FLOW SIMULATION OF AN OUTPUT SHAFT LINE:

(1)

Master thesis, 30 ECTS

FLOW SIMULATION OF AN OUTPUT SHAFT LINE:

Capacity analysis and optimization of the overall

process efficiency

Paulina Sjöholm & Hannah Åström

(2)

This Master Thesis was conducted at the Scania plant in S¨odert¨alje during spring 2019. This project concludes the Master of Science in Industrial Engineering and Management program at Ume˚a Uni- versity.

Firstly, we would like to thank Ulf Bjarre for believing in us from the very beginning, giving us the opportunity to conduct this thesis at the Transmissions Manufacturing department. His dedi- cation, knowledge and guidance has been vital for this project’s execution.

Further, we would like to thank our additional supervisors at Scania, Sofia Oskarsson and Jens Arnholm, for their tireless support and answers to our never ending questions. Moreover, we would like to give a special thank you to P¨ar M˚artensson, who without obligation took us under his guidance.

We would also like to thank our supervisor at Ume˚a University, Robert Johansson, for his end- less support and exceptional advice. His expertise and ability to see problems in new perspectives has been immensely valuable to this project’s success.

Last, but not least, we would like to thank everyone at the DX department for their contribu- tions. We would also want to give an extra thanks to the DXTAB division, who always made us feel like part of the team. We will look back on this semester with joy and try to remember all the valuable lessons you taught us.

Ume˚a 2019-05-28

Paulina Sj¨oholm & Hannah ˚Astr¨om

(3)

Scania CV AB is a world leading provider of sustainable transport solutions. This includes trucks and buses, as well as an extensive offering of product-related services. This project takes place at the S¨odert¨alje plant and is carried out under the Transmissions Manufacturing department, with the output shaft line as the main focus. Due to a higher scale of product demand, the department aims to extend the line’s production capacity to include both additional components and a higher production volume.

The line under investigation consists of eleven serial processes. The project sets out to investigate the line’s dynamic, find a theoretical maximum capacity for different product types and derive a reasonable OPE goal. Moreover, the project’s final objective refers to how un-utilized capacity can be revealed.

The project’s results are delivered in three distinct parts. Firstly, a complex and thorough simulation model is delivered to the department alongside usage instructions. The model in its entirety is found in appendix A3. It is crucial to ensure that the model describes the real system. There- fore, the second result is an extensive model verification and validation. Lastly, results to where un-utilized capacity can be found is presented.

General findings are that, reducing the cycle times to the times stated in the buy-in contract alongside with reducing the length (but not frequency) of shutdowns gives a considerable capacity enhancement. All further optimization endeavours assumes that the line’s cycle times have already been reduced to the purchased times. Moreover, it is possible to increase product type one’s manufacturing capacity by 11,4%, product type two’s capacity by 13,1% and product types three and four’s capacity with 7,2%. The capacity enhancements depends on several parameters for the different product types.

(4)

Scania CV AB är världsledande leverantörer av h˚allbara transportlösningar. Dessa involverar last- bilar och bussar, tillsammans med ett stort utbud av produktrelaterade tjänster. Detta projekt har genomförts p˚a Scanias produktionsenhet i Södertälje, närmare bestämt p˚a Transmissionsbearbet- ningen, med fokus p˚a bearbetningslinan för utg˚aende axel. P˚a grund av högre produktefterfr˚agan

¨

onskar avdelningen att ut¨oka linans kapacitet genom att tillverka fler produktertyper samt ¨oka pro- duktionsvolymen.

Bearbetningslinan som undersöks best˚ar av elva stycken seriella processer. Projektet önskar att undersöka linans dynamik, hitta ett teoretiskt maximum för de olika produkttyperna samt presen- tera ett rimligt OPE m˚al för dessa. Slutligen önskar projektet att undersöka hur outnyttjad kapacitet kan ˚aterfinnas.

Projektets resultat presenteras i tre olika delar. Först överlämnas simuleringsmodellen till avdelningen tillsammans med en grundlig genomg˚ang av uppbyggnad och funktion. Modellen i dess helhet

˚aterfinns i appendix A3. Det är avgörande att säkerställa modellen beskriver det verkliga systemet.

Därför inneh˚aller den andra delen av resultatet en omfattande verifiering och validering. Slutligen presenteras resultat rörande var outnyttjad kapacitet kan ˚aterfinnas.

Projektets övergripande fynd är att en reducering av cykeltiderna till maskinernas inköpta cykeltider tillsammans med en reducering i maskinstoppens längd (inte frekvens) skulle öka linans kapacitet avsevärt. Alla vidare beskrivna optimeringsförsök förutsätter att linans cykeltider redan reducerats till de inköpta tiderna. I och med detta kan kapaciteten för produkttyp ett ökas med totalt 11,4%, produkttyp tv˚a med 13,1% och produkttyperna tre och fyra med 7,9%. kapacitetsökningarna beror p˚a flertalet skilda parametrar för de olika produkttyperna.

(5)

Generation 1 - Product type one & Product type two Generation 2 - Product type three & Product type four

Process - The performance of one or several operations on a product Machining - A process which cuts a piece of raw material into a certain shape Blanks - The denotation used for raw materials used at Scania

Line - A number of interlinked processes with the aim to manufacture a product

UGA - The name of the line producing output shafts

OPE - Overall Process Efficiency. Describes how well an investment is used, when intended to be used

Capacity - The amount of products possible to produce given certain circumstances Output - The line’s throughput, mostly mentioned in weeks

Machine Availability - The percentage of time that a machine is available for usage

Downtime - Any time a machine is not processing

Shutdown - Time a machine is not processing due to an unexpected stop

WIP - Work-In-Process. A measure of how many products are at work in a particular station

SPC - Single-Part-Control. Is performed on a given interval, to ensure quality requirements

AU - Remedial maintenance performed by technicians

FU - Precautionary maintenance performed by technicians to reduce downtime

UFO - Maintenance tasks performed daily by operators

PUS - Production monitoring system. Downtime recorded manually by operators

MTBF - Mean Time Between Failues

(6)

1 Introduction

In this section a brief introduction to the company is given, as well as a background to the problem.

Alongside this the purpose and objectives for this project are defined, and the problem statements presented.

1.1 About the company

Vagnsfabriks Aktiebolaget (Vabis) is established in December 1891. Its main products are open goods wagons, baggage carriers and horse-drawn trams. In 1911 Vabis merges with Maskinfabrik- saktiebolaget Scania to form Scania-Vabis. The Head office is now located in S¨odert¨alje, and so is the developement and production of engines, cars and light goods vehicles. In 1969 Scania-Vabis merges with Saab, and the new brand name SCANIA is introduced on trucks and buses. In 1995 Scania again becomes an independent company, as Saab-Scania is divided into two companies (Scania CV AB, 2019).

Over the years Scania CV AB has become a world leading provider of sustainable transport solutions. This includes trucks and buses, as well as an extensive offering of product-related services.

Moreover, Scania CV AB (hereinafter Scania) is also a leading provider of industrial and marine engines. Furthermore, Scania offers insurance and rental services alongside vehicle financing to enable their customers to focus on their core business. The main focus for the plant located in S¨odert¨alje (Sweden), where this project is carried out, is truck manufacturing. This includes all processes from procurement of raw materials and operations of modifying these into the needed components, to assembly and test driving of completed trucks (Scania CV AB, 2019).

Scania has about 50 000 employees, stationed in 100 different countries. The company’s production takes place in Europe, Latin America and Asia and at these sites global interchange of both complete vehicles and components take place. In addition, there are regional production centres in Africa, Asia and Eurasia (Scania CV AB, 2019).

Scania is a member of TRATON GROUP, which is a subsidiary of Volkswagen and a leading commercial vehicle manufacturer world wide, together with MAN, Volkswagen commercial vehicles and RIO. (Volkswagen AG, 2019). The brands aim is to turn TRATON, and all its members, into Global Champions (Scania CV AB, 2019).

(9)

1.2 About the department

The transmission manufacturing department, DX, at the S¨odert¨alje plant manufactures components to the gearbox. The department is broken down into several branches, and this project is carried out under the DXTAB branch. The branch consists of production engineers and process planners in charge of manufacturing bevel gears and output shafts. This project’s focal point is the output shaft’s production line. Figure 1 displays an example of the line’s output products.

1.3 Problem background

To meet a higher demand Scania started a search for an additional supplier for output shafts. After some deliberation they realised that the new suppliers could not achieve sufficient quality requirements and decided to insource the production of this component instead. It fell on the Transmissions department at the S¨odert¨alje plant to complete the production line (Arnholm, 2019).

In general, Scania does not like to rush implementation of processes, but in this case they had to act quickly to make the deadline. They achieved the deadline, and produced the amount required, but not the amount intended. However, for today’s demand with MAN using Scania’s gearbox this capacity will not be enough. Today, two of the five types of output shafts are manufactured here in Södertälje, and the other three are manufactured by a supplier in Sk˚ane. Due to a higher scale of product demand, and to secure delivery as to not be dependent on only one supplier, Scania Södertälje aims to extend their production capacity to include both some additional components and a higher production volume (Arnholm, 2019).

To this day, the line consists of seven serial processes. The complete line, consisting of eleven serial processes, is set to be in full use during summer 2019. To reach its intended production capacity, the line must be optimized. This capacity is affected by a number of variables such as the size and placement of buffers, bottleneck processes, production plans etc (Bjarre, 2018).

Figure 1: Output Shaft

(10)

1.4 Production line

The production line will consist of eleven serial machining processes. Figure 2 shows an overall representation of the output shaft line. The line starts at the input conveyor (1), where the operator manually loads the blanks. The part is transported by a gantry to one of the two turning machines (2)(3). The gantry is an overhead assembly on which carriers are attached, tasked with transporting and load-/unloading parts into machines. All transportation not executed by a conveyor is performed by a gantry. When processed in the turning machines, the part enters the milling machine (4). The next step in the line is the splines making machine (5), before the part arrives in the welding cell(6). This cell includes several processing steps, such as washing, welding, re-heating and cool-down, and is also the last step before induction hardening (7).

Upon exiting the induction hardening, the part is washed (8) and transported on a conveyor (9) to a shot blasting machine (10). This machine removes residue before the part enters the hard turning machine (11). After this process, the part is forwarded to the grinding machine (12). In the following process the part passes through two multi-operational machines (13)(14), these operate simultaneously alongside each other and the part only enters one of them. Before the part ends up on the output conveyor (17) for a last inspection, it is balanced (15), washed (16) and corrosion protected (16).

1. 2. 4. 5. 6. 7. 8.

9.

11. 10.

12.

16. 15.

17.

1. Input Conveyor 2. Turning Machine 1 3. Turning Machine 2 4. Milling Machine 5. Splines Machine 6. Welding Process 7. Induction Hardening 8. Washing Machine 9. Conveyor

10. Shot Blasting

11. Hard Turning Machine 12. Grinding Machine 13. Machining Centre 1 14. Machining Centre 2 15. Balancing Machine 16. Final Wash

17. Output Conveyor

: New area : Processes

: New machine,   not yet in place 3.

14.

13.

Figure 2: Output Shaft Production Line

(11)

1.5 Capacity

The capacity at the S¨odert¨alje plant falls short to manufacture the demanded quantity of output shafts. As a result of this they have to compensate with more production hours to deliver the needed quantity.

Loading onto the input conveyor, tool changes, set-ups, control measurements and final inspections are carried out manually by operators. These operators are responsible for different areas of the line, and knowledge of the tasks of other areas of the line are often limited. These tasks need to be carried out, and are connected to the overall capacity. If an operator ends up with two tool changes at the same time at his/her area, and no other operator knows how to change this tool, the line stops.

Furthermore, the line’s overall capacity and throughput is affected by the machines downtime.

The machines buy-in contracts states that they will have a machine availability of 97-98%. This is not the case, since historical stop data (for some of the line’s machines) exceeds the accepted 2-3%.

Moreover, planned maintenance for the line must also be taken in consideration, which increases the downtime of a machine even further. Scania has a standard OPE goal of 85% of maximum capacity, and the equation for calculating it takes planned maintenance in consideration.

1.6 Purpose and Objectives

The purpose of this project is to map and then optimize the line capacity for output shaft manufacturing. The project results in a simulation model along with a written report containing recommendations for the line.

1.6.1 Problem statements

• What is the theoretical maximum capacity of the line, for all product types?

• With today’s capacity, what is a reasonable OPE goal for daily production?

• By which means can the line’s capacity be increased?

1.7 Delimitations

This project will only examine the new production line, it will not take into account for the previous soft material (pre induction hardening) machining line. The machining processes from the input conveyor up to the welding machine, denoted by the light blue area in Figure 2, is not yet in full production. This is because the machines are new and some of them are still under an installation and test phase. Therefore, the data regarding these processes are estimated from equivalent machines and information provided by the distributors alongside with results from pre-delivery tests and acceptance tests.

Today, the production line manufactures two types of output shafts, however, this project will also incorporate the variables of manufacturing the next generation of output shafts. This is because the next generation of products will be introduced to the market and manufactured at the S¨odert¨alje plant in a near future. The delimitation in the model regarding the new generation concerns limited data availability.

(12)

2 Theory

Theories relevant for the project are discussed in this section. It covers the discrete event modelling and optimization approach as well as terminology related to the areas.

2.1 Simulation

Simulation can be defined as the act of imitation and the simulation process involves designing a model of a system and carrying out experiments on it as it progresses over time. The outputs from the simulation model will be reflective of the behaviour of the real system, if the model is valid.

Therefore, a simulation model can be used both as a design tool to predict the performance of new systems, as well as an analytical tool to predict the impact of potential changes in an existing system (Extendsim, 2017; Banks, 2014).

2.1.1 Advantages and Disadvantages of Simulation

One of the main advantages of simulation is that it enables experimentation in a complex system without interfering with ongoing operations. Some systems are so complex that only a dynamic model will give insight of operations and interactions within the system. A typical example of this would be to try to understand how bottlenecks in manufacturing occur, and when they occur, with regard to different parts. This may be impossible to study by stopping it or examining an individual component in isolation. Simulations also allows for testing of new techniques or strategies without requiring resources for the implementation. The cost of modelling a non-existing system can be very small compared to the investment involving for example, an installation of a new assembly line. Another advantage of simulation is that knowledge, both regarding the interaction between variables and the importance of variables linked to the system’s performance, can be obtained by changing the simulation input and observing the resulting output. The information obtained from simulations can be used to reduce risk and uncertainty in order for a company or an organisation to make informed decisions (ExtendSim, 2017; Banks, 2014; Chung, 2004).

There are also some disadvantages to take into consideration. The first, and foremost disadvantage, can be described by the statement ”garbage in, garbage out”, which means that no matter how well a simulation model is developed, if the data input is not accurate the subsequent output will not be accurate. Hence the data collection is the most crucial, and also difficult part of the simulation process. Moreover, building a simulation model can be both expensive and time consuming.

However, creating a simulation model while skimping on resources often results in an insufficient model. It is therefore important to make sure that simulations are necessary for the given problem.

Another disadvantage of simulation is that the construction of a simulation model hinges on the person building it and their competences and preferences. For example, if two people with slightly different experiences build an individual simulation model of the same system, it is very unlikely that the two models will end up being identical. This can lead to contradictions, which also illustrates to the fact that simulation results often are hard to interpret (Banks, 2014; Chung, 2004).

(13)

2.1.2 Discrete Event Simulation

Discrete Event Simulation is a type of simulation where the system changes state only as events occur, time passing has no direct impact on the model. Discrete event systems are structured to combine elements and resources to imitate a real system. Such examples of elements includes people, procedures, materials as well as resources like equipment, tools and personnel. Each process consists of a series of related activities carried out to achieve a specified outcome. These activities are performed with a certain duration, and usually involves usage of both process elements and resources. Processes are organised around events, such as the receipt of parts or a request for service.

These evens occur at a random, but somewhat predicable, interval. Discrete event simulations allow organisations and companies to examine their fundamental processes from a cross-functional per- spective and ask questions about their organization like “Why?” and “What if?” (ExtendSim, 2017).

An example of a discrete event system is an assembly line in a factory. The individual entities (in this case parts) are assembled based on different events (planned orders, different part specific operations) (ExtendSim, 2017).

2.2 Bottleneck identification

The performance of a production line is often evaluated by the level of throughput. A machine that impedes a system’s performance in the strongest manner, i.e. decreases the throughput, is called a bottleneck. If the location of a bottleneck is accurately identified utilization of finite manufacturing resources can be improved, total cost of production can be reduced, and the system throughput can be increased. One way to identify a bottleneck is to compare two neighbouring machines: if the starvation time of the downstream machine is lower than the blockage time for the upstream machine, the bottleneck is located downstream; otherwise upstream. Here starvation time refers to the time a machine spends waiting for a part to process, hereinafter denoted by idling time.

Blocked time refers to the time a machine is unable to process parts due to the succeeding machine being occupied. In previous work, there are generally two ways of detecting bottlenecks; through an analytical model or through a simulation model (Li et al., 2007; Chiang, Kuo, Meerkov, 2001).

Even though the notion “bottleneck” is widely discussed and used, Li et al claims that there does not seem to exist an interpretation that is uniformly accepted. However, the following definitions can often be found in literature regarding the subject:

• A machine is defined as a bottleneck if it has the lowest isolation production rate compared to the other machines in the system. Production rate is here denoted as the average number of parts produced by a machine per unit of time.

• The machine right after the buffer with the largest work-in-process (WIP) inventory is defined as a bottleneck.

• A machine is denoted a bottleneck if the sensitivity value of the system’s production rate to the machine’s PR is the largest within the line.

If a performance improvement is applied to a bottleneck machine, this would result in a greater system throughput than if this improvement was applied to a non-bottleneck process. If after such an improvement a bottleneck’s process time is reduced to an extent that another machine now has a lager affect on the system, a new bottleneck is found (Li et al., 2007).

(14)

2.3 Overall Process Efficiency

Overall Process Efficiency (OPE) is a frequently used key performance indicator at Scania. It describes how well an investment is used, when it is intended to be used. The equation is defined as

OP E = Produced amount · Cycle time

Planned production time (1)

where the produced amount corresponds to the number of approved articles in the finished stock within the planned production time. The cycle time is denoted by the line’s controlling cycle time, i.e. the greatest cycle time for all machines in the line. The planned production takes predetermined stops into consideration and includes stops like weekends, summer leave, planned maintenance etc.

The output from Equation 1 is an amount, but the OPE performance indicator is referred to in percent. To arrive at this, the equation output is divided by the absolute maximum capacity of the line. This capacity is obtained by dividing the line’s opening hours by the controlling cycle time, resulting in the output amount if the line had no downtime whatsoever.

2.4 Kolmogorov - Smirnov test

The two-sample Kolmogorov-Smirnov (hereinafter K-S) test is a goodness of fit test which investigates whether two probability distributions differ. It does not determine what this distribution is, only if the two independent sets are derived from populations with the same distribution (or the same population). This test is sensitive in both location and shape of the sample’s empirical cumulative distribution function, and is therefore one of the most useful nonparametric methods to compare samples.

Before applying the K-S test the sample’s cumulative frequency distributions is must be determined. The sample’s empirical distribution function is used to calculate the cumulative frequency distribution. For this, the same interval is used for both samples. The next step is to, for each interval, subtract one step function from the other. The largest of these deviations is the focal point for the K-S test.

Let the observed cumulative distribution for one sample be denoted by Sm(X) = ^K_m, where K is the number of data points equal or less than X and m is the sample size. Moreover, let the other sample’s observed cumulative distribution function be Sn(X) = ^K_n, where n is the size of this sample. The two-tailed K-S test has its basis in the following hypotheses;

(H0: There is no significant difference between the two samples distributions

H₁: There is a significant difference, and the samples comes from different distributions The K-S statistic for a two-tailed test is;

D_m,n= max[S_m(X) − S_n(X)] (2)

For the two-tailed test the maximum absolute difference in all directions,D_m,n, is found. The corresponding P-value is compared to a P-value table and the null hypothesis (H₀) rejected if the computed P-value is lower than the one found in the table at a certain level of confidence (Lin, Wu, watada, 2010).

(15)

3 Method

This section describes the methods used throughout the project in order to reach the set objectives.

Banks et. al describes a set of necessary steps for building simulation models. See Figure 3 (Banks et. al).

Figure 3: Steps in a Simulation Study

(16)

The steps mentioned have been translated to fit this project and are presented below.

1. The problem formulation is defined together with representatives from Scania. It is important that those experiencing the problem and the analyst assigned to the problem understand and agree with each other regarding the formulation of the problem.

2. The project plan and the stated objectives indicate the questions to be answered by simulation.

This step includes investigation into whether the problem is simulation-worthy or not, and by which means the problem could be solved.

3. It is important to have a clear concept of the system to be modelled, before the modelling starts. In the model conceptualisation the essential features of the system are selected and necessary assumptions that characterise the system are identified.

4. The data collection is an iterative process, carried out simultaneously to the construction of the model. For each step of the modelling process necessary data is collected and compiled.

5. A simulation software, ExtendSim, is chosen for this project. The model concepts and compiled data is translated and inserted into the model.

6. Throughout the translation process incremental verifications are carried out. This is to ensure that the model works as intended.

7. Validation is an iterative process where the actual system output in the form of historical data is compared to the model output. If the output is not within a desired interval, steps three through seven, are repeated until an accepted result is obtained.

8. The experimental design is necessary in order to distinguish the studies to be carried out.

These alternatives must be determined and are formed together with representatives from Scania.

9. Model outputs are analysed after each experiment. Results from this analysis is used to provide the company with suggestions for improvements.

10. Dependent upon the results from the previous step, more experiments and runs could be necessary.

11. Documentation of both the structure of the model as well as the obtained results are essential.

This project sets out to attain a standardised way of developing future simulation models, therefore a thorough and comprehensible description is crucial.

12. The final step lies beyond the framework of this project. Implementation suggestions shall follow from the analysis.

3.1 ExtendSim Simulation Software

ExtendSim (previously known as Extend) is a simulation software developed by Imagine That Inc.

The first version of the software was released in 1988 (ExtendSim, 2019). ExtendSim incorporates three different modelling methodologies, these include discrete rate- , continuous- and discrete event modelling (ExtendSim, 2017). The methodology for this project is the latter. More information about simulation and discrete event simulation can be found in section 2.1.

ExtendSim is Scania’s default simulation software and for this reason support and guidance for the project can be found at the company. Because of this, ExtendSim is the software chosen for the execution of this project.

(17)

3.1.1 Components in ExtendSim

A simulation model can be created by adding several different blocks from the software’s library.

The most frequently used blocks and its associated variables are explained in Table 1 below.

Table 1: Blocks in ExtendSim

Blocks Name Description

Create This block creates items for the model. The number of different items can be created in various ways, these include randomly, infinite and by schedule.

Set The Set block assigns properties to items. It is possible to assign each item multiple properties, such as attributes, quantity and priority. These properties can be collected from a database.

Get The Get block identifies and displays item’s properties.

Since an item can have several properties it is possible to specify which one the block should display.

Activity The Activity block processes a chosen amount of items for a certain period of time. The processing time can be constant or varying depending on an item’s property.

Shutdown This block provides other blocks in the model with shutdowns. The occurrence and the length of the stops can be either predetermined or random.

Queue Blocks of this type queues and releases items with user preferred settings. There are two different kinds of queue behaviours, sorted queues and resource pool queues.

Resource This block creates a number of resources. There may be blocks that require a certain resource to be able to process, if there is no available resource, the block will not be able to proceed.

Shift The shift block provides a schedule of availability, this can be applied to other blocks such as activities or resources etc.

Select Item Out

This block select’s which output an item should use based on user settings. The settings for the selected output includes random, property and sequencial etc.

Select Item In

Similar to the Select Item Out, this block select’s rooting for items depending on user settings.

Exit The exit passes items out of the simulation and keeps track of the number of items exited.

(18)

3.1.2 Activity States

The activity block records five different states;

• Busy - The block’s busy time corresponds to actual machining time for a machine.

• Idle - Idling time represents the time an activity block is waiting for parts to process.

• Blocked - Blocked time describes time a machine is unable to process parts due to the succeeding machine being occupied.

• Shutdown - Refers to the time a machine is not processing due to mechanical/electrical errors.

• Offshift - An activity block can be dependent on a shift, such as for example being available Monday-Friday buy not on the weekends. This would accord for the block to record Saturday and Sunday as offshift time.

These are tracked as a percentage of time and capacity during the simulation run and can be used to investigate how a machine is performing. However, it is important to note that an activity can only record one down signal at a time. For example, if the activity goes off-shift and during this time receives a shutdown, it will only accumulate off-shift time and no shutdown time (ExtendSim, 2017).

3.1.3 Scenario Manager

ExtendSim’s Scenario Manager is a block found in the software’s library which can be added to any model. The Scenario Manager investigates by which means different compositions of factors affect a model’s outcome.

ExtendSim’s userguide describes a number of steps for carrying out a scenario analysis:

1. Enter the value library and add the Scenario Manager block 2. Add the model inputs (factors) to be included in the analysis 3. Add the model results (responses) to be analysed

4. Choose what to include in the report 5. Determine the design of the scenarios 6. Generate and run the Scenario Manager 7. Analyse the result

8. Export the scenarios to Excel

The Scenario Manager compares a number of scenarios, these scenarios incorporate diverse model factors. The user determines which factors should be included in different ”what if” scenarios. These factors include settings for blocks, databases and tables etc.

Before each simulation run, the Scenario Manager copies the model factors to be investigated in that particular scenario. In order to capture stochastic behaviour each scenario is advantageously run several times. The running time for the Scenario Manager can vary between minutes and hours depending on the complexity and size of the model (ExtendSim, 2017).

(19)

3.2 Data

A reliable and accurate simulation model is directly dependent on quality of the input data. As mentioned earlier, the expression ”garbage in, garbage out” applies for all simulation processes, which implies that a model is only as good as the entered inputs. Therefore, it is of utmost importance that the quality and reliability of the provided input data has been thoroughly analysed through each implementation step. Moreover, each category of data sets have been thoroughly assessed and verified together with specialists within the specific fields.

3.2.1 Cycle Times

The cycle times used for the different machines in this project are denoted by, loading time, running time and recovery time. The different machines have different cycle times, these cycle times depend on the extent of operations performed on the part as well as the product type of the part being processed. Worth mentioning is that, as the machines cycle time depends on the product type being processed, the line bottlenecks will vary together with the product type. Figure 4 displays the machines cycle times depending on product type.

Turning Machine 1 Turning Machine 2

Milling Ma chine

Splines Machine Welding Process

Induction Hardening Shot Blasting Machine

Hard Turning Machine Grinding Machine

Machining Centre 1 Machining Centre 2

Balancing Machine Final Wash

The Line's Cycle Times

Product Type 1 Product Type 2 Product Type 3 Product Type 4

Figure 4: Cycle times for the different product types

The majority of the machines on the line lack the ability to record cycle times. So, to examine the processes cycle times they are clocked manually. Thereafter they are compared to the cycle times previously recorded by the department. Manually and recorded times which corresponds are accepted, and for those not corresponding, the manually clocked times were utilized. As for the machines with an option to access cycle time data, it is retrieved and examined to find a mean value. Lastly, for the machines which are not yet in full use, the cycle times agreed upon with the suppliers are used. For this project, the cycle times for product type three and four are the same.

They are created as different products in the model so that when exact cycle times are obtained these can be easily entered.

(20)

3.2.2 Tool Changes

Machines in the line have both a varying number of tools to change, and different intervals at which they should be changed. To get an accurate representation of this in the model, both the technician in charge of the buy-in of the machine as well as the operators handling it daily are questioned.

With a few runs left of a tools lifetime, the operators are signalled by an indicator that a tool change is imminent. If the operator is busy carrying out another task, not noticing the upcoming tool change, the machine will stop and be inaccessible for an unnecessary period of time. Some machines are equipped with a tool magazine, which makes it possible for the operator to carry out tool changes without interfering with production. This is also relevant for the magazine-change though, if the operator does not notice the upcoming tool change in time, the machine will be again inaccessible for an unnecessary period of time.

The tool changes are today arranged into multiples, which means that the intervals of different tools coincide. The reason for this is to both minimize machine downtime, as well as using the operator’s time most efficiently. Every tool change requires the operator to quality-check the first part passing through, to verify that the tool is correctly set. The manufacturing process will not proceed until the quality-check is carried out, and approved. If not approved, additional changes to the tool parameters are made and another quality-check will be performed on the next part passing through the machine.

3.2.3 Single Part Control (SPC)

Every tenth part is selected out from a handful of the processes by the gantry and quality-checked to make sure that the part dimensions meet requirements. This does not require the line to stop, but calls for an operator to perform the task.

3.2.4 Set-up Times

This project concerns four different product types, divided into two different generations. A Set-up between product types of the same generation is less time consuming than that between product types of different generations. The accurate set-up times for today’s generation of products are obtained through the operators. Moreover, since the new generation of gear boxes is still in the development phase, set-up times between the generations are estimated and will be subject to change.

3.2.5 Gantry

The gantry is very complex to build in the simulation software, due to its very extensive priority scheme. Therefore, the simulation model contains a simplification of the gantry process, but does not include it’s priority scheme. This simplification still includes that each gantry can be used in only one machine at the same time, but does not prioritise which machine to visit first. The gantry process contains the transport time of the products between machines, but does not account for loading time, since this is incorporated in the cycle time.

(21)

3.2.6 Maintenance

Maintenance data partly covers downtime for the machines. There are different kinds of data, collected and executed by different personnel, these are described below.

Maintenance Data (AU)

This data is retrieved from the maintenance database at Scania. It covers unforeseen shutdowns, and with the technology of the line today it is impossible to predict when they occur. This maintenance is carried out by trained mechanics, specialised in these kinds of operations.

The first step is to divide the data, and inspect each machine separately. To incorporate shutdowns in the model, the mean time between failures (MTBF) is compiled. This is carried out by investigating both historical maintenance data from 2018 as well as production hours for the line.

The MTBF is obtained by dividing the production hours with the number of stops. To get an accurate representation of this in the model, the distribution for the MTBF is uniform integer, with a minimum of 0 and a maximum of 2 x MTBF. This distribution was chosen after recommendations from a senior engineer at Scania, specializing in simulation (M˚artensson, 2019).

The recorded shutdown is sorted and examined by a histogram. Different ranges, where frequency and the probability for the different values to occur is compiled. To represent the actual shutdown time in the model, an empirical discrete distribution is used. This representation is likewise chosen after recommendations from the same senior engineer. The different ranges of shutdown length and the probability of occurrence is added to the model.

Maintenance for Operators (UFO)

This type of maintenance is planned by maintenance engineers and carried out by operators at given intervals. Every Monday through Thursday at 9.30 operators stop a given machine for about 30 minutes to carry out easier maintenance activities, the scheme can be seen in Table 2. There are instructions for each machine denoting the activities to be carried out, together with which day of the week the machine is stopped. The model takes into account which day and time it is, and stops the machines at given intervals. The machines in the new area did not have an assigned stop day, so these were given a suitable day.

Table 2: Schedule for the operators maintenance

Monday Tuesday Wednesday Thursday Friday Turning Machine 1 x

Turning Machine 2 x

Milling Machine x

Splines Machine x

Welding Machine x

Induction Hardening x

Shot Blasting x

Hard Turning Machine x

Grinding Machine x

Machining Centre 1 x

Machining Centre 2 x

Balancing Machine x

Final Wash x

(22)

Precautionary Maintenance (FU)

Maintenance of this type is carried out to prevent as much unplanned maintenance as possible.

On the output shaft line the precautionary maintenance is divided into two groups, one for each leg of the line. For each of the distinct legs maintenance is carried out once every four weeks, with different tasks for each occasion. The length of these maintenance stops are dependent on the activities carried out, but are at most four hours long. These variations could not be translated into the model, therefore all precautionary maintenance stops are four hours long.

3.2.7 Production Monitoring Report (PUS)

Data of this kind is reported by operators during their shift, or after it ends. It incorporates further shutdowns during the shift not covered by maintenance data. An example of this can be deviations in the part dimensions resulting from a tool change, this causes the line to stop, but does not require a mechanic.

Since these shutdowns are manually documented, and the hours have to add up, some uncertainty within the data exists. One example of this is that the shutdowns are logged at the end of each shift, giving the total stop time but not the time at which they occur. Despite this, the advantages of using the manually logged shutdown times outweighs the disadvantages. This is because the maintenance data (AU data) is incomplete in describing all of the shutdowns a machine experiences.

Using both of these sources of stop time may entail a certain stop to be represented in both data sets. However, studying selected data samples shows that even though the production monitoring report should contain all downtime, it does not. The intention of this line is to have automatically logged downtime in the future, which will eliminate these sources of errors.

Even if the production monitoring report data does not always give the precise true shutdown time, it serves as better information than no data at all. The process of representing this in the model is the same as the maintenance data. Firstly, it is examined through histograms, and then MTBF is entered with a uniform integer distribution and the shutdowns by an empirical discrete distribution.

Rejects

The rejects are also found in the production monitoring report. These are defined as products which do not meet the set quality requirements. Rejects are given with cause of rejection and the quantity rejected. In the model, rejects caused by each machine are determined by a probability rate. This rate is found by dividing the quantity rejected by the total number of blanks entered. Worth mentioning is the fact that the number of parts entering each machine is reduced the further on in the line a machine is placed.

(23)

3.3 Verification and Validation

Verification of the model is executed all through the development process. This includes running each block with animation features activated to be assured that the model works as intended, as well as verifying that the product types are assigned the correct attributes. Additional verification is also carried out continuously together with a senior engineer, specializing in simulations. A final verification is also completed before validating the model, this consists of investigating machine utilization, resource pool usage and output patterns.

The validation aims to assure that the model describes the actual system’s behaviour. This step is performed after the model is both finalized and verified. The line’s production quantity is disclosed weekly and therefore the model is run for weekly output. The validation process includes running the model 52 times, simulating a year’s production, and comparing it to the actual line throughput 2018. Two model outputs are compared to real system output, one from a model without extreme shutdown times, i.e. only containing shutdown time up to ten hours, and one from a model containing all recorded shutdown. This is to examine which of the two models best corresponds to the real system. The total throughput, mean weekly throughput and standard deviation of both the real system’s and the models results were compiled.

During 2018, 85% of the total output consisted of product type two. Because of this, and the fact that this is the only product type for which the cycle times have been manually clocked, the validation is carried out by simulating only parts of product type two.

To get an impartial comparison between the model output and the actual system output, certain weeks were excluded. These include all weeks which contain manufacturing of other product types than product type two as well as four weeks of summer leave. Moreover, weeks with an abnormally high recorded output were also excluded, since the exceeding amount is most likely due to formerly blocked parts now controlled and accepted. Furthermore, weeks with an abnormally low output are excluded since the model will never be able to imitate this.

Two types of validations are performed. Firstly, the outputs are compiled in charts to visualize similarities. To further validate the symmetry between outputs, a Kolmogorov-Smirnov test is performed. This test is carried out by using a web-based K-S calculator (Kirkman, 1996). To confirm the credibility of the web-calculator the output from its test is compared to the output from RStu- dio’s built in Kolmogorov-Smirnov test. The outputs are consistent and due to the simplicity of the web-based test, it is used throughout the remaining validation.

The model to be validated is run several times and the minimum output recorded from these runs is compiled. Real system output below this value are excluded from the Kolmogorov-Smirnov distribution test. This because the comparison would not be impartial if including real system outputs impossible for the model to achieve.

(24)

3.4 Model Delimitations

In the model the different machines rejects are determined by a certain probability and only one part is rejected at a time. In the real system, there are many parameters inflicting to the part being rejected or not. For example, if there is a premature tool failure in a machine and this is not discovered immediately, a large number of parts may have to be rejected. The number of rejects vary irregularly and it is hard to predict a specific daily, weekly and yearly rejection quantity with today’s conditions. Therefore, the model may fail to account for all weeks with an extreme rejection quantity, as well as it might not account for weeks with an abnormally low rejection quantity. The same goes for machine shutdowns. In the real system, shutdowns can be dependent on each other and a certain machine error can lead to a higher probability of further shutdowns. These dependencies are not captured in the model, and the real system’s days with an abnormally high error frequency will not occur in the simulation runs.

3.5 Bottleneck Investigation

Bottleneck investigation includes finding and analysing a system’s most impeding process. From analysing both the given and measured cycle times, as well as simulation results, the line’s bottleneck changes depending on product type. Worth mentioning is that, product type three and four are not yet fully developed. The cycle times for these parts are therefore highly estimated and subject to change, opting the bottleneck to shift as well.

Table 3 displays the tests to stress the bottleneck processes. The aim for testing is firstly to find out if the bottleneck is in fact used at its maximum capacity, as well as investigate how to increase its capacity. Tests includes studying how varying the multiples and time of tool changes will affect capacity, i.e. varying the intervals at which the tools are changed and the time it takes to execute the change. Moreover, it involves investigating how much a reduction in the bottleneck’s cycle time and the size and placement of buffers gives in the form of output.

Table 3: Experimental Design of bottleneck investigation

Bottleneck Experimental Design

Scenarios Standard Model Theoretical Max Model 1. Tool Changes

• Frequency D

• Time D

2. Varying cycle times D 3. Buffers

• Placement D

• Size D

These parameters will firstly be changed in isolation to find their affect on the system. To quickly test various values for these parameters the Scenario Manager mentioned in section 3.1 is used.

This tool is also used to perform cross-tests, where more than one parameter is varied. The most significant results from these tests will be displayed in the results section. Moreover, the results in their entirety can be found in the ”Results” folder under the transmission database.

(25)

3.6 Experimental Design

This project investigates three distinct base cases, one gives the theoretical maximum output of the line, one gives the current situation and the ultimate gives an optimized output.

3.6.1 Base Cases

The first base case, which gives a theoretical maximum output, is represented by a simulation model lacking any extra shutdown time. This means that the only shutdown the machines experience corresponds to the availability stated in the buy-in contract, on this line varying between 97-98%. In a perfect world, the machines have shutdowns only 2-3% of the line’s total open hours. The model still incorporates set-up times, maintenance for operators, precautionary maintenance and tool changes.

It aims to reveal the line’s maximal capacity and from this, a reasonable OPE goal can be derived.

The first step for investigation of the theoretical maximum capacity includes running a model with a machine availability of 100%. The reason is to partly ensure that this output corresponds to the output obtained by dividing opening hours with the bottleneck’s cycle time, as well as finding the percentage lost due to tool changes. The next step is to add the 2-3% shutdown to machines which has a real shutdown time exceeding this, and to add the actual recorded shutdowns to the machines which do not. The output resulting from this model gives the line’s theoretical maximum capacity.

The second base case aims to map today’s production capacity. This model is a continuation from the first base case model, but incorporates more shutdowns based on historical data. This model gives the current capacity for the line in a somewhat stable process. The model in this case is also used for further validation, comparing it’s output to that of the real system.

The ultimate base case describes the line after optimization. This line has more output than the second base case, but less than the first base case. The final recommendations to the company are based on experiments from the model in this case.

3.6.2 Scenarios

In the ultimate base case various scenarios, seen in Table 4, are tested to reveal where the most enhancement to capacity can be found. The first scenario to investigate is how different lengths of shutdowns in the machines affects the line’s overall capacity. Another scenario includes investigating how different competences of the operators affect the line’s capacity and total output. These tests will be performed using mainly product type two, since these are the cycle times validated.

Table 4: Experimental design of diverse scenarios

Experimental Design

Scenarios Standard Model Theoretical Max Model 1. Shutdown time

• Include all stops D

• Include stops ≤ 10h D

• Include stops ≤ 4h D

2. Operators

• Equal competence D

• Unequal competence D

(26)

4 Results

This section includes the project’s results. It contains model results, showing how specific processes are represented in the model, validation results and lastly results concerning the project’s objectives.

4.1 Simulation model

This section contains model representations of the real system’s events.

Figure 5: Model of UGA-line

Figure 5 displays an overall view of the model. The rest of the model, including all of the line’s machines, can be found in appendix A3.

4.1.1 Hierarchical Model Structure

The simulation model is built according to a hierarchical structure. This means that each of the line’s machines are represented by its own block, in figure 5 represented by grey boxes/squares.

4.1.2 Database

The cycle times for the different product types are stored in a database. This covers machining time in its entirety for all the different processes that are carried out. It also contains output records and activity states from the latest simulation run. machining time for all the different processes that are carried out. It also contains output records and activity states from the latest simulation run.

(27)

4.1.3 Gantry

A model representation of the gantry, which transports the part between the different machines, can be found in Figure 6. The gantries in the model are simplified and does not cover the extensive real life priority scheme.

Figure 6: Model representation of gantry process

Instead, each of the gantries three arms are represented by an individual resource pool in the model, see Figure 7. When a part enters the gantry process, the specific resource pool responsible for this area is required for the transportation to the next machine.

Figure 7: Model representation of gantry resources

(28)

4.1.4 Input and Output Conveyor

In Figure 8 the model representation of the input conveyor is displayed. The process of loading blanks on the conveyor requires an operator, so a resource pool is called upon. In reality, the operator does not load the next blank onto the conveyor as soon as one part has entered the first machine but waits for a few slots to open up. In the model this is represented by five items being batched together, and only then is the operator required for the loading process.

Figure 8: Model representation of Input Conveyor

4.1.5 Set-up Time

Figure 9 displays how a set-up time is represented in the simulation model. For the model to incorporate reality the set-up had to be divided into two separate steps. Firstly, the product generation must be identified. If the generation differs from the previous part, it will be redirected to an activity which delays the part a number of minutes. This activity calls for a resource pool, in this case an operator, to execute the set-up. Similarly, in step two the product type must be identified. If this differs from the previous type it will likewise be delayed and an operator is required to carry out the set-up.

Figure 9: Model representation of set-up times

(29)

4.1.6 Tool Changes

This process starts with a counter, which keeps track of the quantity of parts passed. When this reaches a certain predetermined level, which differs depending on machine, the part will be redirected.

It is redirected as a tool needs to be changed, and a resource is called to complete the task. This is shown in Figure 10.

Figure 10: Model representation of a tool change

4.1.7 SPC - Single Part Control

The single part control is a process repeated on a specified interval given a machine, performed to assure machining quality. This does, like the tool change in Figure 10, start with a counter which records the passing items. If redirected, the part reaches a delaying activity which demands an operator to perform the quality check. Unlike a set-up or a tool change, this process allows for succeeding parts to pass through even if a quality check is taking place.

4.1.8 Rejects

To account for scrap parts, items are redirected to a premature exit according to a given probability.

This is represented by Figure 11.

Figure 11: Model representation of rejects

(30)

4.2 Model validation

To investigate conformance, runs from each of the two models are compared to real system output in a chart. The outputs are arranged in order of magnitude to get a perspicuous view of the similarities.

Validation results for both, the model without extreme shutdown time and the model with extreme shutdown time, are presented in this section as well as the results from the KS-tests.

4.2.1 Model without Extreme Shutdown Time

Output Validation

Real System Output Model w.o Extremes Output

Figure 12: Output comparison between real system and model without extremes

Table 5: Weekly mean outputs and standard deviations

Real system¹ Model w.o extremes

Mean Weekly Output X 0,99X

Mean Weekly standard Deviation x 0,92x

As displayed in Figure 12 the model output conforms well to the real system output. Especially when the real system output is stable, as seen in the middle of the bar chart. However, when the real system’s tail values are too high or low, the model can not properly imitate these fluctuations.

Table 5 stresses this even further as the model’s standard deviation is lower than the real system’s, indication a more stable process.

1X denotes confidential information which is included in the company report.

(31)

4.2.2 Model with Extreme Shutdown Time

Output Validation

Real System Output Model w. Extremes Output

Figure 13: Output comparison between real system and model with extremes

Table 6: Weekly mean outputs and standard deviations

Real system² Model w. extremes

Mean Weekly Output X 1,04X

Mean Weekly standard Deviation x 0,83x

Similarly, output from the model with extremes conforms well to the real system’s output, as seen in Figure 13. The same tendencies are present here, the model fails to fluctuate to the same extent as the real system. Table 6 reveals that the model’s standard deviation is notably lower than the real system’s.

A comparison cumulative fraction plot of the Kolmogorv-Smirnov test for each of the two models can be found in appendix A1, Figures 20 and 21. The models are run 15 times and all outputs are accepted in the test, concluding conformity in the distributions with a confidence level of 99,9%

for the model without extreme shutdown time, and 99% for the model with extreme shutdown time.

4.3 Simulation Results

This section displays the results with respect to the specified objectives.

4.3.1 Theoretical Maximum

Firstly an absolute maximum level of output is calculated by dividing the line’s opening hours by the bottleneck’s cycle time. By running the model with a machine availability of 97-98%, a theoretical maximum output is obtained. From this a reasonable OPE goal is derived.

(32)

To arrive at a reasonable OPE goal, the model calculated output is divided by the absolute maximum capacity. The model calculated reasonable OPE goals, if the new area’s cycle times are set to the ones recorded in the buy-in contract, are displayed in Table 7.

Table 7: Model calculated OPE corresponding to purchased cycle times

Model Calculated OPE

Product type 1 80,3%

Table 8 represents an OPE goal corresponding to a line with current cycle times for the new area.

Table 8: Model calculated OPE corresponding to current cycle times

Model Calculated OPE

Product type 3 79%

Product type 4 79%

4.3.2 Isolated Experiments

Table 9 displays the obtained results from the isolated tests. These were carried out using only product type two. As the results show, there is significant capacity to uncover if shutdowns are reduced. Moreover, it is shown that the number of operators has a lesser impact on total output.

Table 9: Experimental results of diverse scenarios

Experimental Results

Scenarios Standard Model Theoretical Max Model 1. Shutdown time

• Include all stops 90%

• Include stops ≤ 10h 100%

• Include stops ≤ 4h 106%

2. Operators

• Equal competence

- 2 operators 99,2%

• Unequal competence

- 4 operators 100%

(33)

4.3.3 Bottleneck Investigation Results

A bottleneck is defined by blockage upstream and idling time downstream, as mentioned in section 2.2. The following section will display the line’s bottlenecks for all product types.

4.3.3.1 Product Type One

Figure 14 (a) displays manufacturing of product type one when cycle times are set to current machining times. As the figure displays, there are three subsequent bottleneck processes. These are located right at the beginning of the line, with the turning machines followed by the milling machine.

Instead, Figure 14 (b) describes manufacturing of product type one using cycle times recorded in the buy-in contract. Machines with a current cycle time lower than the one recorded in the contract are not changed. It is seen that the milling machine is the line’s bottleneck. It is not the machine with the greatest cycle time, but has frequent tool changes.

20%

40%

60%

80%

100%

Milling Ma chine Splines Machine

Welding Process Induction Hardening

Shot Blasting Machine Grinding Machine

Balancing Machine Finishing Wash

Activity States in Machines

Product type 1: Current Cycle Times

Busy Tool Change Idle Blocked Shutdown Offshift

(a) Current Cycle Times

20%

40%

60%

80%

100%

Milling Ma chine Splines Machine

Welding Machine Induction Hardening

Shot Blasting Machine Grinding Machine

Balancing Machine Finishing Wash

Activity States in Machines

Product Type 1: Purchased Cycle Times

Busy Tool Change Idle Blocked Shutdown Offshift

(b) Purchased Cycle Times Figure 14: Difference in the line between current and purchased cycle times for product type one

Table 10: Difference in output between scenarios

Weekly Output³

Current Cycle Times X

Purchased Cycle Times x

Effect of lowered cycle times + 3,3%

4.3.3.2 Product Type Two

Figure 15 (a) displays manufacturing of product type two when cycle times are set to current machining times. The figure illustrates higher cycle times for the turning processes and milling machine as well as a lower cycle time for the splines machine. These cycle times amount to the milling machine being the bottleneck for product type two manufacturing.

Similarly as above, Figure 15 (b) describes the line manufacturing product type two when cycle times are set to purchased times. For these cycle times the bottleneck machines are the machining centres. They have both a long cycle time as well as frequent tool changes.

FLOW SIMULATION OF AN OUTPUT SHAFT LINE: