Managing predecessor relationships in the Automotive industry

(1)

Managing predecessor relationships in the Automotive industry A case study at Scania Production Zwolle

Student:

Matthijs Loer

Student number: S2886502

Supervisor: dr. N.D. van Foreest

Second supervisor: dr. W.van Wezel

Supervisor Scania Production Zwolle: G.J.Stoffers, MSc

MSc Technology and Operations Management Faculty of Economics and Business

(2)

2

Abstract

Purpose: Documenting and mapping precedence relationships of assembly tasks is a

complex process. A valuable worker utilization tool called Line Balancing is not performable due to undocumented precedence relationships. Therefore, this paper proposes a manual framework to determine which assembly tasks exist, how they can be aggregated and what the corresponding precedence relations are.

Methodology: The method of this paper is a design science approach to create a

generalizable framework to manage precedence relations.

Findings: Firstly, the proposed framework extends existing manual methods by

reducing the time intensity in the precedence extraction phase while safeguarding a higher accuracy than inapplicable automatic approaches.

Secondly, empirical research shows that manual approaches are able to accurately extract the precedence relationships. Furthermore, empirical research shows that the main issue is documenting and representing precedence relations.

Research limitations: The creation of modular structures and assembly tasks are based

on personal interpretation. The framework has not been tested with a Line Balancing heuristic. Therefore, the effectiveness of the framework has not been validated.

Practical limitations: The proposed framework is a manual approach. Therefore, the

method requires a substantial amount of time. Furthermore, the merging process of simulation areas requires coordination and standardization between users. This process is considered to be challenging.

Value: The framework will describe how to generate the resources to apply Line

Balancing. The framework enables manufacturers to further optimize the utilization rate of assembly line workers.

(3)

3

Acknowledgements

Throughout the entirety of this research, I have received support from many people. I would like to thank anyone who was willing to help me. First of all, I would like to thank the constructive feedback from my first supervisor dr. N.D. van Foreest. He guided me throughout the entire research phase. Furthermore, the thoughtful comments of my second supervisor (dr.W.van Wezel) and criticism on my research gap pushed me to identify the core problem of Line Balancing in real-life scenarios. Secondly, I would like to pay gratitude to Scania Production Zwolle for the opportunity of performing my thesis there. This research was impossible to execute without the conversations with assembly line workers, colleagues and my supervisor within the company, G.J.Stoffers MSc. In particular, I would like to thank my supervisor at Scania. He provided me with weekly feedback on this paper and arranged the possibility to work at the assembly line to enable the empirical validation for this research.

(4)

4

List of definitions

Word Definition

Line Balancing Distributing workload across all available workstations inside a facility.

Precedence relation Sequencing requirement which must be satisfied before a final product can be reached.

Precedence graph All the precedence relationships are documented in a single graph.

Independent assembly task This task has no precedence relationship to another task within a wide area of

simulation/ the entire assembly line. Modular structure Decomposing assembly tasks into logical

sets.

Area of simulation The chosen amount of modular structures (and corresponding workstations) to perform the extraction of precedence relations on.

Bill of material (BOM) A bill of materials is a list of the raw materials, sub-assemblies, intermediate assemblies, sub-components, parts, and the quantities of each needed to

manufacture an end product. Position standard A sequential list of assembly tasks

(5)

5

Table of

1 Introduction

Line Balancing has been around for several decades in the academic literature. It is the art of balancing workload equally over workstations inside a manufacturing plant. This way, worker capacity can be utilized as much as possible. Line Balancing has been an optimization problem of significant importance: the efficiency difference between an optimal and a sub-optimal assignment can yield savings (or waste) reaching millions of dollars per year (Falkenauer, 2005). Line Balancing (LB) was born after the necessity of product variation in various industries, like the Automotive industry. Meyr (2004) has identified that product variation is extremely high in the Automotive industry. The high amount of product variation has led to significant trouble in documenting assembly tasks correctly.

In addition, the significant amount of product variation has led to a large amount of predecessor relationships. Baybars (1986) describes the precedence relationship as technological sequencing requirements which need to be fulfilled to correctly create a final product. Line Balancing heuristics require a good representation of these predecessor relationships to generate usable outcomes. However, the problem is that within the Automotive industry the knowledge about managing large amounts of precedence relationships is lacking. Subsequently, the current balancing methods reach relatively low worker utilization rates. Therefore, this paper will focus on how to manage predecessor relationships in the Automotive industry such that heuristics are able to solve real-life cases.

(7)

7

However, the potential of Line Balancing exceeds any decision support system or configuration method available. As previously stated, Line Balancing maximizes the utilization rate of assembly line workers. However, the usefulness of Line Balancing exceeds this singular purpose. This is caused by the fact that the application of Line Balancing forces a manufacturer to accurately describe the precedence relationships. Line Balancing helps to understand assembly processes in the facility. This will contribute to the quality of the final products in three ways. Firstly, the performance of the staff will improve because the quality of the training process will increase. The obtained knowledge from accumulating precedence relations will contribute to the knowledge at the workstations. Subsequently, their ability to train new staff becomes better. Secondly, the consequences of changes to other components at the assembly line can be monitored with relative ease. The influence of changes to components can be directly linked to the predecessors and successors of set component, this to prevent problems in the assembly process. Thirdly, Scholl, Boysen and Fliedner (2008) state that under manual labor, quality suffers if operators are overloaded with work and thus need to work faster. Thus, to ensure that operators have less overload situations, appropriate Line Balancing needs to be performed.

Therefore, this paper will focus on how to shape predecessor relationships in such a way that they can be represented in a precedence graph. This to ensure that an optimal division of labor tasks will be possible in the foreseeable future. This has led to the following main research question: how should predecessor relationships be

managed such that Line Balancing heuristics can be applied to assembly lines?

(8)

8

2 Theoretical Background

The difficulties of managing precedence relationships have been mentioned in the introduction. The management of precedence relationships has been overlooked by the majority of the academic researchers. However, a few researchers dedicated time to map assembly sequence relationships. Therefore, this chapter will start by giving an overview of the existing literature regarding the creation of precedence graphs. Furthermore, this chapter revolves around existing literature on the four characteristics corresponding to the management of precedence relations (identifying precedence relationships, updating precedence relations, defining assembly tasks and the correct construction of modular structures). The benefits of Line Balancing will be discussed in the final subchapter.

2.1 Existing literature on creating precedence graphs

The creation of precedence graphs has been the main difficulty in applying Line Balancing tools in real-life scenarios. This led to low use of Line Balancing tools in manufacturing plants (Chase, 1974;Milas, 1990;Boysen et al., 2008). The lack of knowledge on precedence relationships led to academic research on this topic. The generation of feasible assembly sequences has been researched for over thirty years. The creation of precedence graphs has already been reviewed by Klindworth et al. (2012).

Bourjault (1984) started by creating a question and answer method to establish connections between liaisons (parts as nodes). The validity of the created connections was decided by a domain expert. However, the drawback of the method was the amount of questions necessary to obtain the connections. The amount of questions increases exponentially compared to the amount of connection points. De Fazio and Whitney (1987) adapted the method created by Bourjault in 1984. They were able to reduce the amount of questions necessary to create the connection between the liaisons. Both previous mentioned methods are based on the assembly process. However, Homem de Mello and Sanderson (1991) introduced another approach. These two researchers were able to extract the precedence graph based on a disassembly process. Underlying, this approach documented all possibilities to disassemble a product by examining cut sets of the liaison graph. The liaison graph is an undirected graph that contains parts as nodes, where two nodes are adjacent if they are connected in the assembly (Klindworth et al., 2012).

(9)

9

Recently, two methods were proposed to create the precedence graph in large product variety. Altemeier et al. (2010) created an automatic top down approach to generate a precedence graph. On the other hand, Klindworth et al. (2012) created a basic learning precedence graph concept (BLGC) to accurately approximate the precedence graph.

BLGC is able to create a maximum and minimum graph which together form a good approximation of the actual precedence graph. The maximum graph contains the same nodes as the target graph and at least all the precedence constraints of the target graph. The maximum graph does not contain the identified independencies. A relationship is independent when assembly tasks can be performed in either order. Assembly sequences are used to identify whether the initial considered precedence relationship can be removed or not. Therefore, the limitation of the maximum graph, is the fact that sufficient assembly sequences are needed to limit the amount of excessive potential precedence relationships.

The minimum graph is a precedence graph that contains the same nodes and task times as the target graph, but only a subset of precedence restrictions of the target graph. In contrary to the maximum graph, all precedence relations described by the minimum graph are valid, whereas the independencies are temporary (Klindworth et al., 2012). The precedence relationships are confirmed with the use of CAD-data or expert interviews. If no information is available, the minimum graph contains no arcs and powerful methods for the well-known bin packing problems can be used to solve the ALBP based on the minimum graph and thus get the lower bounds for the target graph (Scholl, Klein and Jürgens, 1997;Otto and Otto, 2014) If the lower bound is equal to the number of stations in a feasible solution (deduced from the maximum graph), an optimal solution of the target graph has been found (Otto and Otto, 2014). The results of the computational tests of Klindworth et al. (2012) confirm that in many cases the lower and upper bound are equal or very close to each other.

The basic learning precedence graph concept has got several disadvantages. Firstly, the maximum graph in BLGC is constructed based on former production plans (Otto and Otto, 2014). Thus, it achieves its effectiveness only several months after the start of production (Otto and Otto, 2014). In quickly changing working environments this significantly reduces the usefulness of the results. Furthermore, the usefulness of the maximum graph highly depends on the availability of sufficient assembly sequences. In addition, Otto and Otto (2014) encountered that in practice it is not uncommon that independent assembly tasks are not identified because the assembly tasks are never interchanged with each other in assembly sequences. Otto and Otto (2014) add that some sections of the assembly line are seldom re-balanced. The limited variability of production plans, which is widespread in the Automotive industry, significantly restricts the effectiveness of BLGC.

(10)

10

the quality of the minimum graph. However, the main disadvantages stated in the previous paragraph have not been handled.

The other method for creating a precedence graph was made by Altemeier, Brodkorb and Dangelmeier in 2010. It is a top down approach for the generation of a precedence graph under influences of high product variety. The disadvantage of this method is the fact that the presented method is very error prone. On the contrary, the method relies on resources which should be available at all manufacturers. The precedence graph is automatically derived from the bill of materials (BOM) and buildability rules as well as existing solutions for the assignment of assembly tasks to workstations. (Altemeier et al., 2010). Consequently, the Line Balancing method will always find a solution based on this approach. Altemeier et al. (2010) will not be handled extensively because the method is error prone.

Concluding, the methods created by Altemeier et al. (2010) and Klindworth et al. (2012) showed promising potential, but still had limitations in terms of high error sensitiveness or dependencies on resources. Therefore, this paper will focus on a manual approach which needs a limited amount of data. Whilst proposing requirements for software which limits the errors in documentation of precedence relations.

2.2 Identifying precedence relationships

“Assembly tasks cannot be assigned to workstations arbitrarily because of the technological sequencing requirements, known as precedence relations; the processing of a task may not start until certain tasks, i.e., its immediate predecessors have been processed” (Baybars, 1986). This description does not cover the entire use of precedence relationships. Scholl, Boysen and Fliedner (2007) state that not every precedence relationship is based upon strict technical requirements, sometimes the order of assembly tasks is arranged due to increased efficiency. Therefore, solely considering technological precedence relationships can lead to Line Balancing results which are unusable.

The previous paragraph illustrates the complexity of precedence relations. This complexity is caused by the following tradeoff: flexibility versus ergonomics. Precedence relationships should be considered when it significantly contributes to the efficiency of assembly tasks. Furthermore, this positively influences the variation of assembly task times and the duration of assembly tasks. However, if the precedence relationship is not considered, Line Balancing tools have more assignment possibilities. Subsequently, the chance is higher that the assembly line will obtain a more equally distributed workload across all workstations. The data to obtain precedence relations can be generated with the use of interviewing assembly line workers (domain experts), CAD-tools to identify technological sequencing and deducing (in) dependencies with the use of modular structures.

(11)

11

Therefore, both manual approaches will be evaluated. Interviewing is considered to be a time-intensive method of extracting precedence relations (Halpern, Sarisamlis and Wand, 1982;Otto and Otto, 2014). Furthermore, the amount of questions necessary to obtain the precedence relationships has a quadratic relationship with respect to the amount of assembly tasks. This makes it nearly impossible to manage several thousands of assembly tasks which are present at modern assembly lines (Otto and Otto, 2014). Therefore, interviews are not justifiable as a stand-alone data collection method (Otto and Otto, 2014). Thus, other methods are also necessary to collect data.

Modular structures could provide one of these supporting methods. Modules are defined as groups based on certain characteristics, each with different functionality and interfaces (Baldwin and Clark, 1997; Wilhelm, 1997; Otto and Otto, 2014). Modular structures were created to produce a large variety of products while being able to minimize the costs (He and Kusiak, 1997). He and Kusiak (1997) touched the positive influence of modularity on Line Balancing results. Their proposed heuristic showed better results for Line Balancing with the use of modularity than the initial situation of several problem sets. However, the application of modularity for Line Balancing purposes has not been touched in any paper.

Modules are manageable and relatively cheap way to collect information on the precedence relations between modules. In line with the definition of modules and insights generated in the automotive industry, the assumption can be made that all the assembly tasks from different modules are connected with the same kind of precedence relations (Otto and Otto, 2014). In other words, either all the assembly tasks from module A precede each of the tasks from module B, or vice versa, or all the assembly tasks from module A are independent from each task in module B (Otto and Otto, 2014). The previous described definition of modular structures allows us to justify simplifications in Line Balancing as long as each task inside a modular structure is completely considered in the balancing method, the result strictly satisfies the cycle time restriction and the precedence constraints within the module is satisfied. Concluding, the absence of appropriate CAD-data has led to the inapplicability of automatic tools which are able to generate precedence graphs. Furthermore, interviewing as a stand-alone collection method is too time-consuming. Finally, the benefits of modular structuring have been proven. However, how to compose modular structures for Line Balancing purposes has not been touched yet. Moreover, the combination of modular structuring and interviewing has not been considered in any approach. Therefore, this paper will use a multi-method approach to obtain the precedence graph.

2.3 Updating the precedence graph

(12)

12

academic fields to create requirements to facilitate the need for an updated precedence graph. For instance, software maintenance learns us that consistency in the documentation is key to reach up to date systems (Correia et al., 2009). Frankel et al. (2002) emphasize the importance of continuous loops of feedback to identify changes to relationships as quickly as possible. Furthermore, continuous feedback loops could contribute to decrease the sensitivity of errors for mapping new precedence relations.

2.4 Defining assembly tasks

The management of precedence relationships has been overlooked by the majority of the academic researchers focusing on Line Balancing. However, the definition of an assembly task has been touched by Baybars in 1986. He defined an assembly task as the smallest indivisible work element. Furthermore, Baybars (1986) adds to the previously mentioned definition that an assembly task cannot be divided between two or more stations without conflict. Scholl and Becker (2006) describe assembly tasks as the total amount of work inside a production facility partitioned in sets of elementary operations.

The definition of what an assembly task entails remains relatively vague. It is still unclear to what extend assembly tasks need to be considered for Line Balancing. This mainly stems from indivisible work elements that need to be deduced from the initial source of information (worker instructions/position standards). Consequently, the list of assembly tasks would be long. As mentioned before, the documentation and extraction of precedence relationships between a substantial amount of assembly tasks will be impossible to perform manually. On the contrary, supporting automatic tools are not able to accurately describe the precedence graph. Thus in the context of Line Balancing, the definition of an assembly tasks is complex. This is caused by the fact that the requirements for defining assembly tasks are not stated in previous literature. However, Antani (2014) identified that the advantage of splitting assembly tasks into indivisible elements is limited. Antani (2014) adds that there is no additional advantage between splitting assembly tasks or clustering small assembly tasks (with closely linked processes) to a sub-assembly in particular cases. Furthermore, the advantage of clustering assembly tasks is that small assembly tasks do not get fragmented during the task distribution process (Antani, 2014).

Pears (2015) described task grouping constraints which can be manipulated to define assembly tasks. The following constraints can be used to group assembly tasks:

1. Upon finishing a task, the worker has a part or tool in hand, intended for usage on another task.

2. Some assembly tasks require a follow-up self-inspection of work performed, which is a separate task.

3. Part scanning assembly tasks exist to assure that a later installation task uses the correct part.

The complexity of defining assembly tasks is related to the tradeoff between obtaining sufficient variability for useful Line Balancing results and the time intensity of mapping precedence relations. Just like there is a limit for splitting assembly tasks into

(13)

13

The usefulness of the method decreases when clustering is performed extensively. Therefore, the definition of an assembly task is highly important to obtain valid Line Balancing results. The requirements of what a suitable assembly task for Line Balancing is, is not yet documented by any researcher. Therefore, this paper focuses on describing how an assembly task should be defined.

2.5 Creating modular structures

Modularization originated from the need to simplify the management of complex systems (Pandremos et al., 2009). Its general purpose is to decompose the complex system into constituent parts that might break apart “naturally” without destroying the integrity of the whole (Sako and Murray, 1999;Pandremos et al., 2009). Baldwin and Clark (1997) defined modularity as a concept that is applied to manage complex systems, by breaking them down into parameters and tasks that are interdependent within and independent across the modules.

Pandremos et al. (2009) describe the general definition within the Automotive industry: “A group of components, physically close to each other that are both assembled and tested outside the facilities and can be assembled very simply onto the car”. Pandremos et al. (2009) also provided requirements for the number of modular structures: tolerance issues, handling of sub-assembly and accessibility for joining. However, the main disadvantage of this paper is the fact that they do not consider modular structures in correlation with Line Balancing. Therefore, the transfer of knowledge may not be valid.

Several works consider the product architecture as the baseline for product family development (Jiao and Tseng 1999; Dahmus et al., 2001;Mikkola and Gassmann, 2003;Otto, 2001;Jose and Tollenaere, 2005). Jose and Tollenaere (2005) state based on Ulrich and Eppinger (2000) that there exist two dimensions in product architecture: the functional one, which is the group of operations and transformations that contributes to the general functionality of the product, and the physical one, which refers to the group of physical components and assemblies that enable a function. Ulrich and Eppinger (2000) created a four-step guide to establish modular product structures:

1. Develop a conceptual model of components and functions

2. Cluster the elements of the scheme and regroup components inside the modules.

3. Detect interfaces and modules based on a created geometric layout 4. Identify fundamental and incidental interactions in the scheme.

Based on the above, general requirements can be drawn to create modular structures which can be manipulated for Line Balancing purposes. The final requirements will be based on a combination of empirical research and academic literature. These requirements will be discussed in the results section.

2.6 Clarification for Line Balancing in real-life scenarios

(14)

14

work force was utilized for the creation of Automotive vehicles. On the contrary, private communication with Dr. Altemeier emphasized the potential of Line Balancing heuristics. He stated that worker utilization rate could reach up till 90 percent. Pröpster et al. (2015) already proved the usefulness of Line Balancing. The simulation at a truck manufacturer showed that Line Balancing could reach 85 percent staff utilization. Furthermore, the truck manufacturer showed high correlation with the chosen case study. The correlation is determined by the similar worker environment and worker arrangements. Both truck manufacturers make use of “jumpers”. The use of jumpers serves as an additional worker to the core workforce in order to provide support at short notice in the event of capacity bottlenecks to finish the product within the cycle time (Becker and Scholl, 2006). In addition, Pröpster et al. (2015)

prove that Line Balancing can contribute to better allocation of jumpers. Empirical research at the case study suggested that the allocation of jumpers is not

(15)

15

3 Methodology

3.1 Introduction to Scania Production Zwolle

Scania Production Zwolle (SPZ) is part of the Scania Group, which is a Swedish vehicle manufacturer since 1891. The production site in Zwolle was founded in 1964 and it solely focuses on the production of trucks. Over the years, the production process at SPZ has been changed significantly. Quality and safety have increased in importance, lean and quality improvement principles have been incorporated in the core business. One of the characteristics of the Automotive industry, is that the products have modular designs. This means that every assembly process adds a specific feature to the truck. However, product types follow mostly the same assembly procedure.

Scania is having trouble with distributing the workload of assembly tasks. Currently, jumpers are introduced to work on product variants which need additional capacity to finish the set of tasks within the desired cycle time. Jumpers are doing very specific assembly tasks, which leads to low efficiency under jumpers. Some jumpers

are supporting co-workers at their station, which is outside their authority. This not only leads to other workers doing less work, but also covers up overload

situations caused by incorrect task division. On the contrary, other jumpers are doing non-value adding tasks. Line Balancing to distribute the workload of the entire plant should theoretically be possible. However, the problem of lacking precedence relations must be overcome to make this possible.

Currently, the worker tasks are arranged in position standards and Line Balancing is performed subjectively under the supervision of team leaders. Position standards are sets of lists in which all the assembly tasks and their assembly times are denoted in a sequential order for a specific position at the workstation. Although, the assembly tasks are all documented on paper, the predecessor relationships have not been documented. Therefore, an enormous amount of data needs to be organized and grouped before Line Balancing can be performed. Consequently, this report will provide a framework to organize the data inside Automotive producers.

3.2 Design Science

This paper is based upon design science. Design science is a scientific approach primarily focused on problem solving and discovery as opposed to the accumulation of theoretical knowledge (Hölmstrom et al., 2009). Furthermore, design science focuses on the development and performance of designed artefacts with the intention of improving the overall functional performance. This method is chosen because of the lack of generalizable information in the Automotive industry. This paper aims to satisfy the need for the management of predecessor relationships beyond case level.

The Hevner cycles will be used to generate the knowledge for this paper. Figure 1 displays the application of the Hevner cycles to the problem context. The three cycles are denoted as: the relevance cycle, the design cycle and the rigor

(16)

16

the framework. The design cycle will be used as an iterative process to build a framework for the management of predecessor relationships. At last, the rigor cycle will be used to extend the knowledge base with a framework to manage precedence relationships. The knowledge base will be extended by extensively using and adapting available academic research of different fields. More explanation about the application of the Hevner cycles will be outlined in the next paragraphs.

Figure 1: The three cycle approach of Hevner (2007) specified to this problem

3.3 The three cycles of Hevner

3.3.1 DSR: Design Cycle

(17)

17

all of these alternatives against the requirements set in the first phase until a satisfactory design is achieved.

3.3.2 DSR: Rigor Cycle

The rigor cycle reviews existing knowledge to ensure its innovation (Hevner, 2007). Therefore, the rigor cycle starts by evaluating and questioning the current knowledge base. The first phase emphasizes on the shortcomings of the existing academic literature in the field of interest. Hevner et al. (2004) strengthen this statement by mentioning that researchers should thoroughly research and reference the knowledge base in order to guarantee that the produced designs are research contributions and not routine designs based on the application of well-known processes. The second phase is characterized by developing the right method and research set-up to be able to solve the gap in the literature. By following the method created in the second phase of the research a solution will be provided. This solution has to be tested in several real-life cases to ensure the credibility of the generated knowledge. Furthermore, the evaluation of the knowledge will be based on the fact whether or not the research gap proposed at the beginning of the cycle is satisfied. 3.3.3 DSR: Relevance Cycle

(18)

18

4 Research set-up and personal execution

The limitation of line balancing heuristics is that an accurate representation of the precedence graph is currently unavailable. Therefore, empirical research was performed to determine which obstacles are encountered when performing a combination of manual precedence extraction processes.

Currently, the following difficulties are encountered when managing precedence relations based on initial literature review:

1. It is currently unknown what the optimal size is of a precedence graph. This is caused by the fact that it is currently unknown how many relationships can be managed by a single person.

2. It is unknown how the initial source of information (worker instructions) needs to be adapted.

3. What is the optimal way of extracting precedence relationships in real life scenarios.

4. The working environments change often, which causes inaccuracies in the precedence graphs quickly. Therefore, the precedence graph should be managed regularly. However, the optimal way to accommodate this need has not been researched yet.

The creation of a precedence graph was empirically investigated with the use of modular structuring and interviewing. In this empirical section above mentioned difficulties were discussed. Initial literature review and discussions with colleagues at the case company formed initial thoughts of the applicability of manual approaches. These hypotheses were characterized by a high amount of uncertainties and lack of practical knowledge. Therefore, the following initial thoughts were tested in a practical environment at Scania Production Zwolle:

1. The creation of a precedence graph is a complex process because of the high amount of assembly tasks and their corresponding precedence relationships. 2. The interviewing process should be based on Altemeier (2009) and Otto and

Otto (2014). In these articles the way interviewing needs to be performed and how to achieve the desired output of interviewing is discussed extensively. 3. How modular structuring should be performed for line balancing purposes is

unknown. However, it was already discovered that assembly tasks which are present in a modular structure have to be closely correlated with one another. As mentioned in the initial literature overview, assembly tasks can be grouped together if there are similarities in the functionality of the assembly tasks. The empirical research will show if additional characteristics influence the content of a modular structure.

(19)

19

as different in any shape or form, then these assembly tasks can be grouped together.

5. Discussions with colleagues at the case company and dr. Altemeier indicated that concerns exist for the validity of the interviewing results. Although, the interviewing process is a known approach, it still seems undesirable to solely rely on interviewing results extracted from a single experienced assembly line worker. Therefore, a control mechanism should be introduced to increase the validity of the results.

6. The hypothesis of this research is that splitting the facility in simulation areas makes the management of precedence graphs possible with the use of a combination of manual approaches. This initial thought is based on the estimated complexity reduction (with the use of modular structuring and task grouping). It is considered that this suffices the application of interviewing for the creation of a precedence graph.

The main goal of the empirical research is to test the above mentioned hypotheses and use this knowledge to encourage future academic research.

Creation of modular structures

This first step is influenced by the current distribution of the case company. The case company already distributed work over workstations and already created sets of workstations (work areas). The current distribution can be considered as an initial starting point for the distribution of simulation areas. However, it is unknown whether or not the current distribution is sufficient for line balancing purposes. Therefore, the correlation between workstations should be reviewed. This need was accommodated with the use of modular structuring. Initial literature review suggested the possibility of using modular structuring for the organization of assembly tasks on a large scale. After modular structuring is applied, the similarities and differences can be discussed and appropriate decisions can be made about the current division of areas. The main question which arises now, is how modular structuring should be applied to correlate assembly tasks. Jose and Tollenaere (2005) and Baldwin and Clark (1997) provide insight in how modular structuring can be applied. Based on their statements the following characteristics can be considered to obtain the correlation between assembly tasks to create modular structures: location at which the assembly tasks are performed, tools used for the assembly task and the functionality of a part/component.

Figure 2: example of a modular structure

(20)

20

on the truck. Furthermore, the functionality of the assembly tasks are all similar. All assembly tasks are based on preparing the radiators placement. Finally, there is no heavy machinery or tools involved in the preparation for the placement of the radiator. Concluding, all characteristics are in line with one another and therefore the assembly tasks can be organized in the same modular structure “front end left hand side”.

After the assembly tasks are all organized and placed in a modular structure, a set of modular structures can be selected on which the extraction of precedence relations will be performed. Important note: this decision was influenced by the initial thought of creating a precedence graph for a simulation study. This initial idea was proposed by the case company to determine whether automatic line balancing is a possibility for the future. The chosen workstations were characterized by the degree of knowledge of the assembly line workers (for interviewing) and the high degree of jumpers at the workstations. In addition, the main information sources at the case company had good contact with chosen workstations. Based on these comments an initial small sample of three workstations were chosen out of over 50 workstations. This initial sample was relatively small because of the following reasons:

1. Literature suggested that plain manual approaches (like interviewing) were extremely time intensive. Therefore, it was unknown how long the extraction of the precedence relations would take. Thus, starting with a small sample would lead to sufficient time to reflect on the manual extraction process. 2. The initial idea was to perform a line balancing heuristic to observe any

differences, this idea was later on rejected due to lack of performance of a line balancing heuristic for small samples (limited variability) and time related issues.

3. No documentations structures or guides are available to organize all the confirmed precedence relations in an automatic environment (Excel or other programs). Therefore, the best way to relate precedence relations had to be determined.

One of the issues in extracting precedence relationships for an entire facility is that the amount of interrelationships become exorbitantly high. Furthermore, determining where to start extracting precedence relationships when considering thousands of tasks is a complex task. Therefore, empirical experiences showed that there is a need for structuring thousands of tasks before actually starting the manual extraction process. These personal experiences revolved around visualization difficulties (between big components/modular structures/workstations) and selecting an area on which the precedence relationships will be found. Literature suggested a method to accommodate this need: modular structuring. However, up till now this method has not been used to fulfill this need.

(21)

21

assignments: Falkenauer (2005) describes this as the identity of a workstation. Literature added that there are more factors which bound the assignment possibilities of tasks. This will be extensively elaborated in the explanation of the framework in Chapter 5.

While creating the modular structures, discussions with employees at Scania production Zwolle led to the following piece of information. They suggested that there exist different views on the organization of assembly tasks. This suspicion was confirmed during interviews with assembly line workers. The initial modular structures can be interpreted in better ways in some occasions. This was observed after the evaluation of initial modular structures by assembly line workers. Therefore, the initial modular structures had to be reviewed and questioned before starting with the extraction of precedence relationships. This can be achieved by proposing a set of modular structures at each of the involved workstations and discuss the previously obtained modular structures.

Furthermore, available worker instructions at the case company already suggested that working instructions are too elaborated and lead to an excessive amount of interrelationships. Therefore, the initial source of information had to be reworked to make it suitable for line balancing. However, it is currently unknown what the best way is to rework working instructions to eliminate the excessive information. The main problem is that it is unknown to what extend assembly tasks can be grouped whilst not harming the validity of line balancing results. Therefore, literature review was performed to find more information on assembly task grouping. Although, literature did not clearly handle the previously mentioned issue, general guidelines and directions were found and customized from previous literature on the creation of modular structures.

The worker instructions were considered as starting point during the definition of assembly tasks. Literature review suggested early on that the amount of tasks should be reduced. However, no sources clearly defined assembly tasks for line balancing purposes. In the initial resource (worker instructions) it can immediately be concluded that the assembly tasks considered in worker instructions have an elaborated view on what an assembly task entails. Figure 2 illustrates the way assembly tasks are arranged in worker instructions. They are focusing on showing new assembly line workers exactly what to do at a position standard. Furthermore, closely examining figure 2 suggests that assembly tasks within worker instructions have similarities. This indicates the possibility of joining tasks to minimize the amount of relationships. However, this immediately raises the question how many similarities are sufficient before separated assembly tasks can be joined together. For example, the only difference between the two tasks in figure 3 is the location at which the studs are placed at the front of the truck. In addition, all the other characteristics of these assembly tasks are identical.

(22)

22

Therefore, based on previous literature and discussions with the interviewees decisions were made whether assembly tasks are viable for the addition of multiple assembly tasks into a singular assembly task. Task enrichment can be partly based on the criteria set for modular structures: task functionality, equipment used for the assembly task and at which location the assembly task was performed. Figure 3 for example displays two tasks which are viable for the reduction into a singular task. These two assembly task are performed at the same location on the truck, the same tools are used for the two assembly tasks and finally the task functionality is exactly similar.

Finally, the initial thought of performing task enrichment based on assembly line worker input was rejected. It became clear that the assembly line workers primarily consider larger components (almost to the extent of a modular structure) as an assembly task. In their opinion, worker instructions consist of a small amount of tasks. If line balancing would be performed based on this particular input of assembly line workers the amount of assembly tasks would be insufficient to create an accurate representation of the reality.

Interviewing process and selection procedure

(23)

23

determined. Therefore, guidelines to determine what a good interviewee entails has to be outlined. Additional literature review was performed and will be outlined in chapter 5.

Determine penalties for valid LB results and identify decision variables

The initial research goal was to determine the accuracy and usefulness of manually created precedence graphs. This would have been done with the use of a simulation study. A long period of time of the research was dedicated to investigate the applicability of the precedence graph. In addition, a significant amount of time was also invested in finding the surrounding factors to enable simulation. Therefore, after the interviewing phase, a significant amount of time was invested in creating an accurate representation of the reality. Because of the fact that the assignment of the tasks cannot be performed randomly based on logical assignment rules.

The identification of decision variables is in line with the motivation of including penalties for line balancing. It was discovered for the creation of a valid line balancing tool. During the creation of the simulation study the following difficulties arose when composing the product input:

1. Certain assembly tasks are only “active” when a specific component is present on the truck. These components need to be filtered and identified on every single truck which is considered as the input of a simulation study.

2. Documentation and management of the components and linking them with the product input is a time-intensive process, especially if the identification of the component is not commonly used in available software.

Enlargement of the simulation area and merging process of simulation areas

The research set up was revolved around the extraction of a precedence graph on a set of workstations. However, the entire facility of the case company has over 50 workstations. Therefore, all simulation areas need to be linked with each other. This effect was simulated by the enlargement of the chosen three workstations and addition of two workstations to this initial sample. The following characteristics were found based on the enlargement of the simulation area:

1. Most of the precedence relations are based within a workstation/modular structure.

2. The first and final task of a modular structure is in a high number of times one of the tasks which needs to be focused on during the extraction of interrelationships with other simulation areas.

(24)

24

5 Framework

5.1 Managing the assembly task mess

In Chapter 3, it was introduced that Line Balancing at Scania is currently performed under the supervision of team leaders. They identify and report exceedances of the cycle time restrictions. Afterwards, they explore together with engineers the possibilities of task allocation. The main goal of a team leader is to minimize the stop time on their own workstation. The side-goal (if possible) is to achieve this objective by minimizing the variation of workload between stations, however this is not their priority. The current approach is considered as suboptimal and reactive. As previously stated, the available position standards are not documented for the purpose of Line Balancing. Therefore, reaching an optimal selection of assembly tasks for workplaces is currently not possible. Subsequently, this paragraph will show how to achieve a precedence graph suitable for Line Balancing in real-life scenarios based on the observations outlined in Chapter 4. The framework depicted in figure 4 shows the steps from dividing the assembly line into modular structures up to correctly updating the precedence graph. The framework is decomposed out of three main stages:

1. Determining simulation parameters: the determination of the area of simulation and an initial educational guess on workstation identity and modular structuring.

2. Gathering data: extract precedence relationships and verify the initial educational guess on modular structures and workstation identities.

3. Validation phase: update the precedence graph and maximize the simulation area to increase the usefulness of the Line Balancing results. Furthermore, in this phase the addition of penalties between positions will be determined to increase the validity of the simulation study.

Figure 4: General steps to create a precedence graph in a real-life scenario

5.2 Stage 1: Determining simulation parameters

5.2.1 Creation of modular structures

(25)

25

multiple interview questions. Each workstation or position standard will be divided into modular structures. Based on this information an appropriate decision can be made about the area of simulation.

The importance of modular structures has been emphasized in previous paragraphs and chapters. However, guidelines to achieve correct modular structures are hard to define. The connection between modular production and Line Balancing has not been made in previous literature. Therefore, the main difficulty in determining correct modular structures is focused around achieving the appropriate size. An excessive amount of modular structures means that the reduction of interviewing questions will be limited. However, a low number of modules leads to a high number of precedence relations inside the modular structures. In this particular case, the time to obtain the precedence relationships inside the modular structure itself will be unnecessarily long.

The tasks within the modular structures should have a certain amount of correlation such that modular structures can be used to generate useful precedence relations. The correlation between tasks is hard to measure, because the interpretation of the user of the framework plays a role. Therefore, the method needs to be structured by standardized rules. This to create the final modular structures. These standardized rules are outlined in the following guidelines:

1. Baldwin and Clark (1997) mention that modules should be independent from each other. Therefore, independent tasks are required to be a positioned in a single module. Furthermore, this allows the user of the Line Balancing tool to choose between balancing between assembly tasks or modular structures.

2. Product architecture can be used to come up with modular structures. If the tasks are all related to one functionality of the final product, then these tasks can be grouped together in a single module. For example, grouping assembly tasks which are all correlated to the assembly of a wheel to the chassis.

(26)

26

other. This knowledge will increase the likelihood of selecting an appropriate area of simulation.

5.2.2 Determine the area of simulation

The area of simulation should be approximately 5 workstations. This limited number of workstations is chosen to decrease the error-proneness of mapping precedence relations manually. Furthermore, the practical reason behind this choice is that it will be easier to assign a single person to determine the precedence relationships on a specific area. Whereas, if the area of simulation becomes larger the number of precedence relationships increases substantially, even to a point that it becomes too much to be done by a single person. This is caused by the fact that the user could easily lose overview of the considered assembly tasks and the corresponding precedence relations. However, the amount of workstations is solely based on practical experiences. This means that the amount of workstations depends on the capacities of an individual to comprehend with a large amount of variables and interrelationships. Another benefit of dividing the assembly line in smaller areas is that the framework can be performed at each simulation area simultaneously. This will decrease the total time necessary for the creation of the precedence graph. The selection criteria for the area of simulation is based on the degree of freedom between the modules. This entails a predetermined number of independent tasks to ensure that the Line Balancing heuristics have sufficient assignment possibilities. This predetermined amount of independency is determined by the size of the area. 5.2.3 Determine the identity of a workstation

Falkenauer (2005) describes this phenomenon as the fact that each workstation has its own characteristics. Examples of these characteristics are space constraints, presence of heavy equipment, capacity of supplies and operational restrictions. Therefore, in this step the identity of the workstation will be reviewed and an educational guess will be made on the previous introduced characteristics. This is important because these characteristics determine which assembly tasks need to be performed at a predetermined position at the assembly line. Furthermore, it could limit the assignment possibilities if certain assembly tasks cannot be performed at other workstations.

5.3 Stage 2: Gathering Data

5.3.1 Interviewing: verifying modular structures

The first step of interviewing starts with validating the accuracy of the initial modules. The modules will be adapted based on domain knowledge and the guidelines for constructing modular structures. Based on this information, the decision can be made to reconsider the division of initial modules. Furthermore, the initial modules will have a higher accuracy over time due to the iterative manner of the creation of the final modules on the area of simulation. In addition, the reconsideration of modules can contribute to a reconsideration of the area of simulation.

5.3.2 Interviewing: Reducing assembly tasks

(27)

27

Although, there have been general guidelines deduced from empirical research at the case study, a logical interpretation to the area of simulation is necessary in extraordinary situations. In other words, the guidelines can be extended to cover case specific situations.

The reduction of assembly tasks is required because the number of tasks need to be in a suitable range for gathering data. As previously mentioned, a high number of tasks leads to an exorbitant amount of time necessary for the extraction of precedence relationships and documenting tooling restrictions. Furthermore, a low number of tasks reduces the freedom of assigning tasks to workstations. It is advised to follow all the guidelines below and verify the number of tasks. The detailed guidelines for reducing tasks are correlated with guidelines to structure modules. However, the guidelines displayed down below show the following minor differences:

1. Directly related tasks with the use of heavy equipment can be deduced to ensure a safe working environment. The handling tasks without usage of the equipment like stripping of cables and tubes can be kept separate.

2. Independent assembly tasks are located in a single module. Therefore, these tasks need to be distinguished from other assembly tasks at all times.

3. Small assembly tasks (task time <5 sec) with no significant operational difference can be grouped together. This is done because of efficiency reasons. Possible gains by assigning these tasks to other workers would be neglected by non-value added assembly tasks.

4. Large assembly tasks (>50 sec) which are followed up by correlated small assembly task (<7.5 sec) can be grouped together. The correlation is based on technological or operational similarities. This guideline is introduced because of efficiency reasons, the non-value adding time of walking of other workers to get to the work location would neglect all possible gains. This guideline is only applicable if guideline 5 is satisfied.

5. Assembly tasks can only be deduced if the set of considered assembly tasks count for the same product variants.

6. Pearce (2015) emphasized that some assembly tasks need to be performed at a specific location. Therefore, these assembly tasks are not allowed to be assigned to another location. This immediately suggests the opportunity to group these tasks together. This process is allowed because it will not hinder the assignment possibilities.

7. Assembly tasks which count for each worker at all times, for example reading the worker instructions and the final check of all assembly tasks can be removed. These assembly tasks have no variation to other stations and therefore cannot be interchanged with each other. The cycle time can be lowered to ensure that the simulation study remains reliable.

(28)

28

adapted to a four-step guide. This is done to ensure that all the possible hurdles encountered during interviewing can be overcome.

1. The first step to collect valid information from interviewing is the selection of the interviewee. The interviewee should at least be a highly skilled assembly line worker or a team leader. This can be judged by an initial competence evaluation based on the information from supervisors.

2. Secondly, the capabilities of the interviewee should be considered and evaluated. This should be done because an intellectual gap between the interviewee and interviewer is not uncommon.

The evaluation can be done based on an introductory meeting in which Line Balancing and data requirements will be explained. After the introductory meeting, the compatibility of the interviewee can be determined by the interviewer.

3. Thirdly, the interviewing procedure can be started. The interview style should follow the requirements and method set by Otto and Otto (2014). They proved that the usefulness of the interviews becomes higher when non-repeating question are used. This is important to mention because the extraction of precedence relationships is built around one single question:” Is there a precedence relation between process a and process b? “(Altemeier et al., 2010). This can be obtained by constantly rephrasing previous mentioned question. Furthermore, the interviewer should try to understand the processes on the area of simulation. This will not only introduce more diversity in the questions, but this will also lead to more knowledge about the production processes themselves. The increased knowledge can be used to critically validate the results from the interviews and the simulation study. 4. The final step of the interviewing framework is to verify the obtained results

from the interviewee. This can be done by questioning the obtained interview results. The importance of the verification is proven by Farr and York (1975). They found out that early influences in the interviewing process tend to have more effect on the final judgement of the usability of the interview.

5.3.4 Interviewing: verification of the workstation identity

The characteristics of the workstation need to be evaluated. The most important characteristics can be identified based on an educational guess. However, the domain experts could validate whether or not the initial guess was justified. Domain experts are defined as highly skilled assembly line workers/team leaders. They have more than sufficient knowledge on a specific area of simulation because of their experience at the assembly line. This process is one of the least demanding processes in the entire framework. However, this process is necessary to validate the initial defined workstation identity.

5.4 Stage 3: Validation phase

5.4.1 Determine penalties to ensure valid Line Balancing results

(29)

29

of penalties between workstations. The reason behind these penalties is that independent assembly tasks could be assigned to workplaces which have the lowest amount of accumulated time. Theoretically this would have been perfect. However, assembly tasks could also be assigned to workplaces at the other side of the facility. This would mean that it is impossible to perform this assembly task within the clocked task time. Therefore, a matrix needs to be developed including all the walking times between work stations. Based on this knowledge, a simulation study can be provided with the input to determine whether or not it is useful to assign a task to a position located at another workstation. The task times included in the matrix can be multiplied with a safety factor. This safety factor should cover the adjustments of an assembly worker to the identity of a new area of simulation.

In addition, another penalty should also be introduced. Pearce (2015) described another requirement for valid Line Balancing results which needs to be satisfied. He mentions that ergonomic considerations need to be taken into account. The Line Balancing solution should not pair too many assembly tasks that require the assembly line worker to push his physical limits. Therefore, each assembly task needs to be categorized based on physical intensity. The practitioner of the framework has the freedom to introduce appropriate boundaries.

5.4.2 Identifying the decision variables for product input

The high amount of product variation has led to a significant amount of assembly tasks. However, not all assembly tasks have to be performed for each production variant. Every production variant has its own assembly sequence. Furthermore, the differences of these assembly sequences are determined by decision variables. For example, the engine stroke volume determines which assembly tasks will be necessary for the engine placement. Therefore, the decision variables need to be recognized based on the differences within position standards.

Historical data can be reviewed in combination with the decision variables. The historical data should include a list of previous produced Automotive vehicles (at least 3 months of data, to ensure that a high percentage of variants are encountered). The decision variables could indicate whether or not an Automotive vehicle was composed out of components which influenced the assembly task sequence. Based on this information, the Line Balancing tool can determine the optimal line configuration on the area of simulation.

The identification of the decision variables can be extracted from worker instructions with low difficulty. However, obtaining the products related to the decision variables is a more difficult process. The common components are known by product engineers and can be easily found. However, some of the components are ‘exotic’ for product engineers. This means that the decision variables need to be searched manually in the available software, which caused this process to be a time-consuming process. 5.4.3 Scaling up the simulation area

(30)

30

will have two primary goals: validation of obtained precedence relations and merging modular structures of different areas.

Focus groups are considered to be a suitable method to discover participants’ feelings, values, attitudes, reactions and experiences about a topic (Gizir, 2007). This is a suitable method to interchange information between the users of the framework. The method has a significant amount of freedom for the participants’ interpretation. Therefore, focus groups can contribute in the creation of a documentation structure which suits the Industry of interest. This will limit the recording differences precedence relations in the future. Furthermore, current differences can be adapted such that the recorded precedence relations are correct. In addition, the structures will be compatible to introduce the following step: merging modular structures. Otto and Otto (2014) showed that modular structures can be used to identify precedence relations between large sets of assembly tasks. This characteristic will be manipulated to scale up the simulation area. The interconnection between the areas of simulation is bounded to the precedence relations between the multiple modular structures of the two areas of simulation.

The merging process between modular structures will be a relatively straightforward process. This is due to the fact that the lay-outs of assembly lines are usually arranged based on technological sequencing requirements for the assembly of a final product. In addition, the amount of modular structures for an area of simulation will be relatively low. A high percentage of modular structures will have precedence relations within the boundaries of the simulation area. Therefore, the amount of precedence relations which are needed to be discovered are relatively low. The domain experts have the knowledge available to easily extract the precedence relationship between the modular structures.

However, difficulties may arise in obtaining precedence relations between independent modular structures. Independent tasks have their own modular structure. Thus, these modular structures only entail a singular task. However, there exist a high probability that these independent tasks have a precedence relationship which is located before the area of simulation. Therefore, during the focus groups these interrelationships need to be discovered. This should not necessarily be an issue when simulation areas are adjacent. However, when the precedence relationship is located beyond adjacency, the extraction of precedence relationships may be significant harder. This could be one of the disadvantages of the proposed framework. This is caused by the fact that the cooperation and knowledge of all the simulation areas may be insufficient. The reason behind this insufficiency is that it is currently unknown if each simulation area has the knowledge about which successors are on their area of simulation.

5.4.4 Updating the precedence graph

(31)

31

(32)

32

6 Empirical results: testing the framework

The created framework was tested to verify whether or not a precedence graph can be created based on a set of manual approaches. The hypothesis was confirmed to be true. Side note: the current framework has only been tested with a sample of 5 workstations. This significantly limits the validity of the results because of the fact that the interrelationships and variables become higher which significantly increases the difficulty to identify and manage precedence relations.

The following conclusions can be drawn based on the empirical evidence created in this research:

1. Modular structuring indeed significantly reduces the time for interviewing. This because large components can be easily linked to each other. Thus, the potential of modular structures was proven in reality. In addition, the applicability of modular structures made it easier for assembly line workers to identify the precedence relationships themselves. This was based on the observation that assembly line workers think in larger chunks when memorizing assembly tasks.

2. The creation of initial modular structuring were challenging based on the outlined guidelines because there is room for debate on the size and amount of modular structures considered within a worker instruction. However, the validation of modular structures should suffice for the correctness of the final modular structures.

3. Task enrichment (task grouping) cannot be performed based on assembly worker input. However, the guidelines in the framework proved to be of significant use. These guidelines reduced (based on the sample) the amount of assembly line tasks with approximately 40 percent. This was vital for the application of the manual approach. The guidelines were literature and experience based. However, the results of the created precedence graph has not yet been tested in a simulation study. Therefore, the validity of the considered assembly tasks has not been verified yet.

4. The validity of the generated precedence relations were considered to be high. This because multiple assembly line workers agreed with the created precedence graph. The introduced “safe gates” (multiple verification checks over the results by assembly line workers) to ensure correct results take up a significant amount of time. However, these so called “safe gates” are necessary because of the high error sensitivity of manual approaches, especially in the documentation phase.

5. The interviewing approach created by Otto and Otto (2014) proved to be of high use and this approach should be widely accepted for the manual extraction of precedence relations.

(33)

33

8. During the case study at Scania a significant amount of work place changes had been introduced. This confirms previously mentioned source

(Swaminathan and Nitsch, 2007) for the need of flexibility to manage

(34)

34

7 Generalizability of the results

In the methodology phase, it was stated that the aim of this paper was to reach results which are useful for the entire Automotive industry. Therefore, the generalizability of the results needs to be evaluated. Hammersly, Gomm and Foster (2000) emphasize that empirical generalization is possible without problem when the case study is (almost) identical to other cases in the Automotive Industry. Furthermore, Tang (2017) states that the characteristics of the Automotive industry are similar within the entire Industry.

Almost all Automotive manufacturers have the following similarities: 1. Problems in documenting precedence relationships.

2. A similar production environment and thus similar tooling restrictions. 3. Approximately the same preceding modular structure due to technological

requirements.

Following above outlined similarities, the generalizability of the results would be valid. However, the following differences between the case study and the Automotive industry can be identified:

1. Each assembly facility is unique and has its specificity.