by establishing a correct distribution model[12, 18].
tp
2.2.2.2 Predictive Maintenance: The predictive maintenance type, the values of some
variables representing the operational capability of the machine or equipment, the status and
capacity of the process should be defined in detail. Physical variables such as temperature,
no
vibration, power consumption can cause functionally are essential indicators for the implementation
of the maintenance. This type of maintenance factor is used to analyze the system for subsequent
maintenance plan decisions[24, 26].
int
2.2.3.3 Preventive (Corrective) Care: It covers the protection of the plant or equipment by
providing satisfactory conditions during the early stage breakdowns and controls. Process
reliability, affordability, and standard compliance factors occur form of preventive maintenance
pr
efforts. It is an essential factor for the general care organization's effectiveness to prevent damage to
machinery and equipment such as design changes, vibration prevention, and isolation of heat-
Pr
e
exposed areas, pre-regulation of equipment that may cause breakdowns and faults. It can be applied
immediately after any breakdowns and faults can be postponed according to the maintenance policy
that contains specific team rules [10, 13, 15, 17].
This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=4308323
3
THE MONTE CARLO SIMULATION
ed
The Monte Carlo method estimates a probability event by generating multiple scenarios on its
statistical parameters. In other words, the Monte Carlo simulation models the random behavior of
the event it is working. An ideal situation in which Monte Carlo applications are used is where the
ev
iew
size of the problem to be solved is one or more than two. Size refers to the number of state
variables. Because in multidimensional situations, it is either very difficult or impossible to obtain
closed-form formulas, and the application of numerical methods may be equally difficult or
impossible. In this section, simulation and random number generation could be examined [14].
X is the expected value of the random variable θ = E (X) with a probability distribution of 𝑃𝑥
estimated value of the sample mean of this variable;
𝑛
𝜃𝑛(𝑋) = 𝐸(𝑋) = 𝑛∑𝑖 = 1𝑋(𝑖)
1
rr
can be predicted with:
𝑋 = (𝑋(1),𝑋(2)…….𝑋(𝑛))
(1)
(2)
ee
It is a random vector consisting of independent and same dispersed components of the random
variable, which depends on the same probability distribution (independent and identically
tp
distributed).
Thus, 𝜃𝑛(𝑋)can be taken an approximate estimator of the expected value (θ) or, more precisely, as
the expected value. The reason for this is that when a large number of these samples are taken, the
no
average of these samples will approach θ and the variance will be Var(𝜃) = 𝑉(𝑋)/𝑛. Another one is
the central limit theorem. In other words, even if the distribution of a mass is not normally
distributed, the mean and variance of the sample averages the standard distribution, provided that
int
the sample volume is large (n³ 30). (n 30).
According to this theorem, the normalized sample means converge to the standard normal
distribution;
pr
n→∞
while;
𝜃𝑛(𝑋) ‒ 𝜃
𝜎𝑛/ 𝑛
Pr
e
𝑍=
→𝑑𝑁~(0,1).
(3)
Standardization is a non-deviant estimator of actual standard deviation;
1
𝑛
(𝑋𝑖 ‒ 𝜃𝑛(𝑋))2.
𝜎𝑛(𝑋) = 𝑛 ‒ 1∑
𝑖=1
(4)
2
The estimation error 𝜃𝑛(𝑋) ‒ 𝜃 is distributed by approximately 𝑁(0,𝜎𝑛/𝑛), allowing us to obtain the
estimated value based on confidence intervals.
This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=4308323
ed
The standard error 𝜎𝑛/ 𝑛 is one of the most important elements of the Monte Carlo method. If we
want to reduce the error value by half, we need to increase the number of samples by four times, or
multiply the number of samples by a factor of 100.
ev
iew
With the simplest Algorithm;
1. Select large a “n” number.
2. 𝑥 = (𝑥1,𝑥2…….𝑥𝑛)is the sample set.
3. 𝜃𝑛(𝑋)estimate the sample mean as E E(X)=𝜃𝑛(𝑋).
3.2 Simulation Model
rr
Random number generation is a result of deterministic processes and algorithms. Uniform
distribution in the unit range can be produced in the appropriate numbers. These numbers can then
ee
be converted to random numbers.
Let X ~𝐹𝑥 distribution uses the number of n and
G_𝐺𝑥:[0,1]𝑛→𝑅 function represents the compound random number.
tp
The distribution of 𝐺𝑥(𝑈1,𝑈2……𝑈𝑛) can be obtained as 𝐹𝑥. The 𝐺𝑥function can be assigned 𝑃𝑥 to
this function;
Cumulative probability distribution function is 𝐹𝑥(𝑥). If 𝐹𝑥(𝑥) continuous, the probability density
𝑓𝑥(𝑥) =
∂𝐹𝑥(𝑥)
∂𝑥
no
function is
(5)
If F_x is discrete, the discrete distribution function is
(6)
int
𝑃𝑥(𝑥) = 𝐹𝑥(𝑥) ‒ 𝐹𝑥(𝑥 ‒ )
A risk function,
𝑓𝑥(𝑥)
pr
ℎ𝑥(𝑥) = 1 ‒ 𝐹𝑥(𝑥)
One of the essential elements of maintenance management is collecting, analyzing, and using data
Pr
e
on equipment failures and repairs [16].
This study estimated the failures that can help determine the time of breakdowns and faults in the
next term using the historic and now failure data. The results were obtained from the data analyzed
by Macro Programming (VBA) in Excel with the Monte Carlo model given in Table 1. The
suggested simulation model is schematized in Fig. 2.
This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=4308323
ed
In the application, the linear programming model has established the suitability of the data analyzed
according to the road map to the normal distribution and the excess of nonparametric data. In the
simulation stage of the problem, the random model was chosen, and the average values were
ev
iew
generated by repeating the model n times. Breakdowns and fault assignments were made to the
sections associated with the resulting probability values, and the performance indicators were
int
no
tp
ee
rr
determined by comparing the results in an excel environment.
Pr
e
pr
Fig. 2. Suggested Simulation Model
This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=4308323
ev
iew
For j =1 to 5
if j=1 then
blm=” WELDING”-Elseif j=2 then
blm=”PRESS”
…
…
Elseif j=5 then
blm=”MACHINE”
End If
If y(1) > y(2) Then
bsf = (y(1) - y(2)) * 24
Else
bsf = (y(2) - y(1)) * 24
End If
'
maintenance
times
difference
If bsf <> 0 And bsf < Big Data
Then
obs = obs + bsf
'
maintenance
time
calculated cumulatively
End If
set (b) = a
End If
Next i
rr
Sub Analysis()
Dim set(Big Data) As Date
Dim a As Date
Dim y(2) As Date
Dim obs As Double
Dim obo As Integer
Dim bsf As Double
Dim blm As String
Sheets("DATA").Select
son=Range("E665536").End(xlUp).Row
ed
Table1. Monte Carlo Simulation Program Code in Macro Programming in Excel (VBA)
Pr
e
pr
int
no
tp
ee
For x = 1 To b
If set (x + 1) > set (x) Then
avgdiff = (set (x + 1) - set (x))
Else
avgdiff = (set (x) - set (x + 1))
For i = 7 To son
End If
If UCase(Cells(i, 8)) = blm Then
If avgdiff < 100 Then
a = Cells(i, 12) + Cells(i, 13).
'Different constraint values
'maintenance date-time combined
were used for each section.
y(1) = Cells(i,19) + Cells(i, 20)
(Purpose: closeness to reality
' combined 'maintenance start date ))
c = c + (avgdiff / b)
and time data
y(2)
=
Cells(i,
17).Value
+ End If
Next x
Cells(i,18).Value
‘combined maintenance end date
Sheets("parameters").Select
time data
For s=1 to 4
b=b+1
's used counter as (for set and Cells(s, j) = c
Cells(s, j) = 1 / c
frequency)
Cells(s, j) = (obs / b)
Cells(s, j) = 1 / (obs / b)
Next s
Next j
End Sub
Periodic (preventive) maintenance policy was determined by making maintenance
assignments to minimize downtime for the estimated faults. Then, conformity analysis was
performed in SPSS program for the model whose maintenance policy was determined, and the
results were compared with the results of the ANOVA test.
This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=4308323
ed
Notations
𝑁𝑖: Sample Size
ev
iew
C: Number of Faults
𝐴𝐴𝑆𝑘: Time between failures for each i component “k-1”through”k”))
𝐴𝐴𝑂𝑆İ: Mean time between failures of component i
𝜆𝑖: failure rate of component i
𝑇𝐸𝐵𝑆𝑖: The component k of i. Maintenance time for repair efficiency (k = 1… ....... N_i)
𝑂𝐵𝑆𝑖 : average maintenance time for component i
𝜇𝑖
: Maintenance rate for component i
rr
𝐴𝐴𝑂𝑆: Average time between failures of all components
𝜆: Average failure rate of all components
OBS: Average maintenance time for all components
ee
μ: Average maintenance rate of all components
M: Number of predictive maintenance activities (sample size)
tp
KBAS: Time between predictive maintenance activities (k = 1 ………… M)
KBAOS: Average time between predictive maintenance activities
OBS: Average preventive maintenance time
π: The average rate of predictive maintenance activities
no
〖CSF〗 _CM: Compound average time between all maintenance activities
ψ: Average rate of all maintenance activities
ODS: Average downtime
int
Functions related to these variables are shown below:
the average time and rate of failure of component i;
∑
𝐴𝐴𝑆𝑡
1
𝜆𝑖 =
𝑁𝑖
𝐴𝐴𝑂𝑆İ
(8)
𝑘=1
pr
𝐴𝐴𝑂𝑆𝑖 =
𝑁𝑖
the average maintenance time and proportion of component i;
∑
Pr
e
𝑂𝐵𝑆𝑖 =
𝑁𝑖
𝑇𝐸𝐵𝑆𝑖
𝑘=1
𝑁𝑖
𝜇𝑖 =
1
𝑂𝐵𝑆𝑖
(9)
Mean duration and rate of failure of all components;
This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=4308323
𝐶
1
𝐴𝐴𝑂𝑆 = 𝜆𝜆 = ∑𝑖 = 1𝜆𝑖
𝜆.𝑂𝐵𝑆 + 𝜋.𝑂𝐾𝐵𝑆
𝜆+𝜋
ed
𝑂𝐷𝑆 =
(10)
(11)
𝑂𝐵𝑆 =
𝐶
∑
𝑂𝐵𝑆𝑖
𝑖=1
𝜇=
𝐶
1
𝑂𝐵𝑆
ev
iew
Average maintenance time and rate of all components;
(12)
The average time and rate between predictive maintenance activities;
𝐾𝐵𝐴𝑂𝑆 =
𝑀
∑ 𝐾𝐵𝐴𝑆
𝑀
𝑘=1
𝜋=
1
𝐾𝐵𝐴𝑂𝑆
(13)
rr
Average duration and average rate between all maintenance activities;
1
𝐵𝑂𝑆𝐶𝑀 = (1/𝐴𝐴𝑂𝑆) + (1/𝐾𝐵𝐴𝑂𝑆)
1
(15)
ee
= 𝐵𝑂𝑆𝐶𝑀
(14)
The structure of the big data-driven maintenance planning-based decision support system proposed
in this study is given in Fig. 3. The sensors in this structure get the information from the machines,
tp
and it helps to decide which machine can be broken or break down in the future depending on the
type of information received. Predictive maintenance is activated here according to the machine
type, and maintenance planning is designed by considering the possible deterioration status of the
Pr
e
pr
int
no
machine.
This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=4308323
ed
ev
iew
rr
ee
4
tp
Fig. 3. Big data driven maintenance decision support model
APPLICATION
no
In this section, collecting the necessary data is related to big data IoT & maintenance planning from
sensors. We mentioned that the presence of automation systems that increase machine and product
variety, continuous operation or failure conditions, and complexity in fast production enterprises
increases the number and rate of failures by adding to the existential degradation tendency of the
int
equipment [22].
In this study, the data set has been kept in the 15 months of an organization with a production
pr
capacity of 2,200,000 meters/metal sheet, which has approximately 3000 employees operating in a
3-shift system and operating in Turkey (Table 2) examined. The system was stochastically handled.
Pr
e
The Monte Carlo simulation technique was preferred for modeling due to the importance of the
necessities to know the failures in the system, the downtimes caused by them, and the probability
situations.
This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=4308323
First of all, the problem was analyzed with VBA in Excel, and the performance results were
ed
revealed. In the next step, random numbers were generated with the Monte Carlo model, and it was
estimated how the system would behave in the next year.
ev
iew
Unplanned maintenance and predictive maintenance activities were calculated separately to obtain a
combined average time in the analyzed data set. Since the operating conditions of each section of
the enterprise consist of 5 different sections, which are different from each other, the faults
occurring are calculated separately by indexing and subsequently adding combined average time.
Various variables and formulations for failure and maintenance calculations are introduced
below[20, 21, 23].
Table 2. Section Indexes
WELDING
1
MACHINE
2
PACKAGING
3
4
PAINT
5
ee
İ
PRESS
rr
INDEX
The planning process includes the tasks that are required to perform a job are determined before
starting a job. The work is monitored, controlled, analyzed, and reported, as well as the planned
tp
work stages and times of execution. Maintenance needs three resources: labor, materials, and
equipment. Both the most valuable and the most challenging resources to control are labor
resources. Effective use of scheduling depends on labor productivity [25].
no
It was observed that machine and equipment failures in different sections (welding, press, machine,
paint, packaging) caused too much downtime, as observed in Table 3. Fig. 4, breakdowns and faults
increased during increasing periods, and frequent breakdowns were observed. Although the number
int
of failures observed in the ongoing month after the periods in which the predictive maintenance
policy is implemented, the direction of the failure trend did not change, and these maintenance
Pr
e
pr
practices were deemed insufficient.
This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=4308323
Montly Breakdown-Maintenance Distribution
ed
60
50
30
20
10
0
0
2
4
6
8
Breakdown
10
12
Maintenance
ev
iew
40
14
16
Fig.4. Monthly Breakdown and Maintenance Distribution
PROBLEM
JOB
DIAGNOSIS DATE
PRODUCTION
AND HOUR
HAULT
DATE
BREAKDOWN
WELDING
YES
BREAKDOWN
WELDING
YES
13.05.2013
10.05.2013
PRESS
YES
14.05.2013
BREAKDOWN
WELDING
YES
16.05.2013
BREAKDOWN
WELDING
NO
WELDING
YES
WELDING
YES
BREAKDOWN
WELDING
YES
MAINTENANCE
WELDING
BREAKDOWN
17.05.2013
ENDING OF
AND HOUR
DATE
HOUR
IS
PROBLEM
HOUR
SOLVING?
DATE
HOUR
17:00
10.05.2013
17:00
14.05.2013
13:00
NO
13:00
13.05.2013
13:00
13.05.2013
13:20
YES
14:00
14.05.2013
14:00
14.05.2013
18:00
YES
11:30
16.05.2013
11:30
16.05.2013
11:50
YES
17.05.2013
11:05
17.05.2013
18.05.2013
06:45
18.05.2013
06:45
18.05.2013
18.05.2013
07:00
18.05.2013
07:00
11:05
20.05.2013
18.05.2013
11:35
YES
08:45
YES
07:30
YES
15:35
22.05.2013
15:40
22.05.2013
15:45
YES
NO
21.05.2013
09:35
21.05.2013
16:00
21.05.2013
16:25
YES
YES
22.05.2013
15:00
22.05.2013
15:05
22.05.2013
15:10
YES
NO
23.05.2013
11:05
17.05.2013
11:05
23.05.2013
11:35
YES
06:45
no
BREAKDOWN
HOUR
tp
MAINTENANCE
BEGINNING OF
MAINTENANCE DATE MAINTENANCE DATE AND
ee
ORDER TYPE
DEMANDED
rr
Table 3.Designed Data Set
WELDING
BREAKDOWN
PACKING
BREAKDOWN
WELDING
YES
23.05.2013
BREAKDOWN
WELDING
YES
24.05.2013
BREAKDOWN
PAINT
YES
25.05.2013
int
BREAKDOWN
MAINTENANCE
PAINT
NO
25.05.2013
BREAKDOWN
WELDING
YES
26.05.2013
06:45
07:00
15:35
09:35
15:00
18.05.2013
23.05.2013
08:45
YES
18.05.2013
07:00
24.05.2013
07:30
YES
22.05.2013
15:40
25.05.2013
15:45
YES
16:00
25.05.2013
16:25
YES
15:05
26.05.2013
15:10
YES
21.05.2013
22.05.2013
pr
Maintenance planning is aimed to prevent downtime and quality defects caused by production
failures, downtime, and repair activities according to the Monte Carlo approach in this study. The
Pr
e
maintenance policy was modeled with the data analyzed, then developed by testing with the VBA
program-based Monte Carlo simulation technique. With this solution approach, it is aimed to
determine periodic preventive maintenance intervals.
This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=4308323
4.1
Data analysis
ed
The columns of the multi-dimensional data that are analyzed are classified as simplified as
follows. The section in which requests type occurs and when downtime is caused by maintenance,
maintenance times, and downtime depending on repair activities are calculated from the columns.
ev
iew
Data for the past two years are divided into monthly periods (Table 4), and the performance of
the system is observed based on these periods. The installation of a new machine is considered an
intuitive maintenance intervention. It can lead to breakdowns and faults in the system that require
compulsory intervention and prediction of failure, or production downtime, and the distribution of
the observed results over the 15-month period is shown in Fig. 5.
rr
Montly Breakdown-Maintenance
Distribution
50
1
2
3
4
5
6
7
8
Compulsory Maintenance
9
ee
0
10 11 12 13 14 15
Heuristic Mantenance
Fig. 5. Monthly Breakdown-Maintenance Distribution
DIAGNOSIS
PROBLEM DATE
AND TIME
no
ORDER TYPE
DEMANDED
DEPARTMENT
tp
Table 4. Classified Data Set
DATE
BREAKDOWN
WELDING
BREAKDOWN
WELDING
10.05.2013
13.05.2013
FINDING
PROBLEM
DEPARTMENT
TIME
17:00
13:00
MAINTENANCE
BEGINNING
DATE AND
TIME
DATE
MAINTENANCE
ENDING DATE AND
TIME
TIME
DATE
TIME
14.05.2013
13:00
13.05.2013
13:20
14.05.2013
18:00
16.05.2013
11:50
17.05.2013
11:35
18.05.2013
08:45
18.05.2013
07:30
22.05.2013
15:45
21.05.2013
16:25
22.05.2013
15:10
1
Mechanic Maintenance 10.05.2013
7:00
1
Mechanic Maintenance 13.05.2013
3:00
MAINTENANCE
BREAKDOWN
PRESS
WELDING
WELDING
pr
BREAKDOWN
int
1
NEW
BREAKDOWN
Pr
e
BREAKDOWN
MAINTENANCE
BREAKDOWN
WELDING
WELDING
WELDING
14.05.2013
16.05.2013
17.05.2013
18.05.2013
14:00
Mechanic Maintenance 14.05.2013
4:00
1
11:30
Mechanic Maintenance 16.05.2013
1:30
1
11:05
Mechanic Maintenance 17.05.2013
1:05
0
06:45
Mechanic Maintenance 18.05.2013
6:45
0
18.05.2013
07:00
Mechanic Maintenance 18.05.2013
7:00
1
20.05.2013
15:35
Electrical Maintenance
22.05.2013
5:40
1
WELDING
21.05.2013
09:35
Machine Maintenance
21.05.2013
6:00
1
WELDING
22.05.2013
15:00
Electrical Maintenance
22.05.2013
5:05
This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=4308323
As shown in Fig.6, the increase in the number of breakdowns in the ongoing periods proves how
ed
much it is necessary to plan this problem appropriately. Also, reducing the number of failures
during the period following intuitive maintenance operations should not be overlooked.
Breakdown
Maintenance
Linear (Breakdown)
60
50
40
30
20
10
0
2
3
4
5
6
7
8
9
10
11
12
13
14
15
rr
1
ev
iew
Distribution
Fig. 6. Breakdown with Increasing Tendency
ee
The model established by the data distribution is associated with macro programming (VBA) and
maintenance and failure data for each section; breakdowns-between-maintenance times,
maintenance times, and probabilities were calculated by the appropriate units (day, hour, etc.) on
tp
average.
The VBA codes were used to measure the analysis, together with their descriptions. The
request types included in the data set were calculated separately by the appropriate units (days,
no
hours), mean duration between failures, average maintenance time and rates, predictive
maintenance activities, and maintenance intervals for each component. Each department’s
performance values are shown in Table 5.
int
Tablo 5: Each department’s performance values
WELDING
PRESS
PACKING
PAINT
MACHINE
1,0769
0,9286
3,1841
0,3141
26,2166
0,0381
9,9379
0,1006
45,6586
0,0219
Day
Day
4,3515
7,4985
18,3854
29,1139
3,1250
Hour
0,2298
0,1334
0,0544
0,0343
0,3200
Hour
AAOSi
I
pr
OBSi
i
Pr
e
The performance indicators in Table 6 were determined by combining all sections due to the effect
of faults occurring in all sections on the overall system performance starting from Table 7. The
effect of the intuitive maintenance sections on the whole system is shown in Table 9.
Table 6. Integrated Maintenance Performance Indicators
This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=4308323
ed
Day
Day
Hour
Hour
ev
iew
AAOS
OBS
ALL
DEPARTMENTS
2,8955
0,3454
8,3401
0,2277
Table 7. Heuristic Maintenance Performance Indicators
ALL
DEPARTMENTS
KBAOS
17,4704
0,0572
11,75
OKBS
0,0851
Hour
rr
5
Day
Day
Hour
RESULTS
All cases that require compulsory or intuitive intervention are combined from the sections, and the
ee
average performance time (BOScm), maintenance rate (), and average downtime (ODS) are
determined as the leading performance indicators in Table 8.
tp
Table 8. Whole System Performance Indicators
2,4838
Day
0,4026
Day
ODS
5,9858 Hour
no
BOScm
Table 13 shows the screen output of the Monte Carlo simulation. X is the expected value of the
random variable θ = E (X) with a probability distribution of the Px estimated value of X the sample
int
mean of this variable;
𝑛
𝜃𝑛(𝑋) = 𝐸(𝑋) = 𝑛∑𝑖 = 1𝑋(𝑖)
1
(16)
can be predicted by the formulation shown in the form. With this formulation, planned or unplanned
pr
maintenance that will occur in each section was estimated. We also mentioned that the number of
failures in the next period decreased as maintenance was performed. The probability of a machine
Pr
e
being maintained to not fail again in the remaining months is 40%. Estimated failures and
maintenance operations were assigned to the months distributed randomly by simulation using the
following formulation.
𝜋
𝑛
𝜃𝑛(𝑋) = 𝐸(𝑋) = 𝑛∑𝑖 = 1𝑋(𝑖).𝜓
1
(17)
This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=4308323
The average rate of the predictive maintenance activities was used. The estimation coefficient
ed
and the average maintenance rate obtained from the whole system were run by five iterations (1000
times) to ensure the closeness to reality with the predicted failure data (Fig. 7). In the future periods
and the periodic maintenance, numbers were determined from the data obtained. In each iteration,
10.00
1500
1000
9.80
1000
500
9.60
50
100
1
2
250
500
9.40
0
3
n
4
5
Periodic Maintenance
ev
iew
approximately ten monthly maintenance numbers have emerged.
rr
Fig. 7. Periodic maintenance results with iterations
The Monte Carlo model provides information about how much maintenance breakdown, halt
and fault data performed in randomly generated numbers according to the observation values and
𝑁𝑖
𝐴𝐴𝑆𝑡
𝐴𝐴𝑂𝑆𝑖 = ∑𝑘 = 1
ee
probability distribution of the historical data in Table 9.
1
𝜆𝑖 = 𝐴𝐴𝑂𝑆İ
𝑁𝑖
(18)
Eq. 18 used for accumulated probability distribution was estimated. Moreover, again, the monthly
tp
breakdown distributions of the historical data are shown in Table 10, and monthly breakdown
maintenance assignments are made.
Table 9. Probability of Breakdowns and Failures in Departments
OBSERVATION
VALUE
PROBABILITY
no
ORDER
WELDER
%
RANGE
0,63
0-0,63
PRESS
135
0,25
0,64-0.88
PAINT
47
0,07
0,89-0,96
PACKAGING
10
0,02
0,97-0,99
MACHINE
7
0,01
0,99-1
int
350
Based on the data, the main performance indicator in which the Average Maintenance Time (OBS)
pr
and Average Preventive Maintenance Time (OBS) are combined in the system is the average results
of the average downtime;
𝜆.𝑂𝐵𝑆 + 𝜋.𝑂𝐾𝐵𝑆
=
𝜆+𝜋
Pr
e
𝑂𝐷𝑆 =
= 3,591499.
It was found that the average stop time obtained in the historical data is less than 2,5 hours. Even
though the fault data in the simulation model arises from the number of failures in the historical
data, it is possible to capture this performance value if the maintenance planning is done ten times a
month and an assignment is made with a schedule outside the working hours that can be fully
This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=4308323
controlled. Similarly, the system can be updated by making more periodic assignments to the
ed
department or machines where the number of failures is observed.
It was found that the average stop time obtained in the historical data is less than 2,5 hours. Even
though the fault data in the simulation model arises from the number of failures in the historical
ev
iew
data, it is possible to capture this performance value if the maintenance planning is done ten times a
month and an assignment is made with a schedule outside the working hours that can be fully
controlled. Similarly, the system can be updated by making more periodic assignments to the
department or machines where the number of failures is observed.
Although it is assumed that the maintenance data is typically distributed in each section, it was
decided to perform an ANOVA test based on the assumption that it was a dependent situation
considering that it was distributed in 5 different sections.
rr
ANOVA method with a 95% confidence interval was used to test the proximity of the model
obtained by using the Monte Carlo simulation technique with the data set, and the results given in
Table 10 were obtained. According to these results, the Bias 0.000 value and the Sig. = 0.094 value
ee
was found to be significant. ANOVA test results are shown in Table 11.
Table10. SPSS Statistical Analysis Results
Descriptives
tp
ORDER TYPE
Total
N
no
Statistic
Bootstrapa
95% Confidence
Interval
Std.
Bias
Error
Lower
Upper
549
0
0
549
549
1,09
,00
,01
1,07
1,12
Std. Deviation
,342
,000
,030
,284
,403
Std. Error
,015
Mean
Lower
Confidence
Bound
Interval for
Upper
Mean
Bound
1,06
int
95%
1,12
NEW
Maximum
BREAKDOWN
Pr
e
pr
Minimum
Table 11. ANOVA test results
ANOVA
ORDER TYPE
Between Groups
Within Groups
Sum of Squares
df
Mean Square
,929
4
,232
63,333
544
,116
F
1,996
Sig.
,094
This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=4308323
64,262
548
ed
Total
In addition, with the large sample size, the monthly expected value (number of maintenance) for
Monte Carlo simulation was increased and the Chi-Square test was applied in SPSS and the results
Table12. SPSS Reliability Analysis (%95)
DEPARTMENT
PACKING
TYPE
DOWN
MACHINE
8a
42a
337b
6a, b
1,6%
8,3%
66,5%
1,2%
80,0%
89,4%
96,3%
85,7%
MAINT- Count
1a, b, c
4c
7b
1a, c
NANCE
3,0%
12,1%
21,2%
3,0%
10,0%
8,5%
2,0%
1a
1a, b
6a, b
% within ORDERTYPE
11,1%
11,1%
66,7%
% within SECTION
10,0%
2,1%
10
47
1,8%
8,6%
100,0%
100,0%
% within ORDERTYPE
% within SECTION
% within ORDERTYPE
% within SECTION
NEW
Total
WELDING
Count
Count
% within ORDERTYPE
114a
507
22,5%
100,0%
84,4%
92,3%
20a, c
33
60,6%
100,0%
14,3%
14,8%
6,0%
0a, b
1b
9
0,0%
11,1%
100,0%
1,7%
0,0%
0,7%
1,6%
350
7
135
549
63,8%
1,3%
24,6%
100,0%
100,0%
100,0%
100,0%
100,0%
tp
% within SECTION
PRESS
rr
BREAK- Count
Total
ee
ORDER
PAINT
ev
iew
in Table 12 and Table 13 are demonstrated.
Table 13. Monte Carlo Simulation Conformity Test
no
Monte Carlo Conformity Test
df
35,194a
8
Likelihood Ratio
31,066
8
Fisher's Exact Test
34,835
pr
Value
549
Chi-
int
Pearson
Monte Carlo Sig. (2-sided)
95% Confidence Interval
Asymptotic
Significance
(2-sided)
Lower
Significance
Bound
Upper Bound
,000
,009b
,001
,017
,000
,000b
,000
,005
,000b
,000
,005
Square
N of Valid Cases
a. 7 cells (46,7%) have expected count less than 5. The minimum expected count is ,11.
Pr
e
b. Based on 555 sampled tables with starting seed 743671174.
It was found that the minimum expected value was 11; it was well-matched with the Monte
Carlo simulation model with ten maintenance assignments in Table 13.
This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=4308323
6
CONCLUSION
ed
The preventive maintenance approach in the maintenance planning process aimed to develop
a sustainable maintenance policy of the system without any disturbance and failure in the long term
and without any decrease in production quality. In addition, it is aimed to determine the preventive
ev
iew
maintenance policy by analyzing the existing system and predicting future situations through
simulation.
The results are classified according to the closeness to reality, and the duration of the monthly
downtime, which is the main performance indicator, is calculated as approximately ODS = 6 hours
for the historical maintenance data. This breakdown time is the actual loss of time that affects
various actions such as production employee idle, product loss, and production time or unforeseen
rr
costs. Quality of production is reduced due to production losses, waste rate, and defective
production due to machinery and workers working at a standstill.
ee
The system is determined as ODS = 2,5 hours with the simulation model, an order has been
created in the complex system, and the periods considered idle have been reduced with ten
tp
maintenance assignments in each period for the coming months.
The most relative benefit of the model to the system is renewable. As the real-time data
continues to be stored, then historical system data will be refreshed, and authorized persons will
no
monitor new system performance. The model can be re-engineered at the same time, allowing
advanced analysis for unwanted situations.
References
int
[1] T. Nguyen, L. Zhou, V. Spiegler, P. Ieromoachou, Y. Lin, Big data analytics in supply chain
management: A state-of-the-art literature review. Computers & Operations Research Oct. 2018; 98,
254-364.
pr
[2] A. Siddiqa, İ. Abaker, T. Hashem, İ. Yaqoob, M. Marjani, S. Shamshirband, A. Gani, F.
Nasaruddin, A survey of big data management: Taxonomy and state-of-the-art. Journal of Network
Pr
e
and Computer Applications Aug. 2016; 71, 151-166.
[3] K. Moharm, State of the art in big data applications in microgrid: A review. Advanced
Engineering Informatics, V. 42 (Oct. 2019) 100945
[4] Z. You, C. Wu, A framework for data-driven informatization of the construction compant.
Advanced Engineering Informatics Jan. 2019; 39, 269-277
This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=4308323
[5] S. Aheleroff, X. Xu, Y. Lu, M. Ariatızabai, J.P. Velasquez, B. Joa, Y. Valencia, IoT-enabled
ed
smart applications under industry 4.0: A case study. Advanced Engineering Informatics Jan. 2020 ;
43, 101043
[6] S. Faghih-Roohi, M. K. Xie, N. Ming, Accident risk assessment in marine transportation via
ev
iew
Markov modelling and Markov Chain Monte Carlo simulation. Ocean Engineering 2014; 91, 363–
370
[7] A.A. Mendes, M.W. Lorenzoni, Analysis and optimization of periodic inspection intervals in
cold standby. Journal of Manufacturing Systems 2018; 49, 121–130
[8] V.K. Gaurav, C. Sharma, Estimating health risks in metal contaminated land for sustainable
agriculture in peri-urban industrial areas using Monte Carlo probabilistic approach. Sustainable
Computing: Informatics and Systems 2019 ;https://doi.org/10.1016/j.suscom. 2019.01.012
rr
[9] J. C. P. Cheng, W. Chen, K. Chen, Q. Wang, Data-driven predictive maintenance planning
framework for MEP components based on BIM and IoT using machine learning algorithms.
Automation in Construction Apr. 2020;112, 103087
ee
[10] I.P.S. Ahuja, J.S. Khamba, Total productive maintenance: literature review and directions.
International Journal of Quality & Reliability Management 2008; 25, 709-756.
[11] A. Airabghi, A. Tiwari, State of the art in simulation-based optimisation for maintenance
[12] P. Callewaert,
tp
systems. Computers & Industrial Engineering Apr. 2015; 82, 167-182.
W.J.C. Verhagen, R. Curran,
Integrating maintenance work progress
monitoring into aircraft maintenance planning decision support., Transportation Research Procedia
no
2018; 29, 58-69
[13] S.O. Duffuaa, M. Ben-Daya, K.S. Al-Sultan, A.A. Andijani, A generic conceptual simulation
model for maintenance systems. Journal of Quality in Maintenance Engineering 2001; 7(3), 207 –
219.
int
[14] D. Bumblauskas, A Markov decision process model case for optimal maintenance of serially
dependent power system components. Journal of Quality in Maintenance Engineering 2015; 21(3)
271 – 293
pr
[15] B. Laurel, (Ed.). The Art of Human-Computer Interface Design (1990);Reading, MA:
Addison-Wesley.
Pr
e
[16] M. Savsar,
Analysis and modeling of maintenance operations in the context of an oilfilling
plant. Journal of Manufacturing Technology Management 2011; 22 (5), 679 – 697
[17] T. Rosqvist, K. Laakso, M. Reunanen, Value-Driven Maintenance Planning for a Production
Plant. Reliability Engineering and System SafetyJan. 2007; 94(1), 97-110
[18] M. Rausand,
A. Hoyland, System Reliability Theory: Models, Statistical Methods, and
Applications. John Wiley & Sons Inx. 2004; Hoboken, New Jersey, ISBN 0-471-47133-X
This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=4308323
[19] J. Sarkis, D.H. Liles, Using IDEF and QFD to develop an organizational decision support
ed
methodology for the strategic justification of computer-integrated Technologies. lnternational
Journal of Project Management 1995; 13(3), 177-185.
[20] T. A. Salman, S.M.A. Suliman, Modelling preventive maintenance for auxiliary components.
ev
iew
Journal of Quality in Maintenance Engineering 2008; 14(2), 148 – 160
[21] L. Swanson, Linking Maintenance Strategies to Performance. International Journal of
Production Economics 2001; 70, 237–244
[22] M. Ben-Daya, S.O. Duffuaa, Maintenance and Quality: The Missing Link. Journal of Quality
in Maintenance Engineering 1995; 1 (1), 20-26.
[23] E.N. Pistikopolos,
C.G. Vassiliadis,
J. Arvela, L.G. Papageorgiou, Interactions of
maintenance and production planning for multipurpose process plants – a system effectiveness
rr
approach, Industrial and Engineering Chemistry Research 2001; 40(3)195-207.
[24] A. Portioli-Staudacher, M.Tantardini, Integrated maintenance and production planning: a
model to include rescheduling costs. Journal of Quality in Maintenance Engineering 2012; 18(1),
ee
42 – 59
[25] N.M. Paz, W. Leigh, Maintenance Scheduling: Issues, Result and Research Needs.
International Journal of Operations & Production Management 1994; 14(8), 47–69
tp
[26] R.Dekker, Applications of Maintenance Optimization Models: A Review and Analyis,
Reliability Engineering and System Safety 1996; 51 229.
no
1. International Journal of Information Management
Computers in Industry
6.1
Aims & Scope
int
The aim of Computers in Industry is to publish original, high-quality,
application-oriented
research
papers
that:
Pr
e
pr
• Show new trends in and options for the use of Information and
Communication
Technology in
industry;
• Link or integrate different technology fields in the broad area
of computer
applications for industry;
• Link or integrate different application areas of ICT in industry.
General
topics
covered
include
the
following
areas:
• The unique application of ICT in business processes such as design,
engineering, manufacturing, purchasing, physical distribution, production
This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=4308323
ev
iew
ed
management and supply chain management. This is the main thrust of
the journal. It includes research in integration of business process
support,
such
as
in
enterprise
modelling,
ERP,
EDM.
• The industrial use of ICT in knowledge intensive fields such as quality
control, logistics, engineering data management, and product
documentation
will
certainly
be
considered.
• Demonstration of enabling capabilities of new or existing technologies
such as hard real time systems, knowledge engineering, applied fuzzy
logic, collaborative work systems, and intelligence agents are also
welcomed.
• Papers solely focusing on ICT or manufacturing processes may be
considered
out
of
scope.
Technologically
Application-oriented
Representative
for
outstanding
with
a
research
at
ee
-
rr
A continuous quality policy, based on strict peer reviewing shall ensure
that
published
articles
are:
and
front-end
generalised
message
an
international
level
no
tp
Benefits
to
authors
We also provide many author benefits, such as free PDFs, a liberal
copyright policy, special discounts on Elsevier publications and much
more. Please click here for more information on our author services.
Pr
e
pr
int
Please see our Guide for Authors for information on article submission.
If you require any further information or help, please visit our Support
Center
This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=4308323