If down, do

(1)

Master’s thesis

If down, do

Selecting opportunistic maintenance tasks during unexpected downtime

Author

ing R. Schuurman S2395800

Date 21-08-2014

University of Groningen

Supervisor

Faculty of Economics and Business

dr. N.D. van Foreest

Technology and Operations Management

Co-assessor

(2)

Abstract

Unexpected downtime in continuously operating plants often result in big losses. To minimize unexpected downtime, preventive maintenance is widely regarded as the best solution.

However, when preventive maintenance is needed on all machines in a multi machine series configuration, the plant has to shut down in its entirety for the required tasks. If more tasks can be done simultaneously, the downtime can be used more efficient. Opportunistic maintenance, which is the execution of preventive maintenance tasks before they are needed, but because the opportunity arises, might lower overall costs.

In current literature, opportunistic maintenance is only considered in combination with preventive maintenance, because preventive maintenance can be planned. To be able to perform opportunistic maintenance during unexpected downtime, a fast and accurate selecting method for opportunistic maintenance tasks is needed, to efficiently select the best candidate maintenance task.

The system developed in this research can select all maintenance tasks which are possible to do in the current time window. The selecting procedure is based on available parts, tools, duration of tasks and the available time in the current time window, skill level of the present employee(s) and the state of the factory. A prioritizing method is built to show which task is the most profitable to do if multiple tasks are possible to do in the current time window.

This research is conducted at FrieslandCampina Bedum, a cheese and cheese-whey producing plant in the Netherlands.

Keywords: Opportunistic maintenance, task selecting, maintenance, maintenance continuously operating

(3)

If down, do. Selecting opportunistic maintenance tasks during unexpected downtime. 3 Preface ... 4 1. Introduction ... 5 1.1. Research questions... 6 2. Theoretical background ... 7 2.1. Maintenance... 7

2.2. Optimum age based preventive maintenance ... 7

2.3. Opportunistic maintenance ... 9

2.4. Variables for opportunistic maintenance ... 10

3. Methodology ... 12

4. Findings... 14

4.1. Task selecting ... 15

4.2. Task prioritizing ... 16

5. Decision support system ... 21

5.1. Database structure ... 21

5.2. Working with the decision support system. ... 22

(4)

If down, do. Selecting opportunistic maintenance tasks during unexpected downtime. 4

Preface

This report is the result of my master’s thesis performed at FrieslandCampina Bedum, The Netherlands. It forms the end of my Master of Science degree on Technology and Operations Management at the University of Groningen.

Although in the end, I sign this report with only my own name, a lot of people helped me to get most out of this report. To start, I would like to thank my supervisor Nicky van Foreest for giving me the freedom to work on the subject I liked and the given feedback which helped me finishing this report. I would also like to thank all the employees at FrieslandCampina Bedum, especially Edwin Kreder who helped me to get the correct data I needed to develop the decision support system. I would also like to thank my fellow students who gave feedback and helped me with valuable insight into solving the problems I faced.

During this research I learned a lot about maintenance, rule based expert systems and working independent on my own project responsibly. I found these subjects and skills very interesting and informative and I am sure these skills and knowledge will help me in my career.

(5)

1. Introduction

Unexpected downtime in continuously operating plants often results in big losses. To minimize unexpected downtime, preventive maintenance is widely regarded as the best solution (Laggoune, Chateauneuf, & Aissani, 2009; Xiaojun Zhou, Xi, & Lee, 2009). However when preventive maintenance is needed on all machines in a multi machine series configuration, the plant has to shut down in its entirety for the required tasks. Opportunistic maintenance, which is the execution of preventive maintenance tasks before they are needed but because the opportunity arises, might lower overall costs as a single shutdown but with more tasks simultaneously can be less costly than the sum of all costs when shutting down the plant multiple times (Laggoune et al., 2009).

This research provides a design science study within FrieslandCampina Bedum (FCB), a producer of cheese and milk-whey in the Netherlands. Currently the maintenance department is under constant pressure to keep the factory up and running with as little planned and unexpected downtime as possible. Over 2013, production line ‘Cheese 1’ had 20.4 hours of planned downtime while it had 331.7 hours of unplanned downtime. When one machine is down, the complete production line is down, so the question arises if it is possible to carry out preventive maintenance tasks during unexpected downtimes and to find a procedure to quickly decide which preventive maintenance tasks to do when unexpected downtime occurs.

The goal of this research is to find a simple-to-use procedure to quickly and accurately decide which preventive maintenance tasks to do when an opportunity arises (unexpected downtime or planned preventive maintenance downtime) in a multi-unit system with perishable goods. If this procedure is quick and accurate, the user will have the opportunity to perform opportunistic maintenance during unexpected downtime. The decision of which tasks to perform should be based on whether a task is feasible for that specific moment, and what the costs and benefits are of executing this task.

(6)

If down, do. Selecting opportunistic maintenance tasks during unexpected downtime. 6 This thesis is structured as follows, in section 1.1 the research questions will be discussed, chapter 2 contains the background analysis. Chapter 3 consists of the used methodology for the research and the 4th chapter gives the findings of this research. In chapter 5 the proposed decision support system is explained. Chapter 6 and 7 are respectively the discussion and conclusion.

1.1. Research questions

Maintenance task selecting and ordering during (un)planned downtime consist of determining the expected length of the downtime, and knowing the length and importance of the potential preventive maintenance tasks to perform. This research only focuses on the actual decision of which tasks to do when downtime occurs. The expected length of the downtime and all possible preventive maintenance tasks with their durations are expected to be known.

Main research question:

“In a continuous multi series production environment we build a decision support system to help decide which preventive maintenance tasks to do when an opportunity occurs to decrease total maintenance costs.”

The main research question is divided into a few sub research questions to give a structure to the research. The answer on all sub research questions combined should be the solution to the main research question. The sub research questions are:

• What are the objectives when implementing opportunistic maintenance? • What are input variables for the decision to do opportunistic maintenance?

• Which costs and benefits are associated when rescheduling a preventive maintenance task when an opportunity arises?

• Which costs are associated when not performing opportunistic maintenance?

(7)

2. Theoretical background

The aim of this research is to find a procedure to quickly and accurately decide which preventive maintenance tasks to do when an opportunity arises (unexpected downtime or planned preventive maintenance downtime). This background section will give more info on what research has been done and what research has not been done but is needed for building a decision support system.

Section 2.1 discusses the characteristics of maintenance, section 2.2 focuses on the concept opportunistic maintenance and in section 2.3 the variables when applying opportunistic maintenance will be discussed. Section 2.4 will give more background on rescheduling preventive maintenance tasks and finally in section 2.5, the conceptual model will be presented.

2.1. Maintenance

Maintenance can be broadly divided into reactive/unplanned and proactive/planned maintenance (Veldman, Wortmann, & Klingenberg, 2011). Figure 2-1 shows an overview of currently known types of maintenance. Unplanned maintenance will be conducted when a failure occurs and the condition before the failure should be restored (corrective maintenance) or when immediately action is needed to avoid dangerous situations (emergency maintenance) (Veldman et al., 2011). Planned maintenance is divided into preventive maintenance and predictive maintenance. Preventive maintenance is scheduled downtime, usually periodical, in which a set of defined tasks are performed (Laggoune et al., 2009). Predictive maintenance is a policy in which selected physical parameters associated with the operating machine are monitored, measured or recorded either intermittently or continuously with the purpose to support decisions related to the operation and maintenance of the operating machine.

2.2.

(8)

Optimum age based preventive maintenance

Barlow & Hunter (1960) introduced the age based preventive maintenance policy in which a part can only be repaired or replaced when the part breaks or when a constant time T is elapsed since the last repair. The goal of the optimum age based preventive maintenance policy is to find the time T at which the mean cost per unit time is the lowest. The variable T is the time between the last maintenance action (preventive or corrective maintenance) and the preventive maintenance action. A higher value of T implies less preventive maintenance but the chance of a breakdown increases with time T which means more corrective maintenance is expected.

This age based policy assumes that the likeliness of a breakdown is increasing over time and the part can be restored to a good as new state. To calculate which time T is optimal to perform preventive maintenance, the costs of preventive, cost of corrective maintenance and the cumulative failure distribution in time (F(T)) should be known. To calculate the optimal level of T, the following formula is used.

( 2-1 )

( 2-2 )

In which is the mean cost per unit time with a time between maintenance jobs of T. This variable shoud be minimized by changing the time T. F(T) is the cumulative failure distribution in time T and c is the fraction of preventive maintenance cost versus corrective maintenance costs. This factor is always between 0 and 1 and can be calculated with formula 2-3.

( 2-3 ) Because it is difficult to predict the time a part failes, failure distributions are used. This however means that there is a chance a part will break in which corrective maintenance has to be performed. If the value of c is low, preventive maintenance is relative cheap and taking risk with a higher T is not economical and the optimal T will be lower. With a high c, preventive maintenance is less favorable and the optimal T will be higher thus more corrective maintenance is needed.

(9)

2.3. Opportunistic maintenance

Opportunistic maintenance (OM) is the performance of non-critical preventive maintenance (PM) tasks when the opportunity arises because of the state of the complete system (Wang, 2002). OM often has great potential costs savings in multi machine series configuration which have high costs of unavailability (Laggoune et al., 2009; Wang, 2002; Xiaojun Zhou et al., 2009).

The current focus in literature of OM is on performing OM while on other machines PM is being performed or in a limited setting with only two machines. Current OM research does not focus on unexpected downtime and multiple unplanned PM tasks which might have to be performed with different characteristics.

To illustrate the benefit of OM, consider the simplified case with a series production set-up with two machines. The first machine has an optimal age based preventive maintenance interval of three days; the second machine has an optimal age based preventive maintenance interval of four days. Both machines have a repair time of one hour. To simplify this example, no corrective maintenance is needed, and when two preventive maintenance tasks are scheduled on the same day, they are executed in parallel. Without OM, the maintenance schedule will look like figure 2-2. It shows that seven out of twelve days have downtime.

Without opportunistic maintenance

Day 0 1 2 3 4 5 6 7 8 9 10 11 12 Count

Machine one: T=3 5

Machine two: T=4 4

Downtime 7

Figure 2-2: Preventive maintenance schedule without opportunistic maintenance

With an OM policy the maintenance schedule will look like figure 2-3. Since both preventive maintenance tasks can be done simultaneously, OM is performed on machine two, while preventive maintenance is performed on machine one. In this schedule, instead of seven, there are only five days which have a downtime and machine two is maintained five times instead of four times. This is favorable if the cost of one time preventive maintenance on machine two is lower than the costs of two hours of downtime of the production plant.

With opportunistic maintenance

Day 0 1 2 3 4 5 6 7 8 9 10 11 12 Count

Machine one: T=3 5

Machine two: T=4 5

Downtime 5

(10)

If down, do. Selecting opportunistic maintenance tasks during unexpected downtime. 10 When performing OM during unexpected downtime, the same benefits will occur because there will be less planned preventive and more opportunistic maintenance and therefore less downtime. However, this will mean maintenance is performed more frequent than the optimal age based maintenance policy because the maintenance is performed before the optimal age based maintenance interval ends.

The challenge with OM during unexpected downtime is the unpredictable nature of unexpected downtime in duration and timing. This may results in a different task selection every occasion because input variables differ and thus a prepared list with preventive maintenance tasks is not a suitable option. A decision support system should help with the decision which task(s), if any, to do simultaneously with the repair of the breakdown. The decision has to be made in the shortest amount of time, because time which is not spend on selecting an opportunistic maintenance task can be spend on executing the task. This might result in more tasks to be able to do and in the end a higher uptime of the plant and lower overall maintenance cost. Therefore the task selection and prioritizing should be quick and accurate. Every decision on which task(s) to do should be based on overall costs and benefits, as performing a task too often will result in higher overall maintenance cost while downtime does not decrease.

2.4. Variables for opportunistic maintenance

In literature variables for selecting maintenance tasks can be found, although the focus on OM is in combination with planned preventive maintenance and not during unexpected downtime. The following variables can be found in literature or are found when interviewing key stakeholders.

OM policies. When deciding which tasks to plan, multiple policies are used. The most common is for

every component to have a time period in its life cycle in which a PM task should be performed if the opportunity arises which ends with the moment the component reaches its reliability threshold and the maintenance will be performed as a standard PM task (Berg, 1978; Laggoune et al., 2009; Xiaojun Zhou et al., 2009). Another model is replacing all components when the opportunity arises (Xiaojun Zhou et al., 2009). In the paper of Zhou (2009) a model is introduced which calculates for every PM task the possible costs savings when that task is performed at an occurring opportunity.

Available time. Current OM policies do not take into account perishable products, however time can

have a big influence in costs when working with perishable goods as products might perish and can be a decisive factor when planning OM tasks during unexpected downtime.

Workforce constraints. Some tasks can only be performed when a certain amount of personnel is

available, or tasks need a certain skill which available personnel does not have. Because of this constraint, certain tasks cannot be performed when the opportunity arises.

State of the factory. During unexpected downtime, the factory might be in a state in which certain PM

(11)

If down, do. Selecting opportunistic maintenance tasks during unexpected downtime. 11 that certain task within a time frame, or a certain tasks takes longer as the factory has to be transferred into another state.

Costs. For every business decision, costs are an important factor. Every OM planning policy should have a

cost consideration. Costs associated with OM are: costs of extra downtime in the current time window, costs of products loss because it will perish, increase of maintenance costs as tasks will be performed more often.

Task prioritization. When selecting which tasks to perform when a breakdown occurs, it is important to

(12)

3. Methodology

This research consists of a design science research performed at FrieslandCampina Bedum, a producer of cheese and milk-whey in the Netherlands. All input variables and the opportunistic maintenance policies are aimed at the production plant of FrieslandCampina Bedum.

A design science research is chosen because the key stakeholders want to change how task selecting during unexpected downtime works as there currently is not a quickly and accurately procedure to decide which preventive maintenance tasks to do when an opportunity arises. This type of problem is a practical problem, which is solved using a design science research. By solving this practical problem, knowledge is gained on how the world works which might solve a knowledge problem (Wieringa, 2009). To structure this research, the research is divided into the following five phases which are proposed by Peffers, Tuunanen, Rothenberger & Chatterjee (2007) for doing design science research. All five phases of the research will be discussed and tradeoffs during the research will be explained.

1. Problem identification and motivation

Based on interviews at FrieslandCampina Bedum and literature, the problem is defined and the value of a solution is motivated. The problem which is defined is the high pressure on maintenance tasks at FrieslandCampina Bedum. This is because of a high cost of downtime and a low budget. A possible solution to this problem is a better utilization of maintenance time and specifically more efficient use of the downtime caused by different failing parts of the plant.

A solution to quickly and accurately decide which task to do increases the time maintenance engineers can work on these tasks and thus increase the number of tasks being done during downtime of the machine. These tasks do not have to be done later when the machine has to be stopped which results in a higher uptime and if done right, a decrease in total maintenance cost.

2. Objective and stakeholders definition

(13)

If down, do. Selecting opportunistic maintenance tasks during unexpected downtime. 13 3. Design and development

In the design and development stage, multiple solutions are considered and the solution which fits the objectives best is further developed. Since the goal of the maintenance engineers is a system which was reliable and fast, a software rule based expert system is chosen because it can process large amounts of up to date information fast and accurately. Since data can be presented in the same way for every occasion, a rule based expert system is proposed.

To satisfy the stakeholders, maintenance tasks which after considering all constraints are possible to do are ranked on their contribution to the lowering of overall costs of maintenance and tasks with a negative influence on lowering the overall costs of maintenance are not able to be selected.

4. Evaluation

The developed solution in the third phase, is tested in the evaluation phase. The model is checked for a small list of possible maintenance tasks to see if the correct tasks are being selected and if the prioritizing is correct. A small list is chosen to be able to check the outcome manually. The scalability of the system is tested by adding multiple parts, tasks and employees.

5. Communication

The final stage of the research is to communicate the results. With this research paper and a presentation at FrieslandCampina Bedum, results are being shown and the opportunistic task selecting and prioritizing system is explained.

FrieslandCampina will be made familiar with the developed system and a manual will be provided.

(14)

4. Findings

The decision of which task to do when the opportunity arises consists of multiple phases. The first phase is estimating the downtime of the current breakdown and assign employee(s) to fix the breakdown. This phase is outside the scope of the research, it is therefore assumed that the estimated downtime is known and it will serve as an input variable in the decision process.

The entire decision process is conceived and is schematically displayed in figure 4-1. The scope of this research consists of two phases, namely the task selecting and the task prioritizing phases. In the task selecting phase all tasks which are feasible to do are selected from a list of all preventive maintenance tasks. The findings on the task selecting phase will be discussed in section 4.1. In the task prioritizing phase the tasks which are feasible to do will be prioritized based on their expected profit. The findings of task prioritization will be discussed in section 4.2.

Figure 4-1: Schematic overview OM task selecting procedure

Task selecting and prioritizing should be done quick and accurate. The task selecting process consists of multiple checks if the current state is feasible for preventive maintenance tasks. The task prioritizing process consists of multiple calculations. Since a computer can do both in a quick and accurate way, both phases are designed in such a way that they can be implemented with software.

(15)

4.1. Task selecting

The goal of task selecting is to select all the tasks which can be done in the current time window. Task selecting should be a quick and accurate process, therefore a rule based expert system is chosen, which can be executed by a computer. A rule based expert system is a system which mimics the problem solving technique from an expert by checking facts versus known rules which are embedded in software and forming conclusions (Negnevitsky, 2001). Since the variables in deciding which opportunistic maintenance (OM) task to select are facts (e.g. is part x on the shelf? Is a person in skill group B present?) a rule based expert system is suitable.

Example of a rule based expert system (Negnevitsky, 2001: 26): If The ‘Traffic light’ is ‘Green’

Then Action is “Go”

If The ‘Traffic light’ is ‘Red’ Then Action is “Stop”

Within the above mentioned rules, an action can be found by following rules when coming across traffic lights, a process in which most humans are experts but computers are not. In OM a likewise expert system can be used to find out which tasks are suitable to perform in the current time window.

Task selecting is done with hard requirements; requirements which cannot be relaxed and must be strictly satisfied (Giachetti, 1998). Every plant has different hard requirements, for FrieslandCampina Bedum, where this research is conducted, the following hard requirements were found:

• Parts and tools on stock - When performing maintenance, parts and tools may be needed to perform a certain task. A rule in the expert system checks if the needed parts and tools for each task are on stock in at least the desired quantity. The current stock level of all parts is held in a database.

• Skill present employee – An employee who is present and is not fixing the breakdown has a certain skill set, and all tasks have a required skill set to be executed. When running the expert system, a comparison is made between desired skill set and available skill set to determine which tasks are feasible. The skill of the employee of every skill group (mechanical, programming, electrical, etc.) should be at least equal compared to the desired level of that specific skill group. • Time available – The estimated downtime should be more than the estimated time needed for

(16)

If down, do. Selecting opportunistic maintenance tasks during unexpected downtime. 16 • State of the plant – Some maintenance tasks need a certain state of the machine, for instance, if

the inside of a pump has to be cleaned, the machine cannot be full with semi-finished goods. When checking all these requirements on the preventive maintenance tasks, a list can be composed with all feasible tasks. Other companies might have different hard requirements, these can be easily added to the rule based expert system.

4.2. Task prioritizing

With the proposed selecting program, tasks are selected which are possible to execute. However, a task might be possible but when that task is executed in that specific time window, overall maintenance costs might be higher compared to when the task is not executed as this certain task is performed too often. For every task a consideration should be made based on the failure rate, cost of repair, cost of failure and time since last repair. A computer might be able to calculate the best possible task to execute, but the final decision on which task to do is taken by the operator.

The optimum age based preventive maintenance policy is a strategy where the mean costs per unit time (η(T)) is the lowest (Barlow & Hunter, 1960). To calculate the mean costs per unit time, equation 2-2 is used. In this formula, c is the cost ratio between preventive maintenance and corrective maintenance. F is the cumulative failure distribution in time T. The failure distribution should be based on the (mathematical) distributions which fits the machine’s failure distribution best. With equation 2-2, an optimum can be found for each machine. In appendix A, a Weibull, normal and gamma distribution are shown with the mean cost per unit time.

The mean costs per unit time has an optimum in which the total costs of preventive maintenance and corrective maintenance are at the lowest point. The costs per unit time resulting from the optimum age based maintenance policy is the starting point for the calculation of possible cost savings when performing opportunistic maintenance.

Opportunistic maintenance(OM) is: The execution of preventive maintenance tasks before they are needed but because the opportunity arises (Laggoune et al., 2009). This means that the task which needs to be done is preventive and not corrective. Implying that the extra costs associated with possible breakdowns do not have to be taken into account and only preventive maintenance costs should be included. The cycle length is the time since the last repair as the costs of OM can be spread over this period to gain the same date type, which is mean cost per unit time.

(17)

om ( 4-1 )

om

( 4-2 ) c is the fraction of opportunistic maintenance cost versus corrective maintenance costs. This factor is between 0 and 1 and can be calculated with formula 4-3.

Cost of corrective maintenance ( 4-3 )

Formulas 2-2 and 4-2 are graphically displayed in figure 4-2 to show the possible cost savings of OM. Both formulas are plotted with a variable time T on the horizontal axis, and a resulting cost per unit time on the vertical axis. In this graph, the solid line is formula 2-2 plotted over time t, which represents the expected costs of preventive maintenance per unit time if a maintenance policy is chosen with the time between preventive maintenance tasks as displayed on the x-axis.

The dotted line is formula 4-2 plotted over time T, which represents the costs per unit time when the task can be executed as an OM task. Even thought it is plotted as a line, opportunities for performing OM tasks do not arise continously. This line only shows what the cost per unit time are when an opportunity arises.

The input failure distribution is a Weibull distribution with α = 2 and β = 6. The value of C, which is the cost fraction of preventive maintenance versus corrective maintenance equals the cost fraction of opportunistic maintenance versus corrective maintenance and is 0.1.

(18)

If down, do. Selecting opportunistic maintenance tasks during unexpected downtime. 18 Figure 4-2 also shows that the costs of opportunistic maintenance will be lower than the costs associated with the optimum age based preventive maintenance policy. This is as expected because the costs of possible breakdowns do not have to be included. Once the costs of opportunistic maintenance is lower than the costs of the optimum age based preventive maintenance policy, performing opportunistic maintenance will yield a lower overall maintenance cost and thus a profit.

With the input variables of figure 4-2, the optimum age based preventive maintenance policy will yield a cost factor of 0.1 per unit time and the corresponding time between maintenance tasks is 2 units time. For opportunistic maintenance, this cost factor per unit time is achieved after 1 unit time. This means, that if the opportunity arises between 0 and 1 unit time of the last maintenance action, OM on this machine is not profitable. However, OM is profitable if the opportunity arises between 1 and 2 unit time since the last maintenance action. The closer the opportunity is to the optimum time between maintenance tasks (t = 2), the higher the profit is when performing this OM task. In the graph, this OM area is marked by a dashed line.The OM period ends when the part reaches its optimum age based time as continuing onwards will increase the risk of higher costs. and preventive maintenance should be performed.

Example:

Consider the same machine as discussed previous with a Weibull failure distribution with α = 2 and β = 6 and a cost factor C, for preventive and opportunistic maintenance of 0.1. In the normal age based preventive maintenance, maintenance is done every 2 time units at an average costfactor of 0.1 per unit time as this is the lowest possible cost per unit time

In table XXXXX the time since last maintenance task (t) is displayed for every time an opportunity occurs or preventive maintenance is planned. If an opportunity occurs, ŋom

_(t)

is calculated and the decision to do opportunistic maintenance or not is given.

Time Opportunity ŋom

(t)

Decision

0.0 No 0.1 Preventive maintenance was performed.

0.2 Yes 0.5 Do not perform OM, as cost per unit time is higher then normal preventive maintenance cost

0.8 Yes 0.125 Do not perform OM, as cost per unit time is higher then normal preventive maintenance cost

1.5 Yes 0.667 Perform opportunistic maintenance, costs per unit time is lower then normal preventive maintenance costs

2.0 No 0.1 Perform preventive maintenance if time since last

maintenance task (T) is 2

3.5 No 0.1 Perform preventive maintenance if time since last

(19)

If down, do. Selecting opportunistic maintenance tasks during unexpected downtime. 19 On time 0.2, 0.8 and 1.5 an opportunity occurs to do OM. Equation 4-2 yields cost factors of respectively 0.5, 0.125 and 0.0667. The first two opportunities have a cost factor which is higher than the optimum age based maintenance policy, making OM nog feasible costwise. However, the third opportunity has a cost factor which is lower than the preventive maintenance costs, and taking this opportunity will lower overall maintenance costs. If this opportunity is taken the next preventive maintenance task is planned on time 1.5 + 2 = 3.5 and not on the previously planned 2.0, if the opportunity is not taken because an other opportunity is better, the planned maintenance on time 2.0 should still be executed.

4.2.1. Calculating net profit.

With the equations 2-2 and 4-2, the costs per unit time are calculated for preventive and opportunistic maintenance as a fraction of corrective maintenance costs. However, these costs should be recalculated to real currency figures so they can be compared to other tasks with different failure distributions and costs. This can be done with equation 4-4.

( 4-4 )

Formula 4-3 still yields a profit factor relative to the corrective maintenance costs of the machine. By multiplying the outcome with the costs of corrective maintenance, the real currency figure can be obtained as shown in formula 4-5.

( 4-5 )

Since the time since the last maintenance action, , , and the costs of corrective maintenance differs for every task and for every moment in time. Therefore, the most profitable task to execute will change constantly.

In a rule based expert system, the profit for every maintenance task which satifies the hard requirements are calculated and ranked in descending order. The highest ranked maintenance task is recommended to perform in the current time window. However, all tasks with a positive profit will be displayed as they all contribute to lower overall maintenance costs.

(20)

If down, do. Selecting opportunistic maintenance tasks during unexpected downtime. 20 4.2.2. Different value for c

The cost factor variable ‘c’ which is calculated in equation 2-3 and 4-2 and used in equation 2-2 and 4-2 is the cost ratio between preventive maintenance and corrective maintenance or opportunistic maintenance and corrective maintenance respectively. This means that the factor will have two values, one for during preventive maintenance (equation 2-2) and one for during OM (equation 4-2). The factor ‘c’ will always be between 0 and 1 because corrective maintenance always has higher costs than preventive or opportunistic maintenance for a certain task because corrective maintenance has the same and more cost factors than preventive and opportunistic maintenance. All factors associated with the different types of maintenance are listed in table 4-1 and will be discussed in the remaining of this section, including their influence on OM.

Corrective maintenance

costs > preventive maintenance costs > Opportunistic maintenance costs

• Costs of parts • Costs of parts • Costs of parts

• Costs of labor • Costs of labor • Costs of labor

• Costs of lost production time • Costs of lost production time

• Costs of perished goods

• Costs of unpreparedness

Tabel 4-1: Cost factor different maintenance types

Costs of parts are costs of replacement part(s), or the costs of repairing a part. Costs of labor are wages paid to employees during their work on a maintenance task.

The cost of lost production time is depending on the length of the downtime and the costs of lost profit per unit time. These costs are not a factor for OM as the machine is already down and these costs are calculated towards the preventive or corrective maintenance task. If the costs of lost production time is relative high compared to labor and parts costs, OM is more often profitable.

The costs of perished goods arises because corrective maintenance is unplanned and the machine might be in the middle of production. The cost of perished goods at FrieslandCampina Bedum are relative high since the cheese will perish if it stays in production too long.

Cost of unpreparedness are all costs resulting because corrective maintenance is not as efficient as preventive maintenance. This is because the breakdown has to be diagnosed first, parts have to be gathered when the machine is down and employees are not always present.

(21)

5. Decision support system

To quickly and accurately handle the task selecting and task prioritizing, a database is designed with a Structured Query Language (SQL) based report. In this chapter, the database structure and SQL query will be explained. Since FrieslandCampina Bedum already uses Microsoft Access for their (maintenance) databases, this program is also used for this decision support system. First, the database structure will be discussed. Then the decision support system, and how to interact with the decision support system is explained and finally the query will be explained.

5.1. Database structure

The database needed for the task selecting and prioritizing is created and displayed in figure 5-1. In this database, all PM tasks can be added to the PM_Jobs table. A name, description, all required skill levels and whether the machine needs to be empty for a certain task can be added. The ‘Time_in_minutes’ attribute is the total time in minutes an A level employee needs for the specific task. Since maintenance jobs may need multiple parts, but parts can also be needed in multiple jobs, a bridge-table is used. In the parts table, all needed parts can be added and the current available stock level can be displayed in the ‘PartQuantity’ attribute.

All employees can be added to the ‘Employees’ table with their name, skill levels and work speed factor. The ‘company’ attribute is used to connect the employees with PM tasks during task selecting.

-Jobid -JobName -JobDescription -SkillRequired1 -SkillRequired2 -SkillRequired3 -SkillRequired4 -SkillRequired5 -Company -Time_in_minutes -Clean_needed -Downtime_if_break_min -Last_maintenance -Optimal_age_based PM_Jobs -PartID -PartName -PartDescription -PartQuantity -Price Parts -ID -JobID -PartID -Quantity PM_Parts_Bridge -EmployeeID -EmployeeName -EmployeeSkill1 -EmployeeSkill2 -EmployeeSkill3 -EmployeeSkill4 -EmployeeSkill5 -Company -Factor Employees -Jobid 1 -Jobid * -PartID * -PartID 1

(22)

5.2. Interacting with the decision support system.

Working with the decision support system can be divided in three phases. The first phase is before any breakdown occurs and can be briefly described as keeping the system up to date. The second phase is during breakdown in which the decision on which opportunistic maintenance to do is taken. The last phase is after breakdown in which the system is updated.

In the next sections the decision support system will be discussed on a step-by-step basis for both the user interface as the backend off the system. This step-by-step analysis is divided into the phases before, during and after a breakdown occurs.

5.2.1. Before the breakdown

The whole decision support system needs to be set-up. This should be done by inputting the right data in the database tables. The database consists of four tables which need to be filled with company specific data. All the tables with the needed data for each field will be discussed.

Figure 5-2: Employee table of decision support system

EmployeeID is the unique key value of the employees table. When entering any other variable on a new line, a EmployeeID will be assigned to that line. After that, all variables have to be added, the description of the variables are:

• EmployeeName: The name of the maintenance engineer.

• EmployeeSkill1-5: Skill level off the employee on a certain field. In this example labeled skill 1 true 5, which can represent skills like mechanical, electric and

programming for example.

• Company: The company the employee works for.

(23)

If down, do. Selecting opportunistic maintenance tasks during unexpected downtime. 23 Figure 5-3: PM_Jobs table of decision support system

JobID is the unique key value of the PM_Jobs table. When entering any other variable on a new line, a JobID will be assigned to that line. After that, all variables have to be added, the description of the variables are:

• JobName: Name of the job.

• JobDescription: Description of the job.

• Skillrequired1-5: Required skill level on a certain field to complete the job. In this example labeled skill 1 true 5, which can represent skills like mechanical, electric and programming for example.

• Company: The company the job is from.

• Time_in_minutes: Time it takes a employee to finish the job at standard speed. • Clean_needed Binairy variable if a job needs a clean machine (not full with semi-

finished products) in which 1 means a clean system is needed and 0 means is does not matter.

• Downtime_if_break_min The time it takes to fix the part if it breaks down. This variable Is used to calculate the costs of corrective maintenance.

• Last_maintenance The date of the last maintenance action on which the part is returned to a as-good-as-new status.

(24)

If down, do. Selecting opportunistic maintenance tasks during unexpected downtime. 24 Figure 5-4: Parts table of decision support system

The parts table is used to track the parts on stock. The data in the Parts table should be up to date constantly. PartID is the unique key value of the Parts table. When entering any other variable on a new line, a PartID will be assigned to that line. After that, all variables have to be added, the description of the variables are:

• PartName The name of the part.

• PartDescription A short description of the part.

• PartQuantity The current quantity of the part on stock.

• Price The price of the part, this will be used to calculate to cost of the maintenance jobs.

Figure 5-5: PM_Parts_Bridge table of decision support system

The PM_Parts_Bridge table is used for connecting parts with jobs. Since each job can have multiple parts, and each parts can be in multiple jobs a bridge table is needed. This also opens the possibility for higher quantities of a part on a certain job. Id is the unique key value of the PM_Parts_Bridge table. When entering any other variable on a new line, a Id will be assigned to that line. The other variables are:

• JobID The unique JobID to which a part is to be assigned.

(25)

If down, do. Selecting opportunistic maintenance tasks during unexpected downtime. 25 5.2.2. During breakdown

Opening the ‘job selection ordering’ report will start the decision process of which opportunistic maintenance task to perform. The query ‘job selection’, which can be found in appendix C, will fill the report and since it has multiple parameters which are unknown to the computer, the computer will ask the user to input these variables.

The first parameter the user is asked to put in is which employee is present. The user should input the “employeeid” of the available employee.

Figure 5-6: Parameter state of the factory

The next parameter the user is asked to input is the current state of the factory, in this case, if the pipes are clean. This is answered with a binary code, in which 1 means the pipes are clean and 0 means the pipes are not clean.

Figure 5-7: Parameter employee present

The last parameter is the estimated downtime. The time input should be in the same time unit as with which the job table is filled.

(26)

If down, do. Selecting opportunistic maintenance tasks during unexpected downtime. 26 Once these three parameters are added, the query will run and will rank all possible jobs in descending order of expected net profit. Since the job selecting uses the parameters, the date and the part stock levels, the result might be different every time. The result will show time and expected net profit for each job, so the user can make an educated decision on which jobs to plan in the current time window.

Figure 5-9: Results on the decision support system 5.2.3. After breakdown

(27)

5.3. SQL query

The SQL query used to fill the Microsoft Access report is displayed in appendix C. This query selects all tasks which satisfy all hard requirements and ranks them based on their estimated profit with the formulas which are described in section 4.2.

The query consists of two sub queries. The first sub query will check if parts are on stock in the needed quantity and will calculate the price of all parts which are needed for every task. This sub query will output for every jobid the partids it contains, and the quantity of that part if it has more on stock then needed for the specific part. It will also output the price per part type.

The second sub query will get all preventive maintenance task info and checks if the task can be done within the time limit, with the present employee(s) and in the current state of the factory. This sub query takes the user input variables employee present, estimated downtime and state of the factory. This sub query will output all job information and employee information and will only output information of a job if a job is feasible based on available employee skill, available time and the state of the factory.

In the main query, both sub queries are combined and the number of part types needed are compared with the number of part types which are fully on stock from both sub query’s. If both numbers are the same, all constraints are met and the job can be executed. The needed info for selecting tasks is selected so this can be displayed in the report. Tasks are ranked on their estimated profit and only those tasks with a positive profit are displayed.

(28)

6. Discussion

The research question this research tried to answer was: “How to determine quickly which preventive maintenance tasks to carry out during (un)expected downtime in a continuous multi series production environment?” With the proposed rule based expert system, a quick selection of tasks can be made and ordered with the proposed task prioritizing policy.

In the remainder of this section, the findings on task selection and task prioritizing of opportunistic maintenance tasks will be critically discussed.

6.1. Task selecting

This research shows that with the rule based expert system and up to date databases, the task selecting can be done in a quick and accurate way. The variables which are used to select the maintenance tasks are parts and tools on stock, state of the plant, time available and skill level of the present employee(s). A critical remark can be made that different companies have different hard requirements which are not researched within this task selecting.

Furthermore, task selection criteria are currently assumed to be a fixed fact. Although this is true for the parts on stock as a part is on stock or it is not, time available displayed as a fixed number can be discussed. Time available and time needed for maintenance tasks are input factors which in the real world would might not have a fixed time, but a probability distribution. Further research should be done on how the proposed system can take variable times in consideration when selecting tasks.

The employee working speed factor used in the calculation of time might be not as strict as assumed in the current system as a skilled worker can have more benefit on some tasks compared to others. Further research should be done on how to make the expected time a worker needs for a task can be calculated more accurate. The current system cannot handle the workforce constraint of multiple people working on a single job, the model can be expanded by adding this opportunity to it.

The last variable, current state of the plant can be expanded as it might be possible to bring the factory in the state needed for opportunistic maintenance and perform this task within the available time limit.

6.2. Task prioritizing

(29)

7. Conclusion

This study found a solution for selecting maintenance tasks and prioritizing them during (un)expected downtime. To quickly and accurately select possible maintenance tasks, a rule based expert system is proposed. Since a computer can process large amounts of data with no errors, it is suitable for this task. The proposed rule based expert system consists of two phases, the first phase is task selecting and the second phase is task prioritizing. In the task selecting phase, all tasks which are possible to do in the current time window are selected. This is done by checking all possible tasks against different input variables. The tasks are selected based on the availability of parts, present employees and their skill-level, time available in the current time window and the state of the factory.

In the second phase, all maintenance tasks which are possible to do in the current time window are ranked based on their potential cost savings compared to their normal costs per unit time if a optimal age based maintenance policy is used.

The proposed rule based expert system can be implemented with databases and structured query language (SQL). For FrieslandCampina Bedum, a Microsoft Access database and SQL report is developed which can be ran by an employee during breakdown. The report will give a list of possible tasks to do prioritized on their potential costs savings in a split second. It is flexible enough to allow additional input variables to be added when they are needed.

Further research should be conducted on how variance in task duration and time available influence the task selecting and prioritizing system.

The research is conducted at FrieslandCampina Bedum, a cheese producing factory in the Netherlands. All input data and variables are based on their factory.

(30)

References

Barlow, R., & Hunter, L. (1960). Optimum preventive maintenance policies. Operations

Research, 8, 90–100.

Berg, M. (1978). General trigger-off replacement procedures for two-unit systems. Naval

Research Logistics Quarterly, 25(1), 15–29.

Dekker, R., Wildeman, R. E., & Duyn Schouten, F. A. (1997). A review of multi-component

maintenance models with economic dependence. Mathematical Methods of Operations

Research, 45(3), 411–435.

Giachetti, R. (1998). A decision support system for material and manufacturing process selection.

Journal of Intelligent Manufacturing, 9, 265–276.

Gopalakrishnan, M. (1997). Maximizing the effectiveness of a preventive maintenance system:

an adaptive modeling approach. Management …, 43 (6)(March 2014), 827–840.

Laggoune, R., Chateauneuf, A., & Aissani, D. (2009). Opportunistic policy for optimal

preventive maintenance of a multi-component system in continuous operating units.

Computers & Chemical Engineering, 33(9), 1499–1510.

Negnevitsky, M. (2001). Artificial Intelligence: A Guide to Intelligent Systems (1st ed.). Boston,

MA, USA: Addison-Wesley Longman Publishing Co., Inc.

Peffers, K., Tuunanen, T., Rothenberger, M. a., & Chatterjee, S. (2007). A Design Science

Research Methodology for Information Systems Research. Journal of Management

Information Systems, 24(3), 45–77.

Veldman, J., Wortmann, H., & Klingenberg, W. (2011). Typology of condition based

maintenance. Journal of Quality in Maintenance Engineering, 17(2), 183–202.

Wang, H. (2002). A survey of maintenance policies of deteriorating systems. European Journal

of Operational Research, 139(3), 469–489.

Wieringa, R. (2009). Design science as nested problem solving. Proceedings of the 4th

International Conference on Design Science Research in Information Systems and

Technology - DESRIST ’09, 1.

Zhou, X., Lu, Z., Xi, L., & Lee, J. (2010). Opportunistic preventive maintenance optimization for

multi-unit series systems with combing multi-preventive maintenance techniques. Journal of

Shanghai Jiaotong University (Science), 15(5), 513–518.

(31)

(32)

APPENDIX B

These three graphs show the average cost per unit time of three different distributions. The normal, gamma and Weibull distribution are shown.

Normal distribution

Mean time between failure: 3 units time

Standard deviation: 1 unit time

(33)

Gamma distribution

Alfa: 5

Beta: 1

(34)

Weibull Distribution

Alfa: 2

Beta: 6

(35)

Appendix C

SELECT partonstock.jobid AS Jobid ,partneeded.jobname AS Job_name ,partneeded.jobdescription AS Description

,partneeded.Time_in_minutes * partneeded.factor AS time_in_minutes

,partneeded.employeeid AS employee_id ,partneeded.clean_needed AS clean_pipes_needed ,format((((partneeded.Optimal_age_based)-(((partonstock.price)/((partneeded.Downtime_if_break_min * 100 ) + partonstock.price ) ) / partneeded.time_since_last_maintenance) ) * partneeded.time_since_last_maintenance)*((partneeded.Downtime_if_break_min * 100) + partonstock.price),"currency") AS profit FROM (Select PM_Parts_Bridge.jobid AS jobid

,min(PM_Parts_Bridge.partid) AS partid

,count(PM_Parts_Bridge.Id) AS tel

,sum(parts.price*pm_parts_bridge.quantity) AS price

FROM Parts

INNER JOIN PM_Parts_Bridge

ON Parts.PartID = PM_Parts_Bridge.PartID Where Parts.PartQuantity >= PM_Parts_Bridge.quantity Group by PM_Parts_Bridge.jobid ) AS partonstock INNER JOIN (Select pm_jobs.jobid AS Jobid

,count(PM_Parts_Bridge.Id) AS tel

,max(pm_jobs.company) AS company

,max(pm_jobs.Downtime_if_break_min) AS Downtime_if_break_min ,max(employees.employeeid) AS employeeid

(36)

If down, do. Selecting opportunistic maintenance tasks during unexpected downtime. 36 ,max(pm_jobs.Optimal_age_based) AS Optimal_age_based

,max(employees.factor) AS factor

,format(max(pm_jobs.time_in_minutes),"fixed") AS Time_in_minutes

,max(PM_Jobs.JobName) AS jobname

,max(pm_jobs.jobdescription) AS jobdescription ,max(pm_jobs.Clean_needed) AS Clean_needed

FROM (pm_jobs

INNER JOIN PM_Parts_Bridge

ON pm_jobs.jobID = PM_Parts_Bridge.jobID)

INNER JOIN Employees

on pm_jobs.company = Employees.company

Where employees.employeeid = employee.present

AND pm_jobs.Skillrequired1 <= employees.EmployeeSkill1

AND (pm_jobs.Time_in_minutes * employees.factor) <= Estimated_Downtime

AND pm_jobs.Clean_needed <= Are_pipes_clean

Group by pm_jobs.jobid

) AS partneeded ON partneeded.jobid = partonstock.jobid

WHERE

partonstock.tel = partneeded.tel

AND (((partneeded.Optimal_age_based)-(((partonstock.price)/((partneeded.Downtime_if_break_min * 100) + partonstock.price ) ) / partneeded . time_since_last_maintenance ) ) * partneeded.time_since_last_maintenance)*((partneeded.Downtime_if_break_min * 100) + partonstock.price) => 0

AND Estimated_Downtime - (partneeded.Time_in_minutes * partneeded.factor) >= (0)

Order by

(((partneeded.Optimal_age_based)-(((partonstock.price)/((partneeded.Downtime_if_break_min * 100) + partonstock.price))/partneeded.time_since_last_maintenance))*