ALLOCATION OF CROSS

(1)

A

LLOCATION OF CROSS

-

TRAINED WORKERS IN MAKE

-

TO

-

ORDER

ENVIRONMENTS

:

THE ASYMMETRY OF PRACTICE

M

ASTER

’

S THESIS

Author Dennis Douna

Student number S2153343 (NL) / B4062623 (UK)

Address Buursterlaan 53, 8435 XM Donkerbroek, The Netherlands Telephone +31 6 50 26 54 17

E-mail address d.douna@student.rug.nl / dennis.douna@outlook.com

First supervisor Dr. M.J. Land m.j.land@rug.nl Second supervisor Dr. Gu Pang

gu.pang@ncl.ac.uk Company supervisor Mr. J. van der Heide

Study Dual Degree Msc Operations Management

Universities University of Groningen, Faculty of Economics and Business Nettelbosje 2, 9747 AE, Groningen, The Netherlands

Newcastle University Business School

5 Barrack Road, Newcastle upon Tyne, NE1 4SE, United Kingdom

(2)

1

A

BSTRACT

Cross-training is an important tool for make-to-order job shops to deal with demand variability. While cross-training is extensively highlighted in literature, many companies struggle to adopt an active worker allocation policy. This is partly caused by different asymmetry aspects in companies, while there are also various factors that affect the adopted allocation policies. This paper introduces five factors that can complicate these allocation decisions: (i) learning and forgetting, (ii) job preference, (iii) workload control, (iv) cross-training patterns and (v) asymmetry aspects. A single case study has been used to investigate the constraining effect these factors can have, while two additional factors were unveiled: (i) responsibilities and (ii) capacity representation. A design that incorporates four cross-training related variables in the capacity representation factor has been presented, such that this factor better supports the worker allocation process. Finally, the findings on the various factors have led to a number of prerequisites, which help companies to exploit labour flexibility.

Keywords: cross-training, job shop, worker allocation, asymmetry, practical constraints

1 I

NTRODUCTION

Due to increased customer demand for variety along with the prevailing uncertainties, flexibility has emerged as one of the most important approaches for achieving a competitive advantage in the manufacturing industry (Liu & Li, 2012). In companies with a highly volatile demand, simply increasing capacity by hiring new workers or investing in new machines is often not beneficial and comes with risks of expensive underutilisation, while this strategy is also unsuitable for short-term adjustments (Francas, et al., 2011). To be flexible on the shorter term, labour flexibility is the most common alternative (Francas, et al., 2011), within which cross-training is a popular source of flexibility (Olivella, et al., 2013). A cross-trained workforce can bring great advantages to companies, but exploiting these advantages to utilise flexibility is a complex process (Brusco & Johns, 1998). Due to this complexity and the fact that each situation demands unique decisions, it is interesting to investigate how to exploit a cross-trained workforce in practice.

(5)

4 and the effect of learning and forgetting (Kim & Nembhard, 2013). After taking these factors into account, it is ultimately about when to assign a worker to a job, where to assign this worker and who is the worker to assign (Bokhorst, et al., 2004).

Current research focuses on situations without strong inequalities between departments regarding the types of workstations and the level of cross-training. It is however important to notice that job shops are increasingly investing in advanced CNC-machines (Das & Narasimhan, 2001), resulting in a situation where companies can be partly Dual Resource Constrained (DRC) and partly Single Resource Constrained (SRC), while also the role of workers varies over the various workstations. Besides this variation in workstations, the cross-training level of workers may also differ between departments, resulting in a so called multi-level flexibility (Felan & Fry, 2001), which means that departments and workers possess unequal numbers of different skills and that skill levels are varying. The above mentioned practical deviations from perfect theoretical situations make that the theoretical contributions made in literature so far, are difficult to transfer into practical environments. For the purpose of this paper the above mentioned unbalances, inequalities and variations will be referred to as asymmetry aspects.

The goal of this paper is to investigate how various factors affect the worker allocation process and how they should be arranged to facilitate an active allocation policy. Special attention will be given to the impact of asymmetry aspects on the feasibility of the allocation rules that are proposed in literature. The research will be executed by means of an extensive literature review of the field of cross-training, to find out which factors influence the allocation of a cross-trained workforce and how these factors could be used to facilitate the allocation process. Successively, a single case study will be conducted in a make-to-order job shop, to investigate how the allocation process can be constrained by the mentioned factors and if any other factors play a role. New factors will be analysed, after which several design propositions will be introduced that should help to transfer these constraining factors into facilitating factors. The findings of the case study will ultimately lead to a number of prerequisites for the various factors that stimulate the execution of an active allocation policy.

(6)

5 including asymmetry aspects, on the allocation process. A fifth section will present the design of a newly unveiled factor, after which a conclusion, including limitations and subjects for further research, will end this paper.

2 T

HEORETICAL BACKGROUND

The aim of this section is to create an overview of factors influencing the utilisation of a cross-trained workforce in flexible production environments. This section starts with a short overview of the process of creating a cross-trained workforce, after which the general process of worker allocation will be covered. With these steps as a base, different factors will be introduced that influence the decisions on how to allocate a cross-trained workforce. The section will close by presenting a conceptual model, which will be the starting point for the case study in the remainder of this paper.

2.1 Cross-training, the general principles

Both the research and practice of cross-training consider two separate questions, being the question how to cross-train a company’s workforce and how to assign cross-trained workers to the rights tasks (Sayin & Karabati, 2007). Before assignment questions become of importance, a cross-training policy should be developed, on which Bokhorst et al. (2004) identified five main concerns, being the extent of cross-training, chaining, multi-functionality, machine coverage and collective responsibility. Yang (2007) has compared different cross-training policies and, together with different other studies (Nembhard, 2014; Yue, et al., 2008; Campbell, 1999), came to the conclusion that the degree of cross-training has a strong impact on costs and efficiency, meaning that it is not beneficial to cross-train workers beyond a certain amount of skills. Besides the influence of a cross-training policy on the costs and performance, there are also several variables that influence a policy, such as the variety of products a company makes (Yue, et al., 2008), but also learning and forgetting effects (Nembhard, 2014) and behavioural issues, such as motivational problems due to the loss of uniqueness and responsibility (Nembhard, 2007). All these issues and variables show that defining a cross-training policy is a rather complex process that is unique for every company. This also makes the assignment of a cross-trained workforce a complex task, which will be elaborated in the remainder of this section.

(7)

6 investigated the where-rule several decades ago, by comparing five different strategies to decide where to assign a worker: 1) a strategy of random allocation, 2) the workstation which first comes, gets first served, 3) the workstation with a job that was first in the queue, gets served, 4) the workstation with the shortest operations time, gets served and 5) the workstation with the most queueing jobs, gets served. Hottenstein and Sherry (1998) did an extensive literature research on the findings on these strategies, showing that the fifth strategy has the most positive impact on the combination of mean queueing time and queue variance. Besides the where-rule, the when-rule is also a relevant decision element, which can be divided in push- and pull rules (Treleven, 1987). Treleven provided two push rules of labour flexibility: 1) enabling to transfer a worker after the completion of his current job (central when-rule) and 2) enabling a transfer when a workstation is in idle mode (decentral when-rule). On the other hand, Treleven (1987) provided a number of pull rules on when to free a worker for a transfer, which were based on timing variables, such as the time a job has spent queueing. Even though the when-rule is a prerequisite for the where-rule, its influence on performance is smaller (Hottenstein & Bowman, 1998; Treleven, 1987). While the where- and when-rule are rather well-known rules, the who-rule is more recent. Bokhorst et al. (2004) found that a who-rule was missing, while it was of particular relevance in realistic situations, such as where workers have different task proficiencies or different numbers of skills. In general the who-rule is about which worker to assign to a specific task. Bokhorst et al. (2004) compared different strategies to decide who to allocate and showed that allocating workers based on priority is the most effective method, which means that a job is always allocated to the worker with the least workload on his unique skill. The research showed that this strategy resulted in the best workload balance and the lowest flow-time of products. It is however important to realise that it is in many situations not possible to select a worker from a pool, since often only a limited number of workers is available.

In general all, the where-, when- and who-rule, are important decisions to make in order to allocate a cross-trained workforce. This research does particularly focus on factors affecting these decisions, which is why the next part of this section will elaborate on literature about the available factors.

2.2 Factors affecting the allocation process

(8)

7 however given more indirectly, with studies only mentioning the link between the factors and the allocation process. In the remainder of this theory section, five of the most relevant factors will be covered by explaining the factors in detail and elaborating on the link with the allocation process.

2.2.1 Learning and forgetting

Sayin et al. (2007) stated that learning and forgetting characteristics should be taken into account when allocating workers to jobs, since they have a significant impact on a system’s performance. For example, frequent shifts of workers between tasks, often results in a loss of productivity due to learning and forgetting effects (Kim & Nembhard, 2013). In general, a company should already consider the presence of learning and forgetting during the training of a cross-trained workforce and adapt the cross-training policy on this, since a high level of learning and forgetting might result in a situation where an increased degree of cross-training leads to a decrease of performance (Yue, et al., 2008). This means that the allocation of a cross-trained workforce does also affect a system’s performance, since people and tasks have different learning and forgetting characteristics. Therefore, Nembhard (2001) came with a heuristic approach, to assign workers based on their learning rate. This research showed that assigning quick-learning workers to short-run jobs and slow-learning workers to longer production runs, resulted in a significantly better working productivity. This shows that learning and forgetting aspects can have a high influence on performance and should be included in the allocation process of a cross-trained workforce, while they could also be already considered during the development of a cross-training policy. Differences in task complexity could for example be a good reason to create asymmetry in the cross-training policy, by keeping specialist workers for complex tasks (Kim & Nembhard, 2013). However, for an asymmetrically cross-trained workforce, limited flexibility might constrain companies to allocate workers based on their learning and forgetting rate.

2.2.2 Job preference

(9)

8 productivity due to the forgetting effect. One of the reasons for job preference, is given by Philipoom and Fry (1999), who mention the effect of ‘cherry picking’. They explain this as the effect where workers pick favourable jobs, which have standards (e.g. process-time) that are easily met. Furthermore, from a more psychological view, Xie and Johns (1995) point out that if workers perceive a misfit between their abilities and the skills demanded by a task, this can result in more stress, which can again be a reason why workers prefer one job over the other job. To cope with these job preferences, Bokhorst (2011) tested the effect of work in process levels, showing that a lower level of work in process resulted in a situation where workers balanced the use of their skills in a better way. However, Fernandes et al. (2014) showed that workload control rules, influencing the work in process, are more difficult to apply in unbalanced job shops, which hinders the use of these methods in many practical situations.

2.2.3 Workload Control

(10)

9 performance of WLC mechanisms, but they also show that the central when-rule can explode the number of worker transfers, which illustrates the relevance of the relation between the execution of workload control mechanisms and the allocation of workers. It is furthermore important to realise that the release of orders is not an independent process, as Land et al. (2014) state that it has to be coupled with priority dispatching rules, which are often based on due dates. This creates a useful link to the research by Olivella et al. (2013), who show the dependence between the allocation of workers and the given due dates.

2.2.4 Cross-training patterns and levels

Another factor, which is of particular relevance for the scope of this research, is the way workers are cross-trained at different departments. Yang (2007) mentions that cross-training workers across different departments, results in a better performance, since the allocation of workers can be adapted to changing workloads. Kher and Malhotra (1994) confirmed this by stating that cross-training workers in at least two departments is a better strategy than training them within only one department. There is however still discussion on the number of workers that should be cross-trained. Felan and Fry (2001), state for example that cross-training only a subset of workers is the best solution, while Slomp and Molleman (2002) suggest an evenly distributed skill level. In case of an uneven distributed level of cross-training, Yang (2007) divided primary workers, who are fixed to one department, and cross-trained workers. He suggests to assign cross-trained workers only to a department, when all primary workers at that department are busy. This shows that the level of cross-training across departments has influence on the allocation of workers, while also factors like transfer delays (Kher & Malhotra, 1994) play a role in this. Furthermore, Bobrowski and Park (1993) did research to workers’ differences with regard to their operational efficiency at different workstations and showed that these differences should also be taken into account when allocating workers to tasks. For example, companies should balance the need for more capacity at a workstation and the loss of efficiency this brings when a less skilled worker is assigned. They showed that applying a where-rule, allocating workers to their most efficient department, scores better than any other type of where-rule.

2.2.5 Asymmetry aspects

(11)

10 heterogeneous labour flexibility (Felan & Fry, 2001). Most papers about the cross-trained worker allocation process do however not take these asymmetry aspects into account and suggest this to be an important limitation of their study, such as Slomp and Molleman (2002) and Yang (2007). Beyond the scope of asymmetry aspects of a cross-trained workforce, there are other asymmetries that complicate the worker allocation process. Of particular relevance for advanced job shops is for example the increasing use of CNC-machines (Das & Narasimhan, 2001), which have different manual labour requirements compared to more conventional machines. With many more, often case-specific, aspects of asymmetry existing, this is a relevant factor to consider, since asymmetry complicates the process of adopting theoretical allocation policies in practice.

2.3 Conceptual model

In the second part of the theory background section, five factors have been introduced that affect the worker allocation process. The conceptual model in figure 2-1 gives an overview of these factors, influencing the exploitation of a cross-trained workforce. As the model shows, an allocation policy that is able to exploit the full potential of the available labour flexibility, should take the mentioned factors into account, for which they should be organised in a way that facilitates clever worker allocations, as suggested in the sections above.

In the remainder of this paper, the factors will be validated and complemented with additional factors found in practice, after which the case study will focus on uncovering the constraining effect the factors can have on the exploitation of labour flexibility.

3 M

ETHODOLOGY

The previous sections have gained insights in the general process of worker allocation and the way in which the five presented factors can complicate this process. To investigate how these factors can constrain allocation decisions in practice, a single case study has been conducted. A single case study enabled opportunities for in-depth research (Voss, et al., 2002), which was necessary to gain useful insights in the different complexities introduced by

Figure 2-1: Conceptual model of factors complicating the utilisation of a cross-trained workforce  Learning and forgetting

 Job preference

 Workload control concepts  Cross-training pattern and levels  Asymmetry aspects

(12)

11 asymmetry and to be able to identify and understand the impact of the factors. The following paragraphs describe the processes regarding the case selection, data collection and data analysis.

3.1 Case selection

For the purpose of this research, it was important to select a make-to-order (MTO) job shop, since MTO companies need flexibility in their production process in order to cope with demand fluctuations, while other types of companies often have opportunities to use inventories for this purpose. A suitable case company should furthermore have a cross-trained workforce that could be used for capacity flexibility and should face asymmetry aspects, such as an unequal level of cross-training between departments. The selected company suited these prerequisites well, as it was aiming for more internal flexibility, but faced some difficulties due to different asymmetry aspects, such as the cross-training level and variety of automated and manual processes. The company had an established cross-trained workforce, but its allocation policy was at a pre-mature phase, making it possible to investigate the constraining effect of the different factors on the yet unexploited labour flexibility. The selected company, located in The Netherlands, operated in the high-precision metal industry, serving mainly high-tech customers in the manufacturing-, medical- and space industry. The company gave place to approximately 100 employees and had an annual turnover of around 15 million euros. Due to its market position, the company faced extreme deviations in demand and could thus benefit from capacity flexibility.

Since the research mainly focused on the worker allocation process and those factors affecting this, the unit of analysis for this research was the worker allocation process of those workers that were involved in the main production process of the case company, which demanded a focus on supervisors and planning employees.

3.2 Data collection

(13)

12 get insights in the level of asymmetry, the complexities on the floor and the presence of the various factors. Afterwards, more structured interviews were held with planning employees and supervisors, to gain information about the current allocation policy, the reasons for this policy and the influencing factors. This was performed in a way that the interviewees first mentioned the factors they considered to be relevant, before they were asked about the factors from the conceptual model. Less focused interviews were used to gain insights in specific factors, such as the effect of job preference, while questions for clarification were answered by directly asking the relevant persons on the floor. These findings were combined with observations and short talks from shop floor tours, in order to create an independent view of the situation.

3.3 Data analysis

The data gathered during the study was analysed in different ways, in order to gain useful outcomes regarding the need for flexibility, the current allocation process, the influence of asymmetry aspects and the role of the various factors. Where possible, data from qualitative interviews was verified by quantitative data about productivity, workloads and throughput times, which helped to improve the validity through triangulation (Eisenhardt, 1989). The ERP production data was translated in Microsoft Excel, in order to create weekly and monthly overviews of workloads, order intake, capacity and throughput times, which helped to gain insights in the case situation. The data from the structured interviews was used to validate the conceptual model and expand it with additional relevant factors, while the key outcomes were compared with literature and used for follow-up interviews. Subsequently, the constraining role of the factors was identified by analysing the combination of interviews and focused observations, after which case specific improvements for these factors were discussed with the key informants. The role of asymmetry aspects was analysed by the investigation of the company’s skill matrix and machine details, combined with the practical limitations planning employees and supervisors faced in coping with these complexities.

4 C

ASE DESCRIPTION

(14)

13 worker could be the bottleneck, making this part of the organisation Dual Resource Constrained (Felan & Fry, 2001). The other part of the machines was partly able to run for several hours without manual intervention and partly able to run with manual attention every several minutes, both depending on the type of order that was being produced. The processing times varied significantly, with an average of 10 hours per process per order. Even though some processes occurred at the beginning of the routing and others at the end, no standard routings or fixed number of routing steps could be distinguished. Based on the company’s functional layout, high variety in processing times and the lack of fixed routings, it can be characterised as a typical job shop (Oosterman, et al., 2000).

Precision Milling Electrical Discharge Machining Assembly Measurement 16 machines 11 workers 9 machines 7 workers 10 machines 4 workers Not relevant 7 workers 6 machines 7 workers Mainly DRC Mainly SRC (machine) SRC (machine) SRC (worker) Mainly DRC

Table 4-1: Overview of the different departments with the type of constraint (workers per shift) 4.1 Required flexibility

The case company faced a strongly varying demand pattern, with unpredictable peaks and drops (Figure 4-1). To keep a production system stable, with controlled WIP-levels and throughput times, a flexible level of capacity is needed. The input-output diagram in figure 4-2 shows however significant fluctuations on the input side, while the output stays rather constant, behaving independently from the demand. As a result, the average throughput time was varying between 2.5 and 3.5 months.

0 2000 4000 6000 8000 10000 12000 Ord er in tak e (h o u rs o f w o rk ) 0 20000 40000 60000 80000 100000 120000 140000 Ho u rs o f w o rk

Input (hours) Output (hours)

(15)

14 Following the demand pattern exactly to keep throughput times constant, required unrealistic production peaks, such as a required production rate of 400% of the capacity and many idle times, which is not desirable. Figure 4-3, demonstrating the actual workload at departments, which is based on ERP-data about the work done compared to the standard capacity allocated to a department, indicates however several opportunities to shift capacity to decrease throughput time variety.

Even though the workload patterns show opportunities to shift capacity between departments and processes, interviews unveiled that in practice fluctuations in demand were mainly covered by three interventions: 1) overwork, 2) outsourcing and 3) a longer delivery time for new orders. The company’s cross-trained workforce was thus barely used to gain flexibility and only if problems with due dates occurred, this step was actively investigated. This is an interesting observation, as this was their main source of labour flexibility, since temporary flex-workers were not an option due to high skill demands.

4.2 Cross-training activities

Since a number of years, the case company introduced a 3x3 cross-training policy, where each worker should be able to perform three different tasks and each task could be performed by three different workers. Figure I in the appendix shows the current skill matrix, where workers are rather asymmetrically trained, mainly within their department. The cross-training plan was mainly introduced to become more robust for the absence of workers (e.g.

0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100% Util isa ti o n

Precision Milling EDM Assembly Measurement Capacity

(16)

15 vacation or illness), while it was also meant to avoid workers becoming idle when their workstation was empty.

The current allocation policy used by the company can be described as a passive decentral when-rule on workers’ primary skills and a central when-rule on workers’ secondary skills. In practice this means that workers kept working on their primary skills as long as there was sufficient work available. When the amount of work decreased to numbers close to zero, workers normally asked their supervisor whether they could assist at different tasks. This could result into three situations:

1) A worker was assigned to a different operation, which could be in a different department. Since workers were only limitedly cross-trained, workers were often assigned to basic jobs, such as assembly and quality checks, as these required less specific skills;

2) A worker was set to idle, which meant he would not perform any valuable operations. This time could be used for small maintenance or cleaning activities, but also brought an increased amount of social interaction between workers;

3) A worker was voluntarily asked to take days off, in case he had an overage on his amount of vacation days from previous years, which saved the company money.

Once a worker was transferred to a secondary skill, the workload at his primary job was leading for when to return, which meant they used a central when-rule here, enabling workers to return to their primary station after each completed job. In practice the workload did not extremely fluctuate within a day, which meant the decision to transfer a worker to a different operation usually lasted for the day, after which it was reconsidered for every other day.

In rare occasions, the cross-trained workforce was more actively used to adapt the capacity. This was however mainly when the company faced short-term problems to achieve the planned due dates. On these moments worker transfers were used as an ad hoc solution, to catch up with the planned orders, but without changing the capacity in the planning systems.

(17)

16

5 C

ASE ANALYSIS

In this section the different complicating factors, which were introduced in the theoretical background, will be investigated from a constraining point of view. The findings from the interviews and quantitative data analysis at the case company will be used to find out how these factors can limit companies in the adoption of an active worker allocation policy. At the end of this section, two additional factors are unveiled, which were found to complement the constraining factors at the case company.

5.1 Learning and forgetting

Learning and forgetting effects were found to play a role in the limited amount of worker transfers, but also in the maintenance of the company’s skill matrix. Selecting a worker based on his learning speed, as suggested by Nembhard (2001), was difficult for two reasons. First of all, the asymmetry of the skill matrix resulted in limited possibilities to choose from a number of capable workers. Secondly, it was considered to be difficult to get a qualitative overview of workers’ learning rates, let alone to catch it in a number, like Nembhard (2001) used in his heuristics approach. In periods of idle time, the learning effect was however found to play a role in the decision whether or not to transfer a worker from his primary task to another task. Interviews with workers unveiled that the learning rate was quite high, due to the advanced- and skill-demanding characteristics of most machines. When workstations became idle and workers were transferred to their secondary skill, this fact was however ignored, since the central when-rule made that workers had to return to their primary skill soon. In practice, this led to low efficiency levels and frustration under workers, which ultimately made workers reluctant to short transfers, resulting in a passive attitude towards the exploitation of labour flexibility. These findings imply that the emotional effect on workers when learning rates are not considered in allocation decisions can be significant, while the productivity based use of learning rates, as suggested in literature (Sayin & Karabati, 2007; Nembhard, 2001), is difficult to perform in practice.

(18)

17 effects are thus not covered by active guidance from planning employees and supervisors, it does not only impact workers’ learning speed (Sayin & Karabati, 2007), but can also have a significant effect on companies’ overall flexibility, which is accelerated by the continuous development of advanced machines. It can hereby lead to an increasingly asymmetrically cross-trained workforce and a loss of previously gained flexibility.

5.2 Job preference

The phenomenon of job preference was clearly present at the case company, with many workers having a preference for their primary machine, at which they felt comfortable and an expert, as explained by Bokhorst (2011). Combined with the fact that workers who were available for a transfer, were often allocated to very basic operations, such as assembly and quality control, resulted in a situation where workers had a strong preference for their primary job and some reluctance to transfers. Interviews with workers showed that this effect was the strongest for elderly workers with a long-lasting history at the company.

During interviews with planning employees and supervisors it was mentioned that this job preference was in practice visible by workers slowing their working pace, to avoid transfers. A comparison of ERP-data about the company’s efficiency levels (actual processing times versus pre-calculated processing times) and workload (WIP levels) confirmed this behaviour, showing a correlation of -0,45, as can be seen in figure 5-1. Workers stated that they slowed their pace partly to postpone transfers, but also to avoid idle time, which was often the result of the company’s passive allocation policy. This behaviour is also mentioned by Schultz et al. (2003) and is unbeneficial, since it postpones the visible factor of workers being available for a transfer, decreases the efficiency and makes capacity calculations inaccurate. With a more involving role of supervisors and planning employees, this effect could be decreased, by transferring workers in an earlier stage, before true idle timing is threatening.

0% 1% 2% 3% 4% 5% 6% 7% 5000 6000 7000 8000 9000 10000 In ef ficie n cy W o rk in p ro ce ss (h o u rs o f w o rk ) WIP Inefficiency

(19)

18 Another consequence of the rather passive allocation policy, where workers had a decent level of autonomy, was that they rarely asked others for assistance. Many workers felt their primary machine to be “their machine”, which made them distrusting others to get involved. Therefore the use of short-term flexibility by workers on the floor was limited, while attitudes towards cross-training were often negative, since workers felt it to be impossible for ‘outsiders’ to learn their skill.

In conclusion, the findings show that job preference does not only decrease the utilisation of available flexibility, as mentioned by Bokhorst (2011), but can also have a significant impact on workers’ productivity levels, which is not yet considered in worker allocation literature.

5.3 Workload control

While WLC-literature suggests a triple layer control of workloads (Fernandes, et al., 2014), the case company followed an immediate order release system. Incoming orders were directly processed by the Planning Department, by giving a start- and end date for each specific process. Subsequently, workers were responsible for making the given deadlines, but had freedom in the sequence in which they processed different orders, with the idea to combine orders to reduce setups. This autonomy in sequencing work, combined with the decentral when-rule on primary skills, made that workers started to work ahead in calm periods. This working ahead did however take away the visual trigger of a worker being available for a transfer, leaving this trigger to be only on the digital schedules.

Week 1 Week 2 Week 3 Week 4

100% utilisation Low planned workload Low planned workload executed

Figure 5-2: Comparison of a 100% planned workload and a low planned workload

(20)

19 shows that in a situation of 100% utilisation, workers have no freedom to cluster orders together, since this will cause delays. The second row shows however a situation with a lower workload, in which a backwards planning mechanism leaves free space between orders. As the third row demonstrates, in which this low planned workload is actually executed, workers start to cluster orders and keep working as long as there are jobs left. In such a situation a worker is still 100% productive at his workstation, but the capacity used to work ahead, could be more useful in bottleneck workstations.

In practice, the absence of a controlled work release system, not being covered by strict instructions from supervisors or planning employees, makes that opportunities to exploit labour flexibility are not identified. On top of this, the visibility of workload situations is reduced, which makes it difficult to oversee the potential benefits of worker transfers, since it feels like taking a worker from where he can be productive, to a place where he is less productive. This means that cross-training is not only a tool to improve the performance of workload control mechanisms (Fredendall & Melnyk, 1998), but that the presence of workload control rules is also a prerequisite to carry out a successful allocation policy.

5.4 Cross-training patterns and levels

Kher and Malhotra (1994) showed that cross-training workers in two departments is the best situation, but interviews with supervisors unveiled that this can be particularly difficult when advanced skills are required to operate machines. The skill matrix in Appendix I shows the cross-training levels for the case company, without an effect of chaining in which a pattern of overlapping skills exists (Brusco, 2008), and with limited cross-departmental skills. Interviews with supervisors unveiled that this limited flexibility led to a situation where a secondary worker was often not available when a workstation needed more capacity, while one’s secondary skill did often not demand extra capacity when this worker was available for a transfer, which together limited the possibilities to actively allocate workers. Focusing cross-training on the few workers with primary skills on tasks that can be handed over rather easily, as pointed by Felan and Fry (2001), would increase these opportunities. It was however found that especially these tasks were often performed by lower level workers, which were less capable to learn multiple skills.

(21)

20 mainly repeating orders, but beyond this distinction no details, such as efficiency differences, were available about the exact value a secondary worker could bring at his secondary workstation. This lack of detailed information limits the opportunities to judge whether a transfer is useful and if the lower efficiency at a worker’s secondary skill is outweighed by the priority of a job. This lack of insights in the consequences of worker transfers, made the case company reluctant to actively transfer workers. Decent documentation of workers’ task proficiencies and efficiency levels is thus not only a requirement to incorporate efficiency differences in allocation policies (Bobrowski & Park, 1993), but also a necessity to foresee the consequences of worker transfers in general.

5.5 Asymmetry aspects

As table 4-1 showed, caused by the increasing use of modern CNC-machines, the type of constraint for a process can vary over departments and between workstations, creating an important asymmetry to deal with. The practice of Dual Resource Constrained- and Single Resource Constrained systems has been extensively covered in literature (Yang, 2007; Kher & Malhotra, 1994), but the presence of varying constraints adds a significant complexity. Since software programming, tool preparation, machine start-up and regular machine operation are often executed by the same group of people, the mix of required tasks influences the extent to which a process is labour- or machine constrained. This was found to complicate the case company’s allocation decisions, which was enforced by the fact that the extent to which machines were able to run unmanned was also order dependent.

(22)

21 In conclusion, the asymmetry aspect introduced in this section, makes that standard when- and where-rules are inadequate to exploit full labour flexibility. This asymmetry especially complicates longer term allocation decisions, since detailed information about orders and production sequences are required. To exploit this hidden flexibility on a longer term, companies could for example implement fixed order sequences, by taking away workers’ autonomy to cluster orders. Short before production, detailed information about the required operations for orders is available, making that thorough communication between workers and planning employees could help to exploit the situation-dependent flexibility on a short-term.

The previous sections have extensively covered the factors that were introduced in the theoretical background. These factors were mainly focused on operations- and worker aspects, but during the research it was not possible to explain all forces that limited the company’s adoption of an active allocation policy with these factors. Observations and interviews showed that some process related factors also played a significantly constraining role. In-depth interviews on these factors unveiled that two process related factors could be distinguished: responsibilities and capacity representation. The findings on these factors will be presented in the two sections below.

5.6 Responsibilities

One of the two additional factors that were found, are difficulties with responsibilities about capacity. These responsibilities were shared between supervisors and planning employees, with supervisors being responsible for determining the workstations’ standard capacities and assuring that these standard output numbers were achieved in practice. Planning employees were on the other hand responsible for utilising the capacity effectively, while meeting customer demands. In practice this meant that short-term worker transfers, to absorb illness or ad hoc problems, could be made by supervisors solely, while longer term capacity decisions, based on demand patterns, were a shared responsibility of supervisors and planning employees. Such a shared responsibility requires communication, which can be difficult due to different interests. Interviews with supervisors and planning employees showed this to be a reason for lately- or even non-executed worker transfers, which could be improved by moving the responsibilities to one party.

(23)

22 which they were assessed by the management. As a consequence, supervisors focused on efficiency rates and preferred to maximise the productive hours, while the usefulness of these hours was of less importance. On top of this, worker transfers across departments were unbeneficial for supervisors, due to the loss of hours on their own department. Together, this contributed to a passive attitude of supervisors towards active worker allocation.

These findings show that the way in which responsibilities are divided and assessed, can have a significant impact on the practice of allocation decisions.

5.7 Capacity representation

Interviews with planning employees unveiled that the case company experienced difficulties to foresee the consequences of worker transfers on the available capacity. This effect was enforced by the fact that the capacity representation in the case company’s scheduling software did not distinguish machine- and labour capacity, resulting in a simplified representation of capacity. As a result of this, worker transfers were usually not translated in adjustments of the available capacity in the scheduling system and were therefore mainly used to absorb ad hoc disruptions from the planned situation. Employees mentioned that the simplified capacity representation complicated the use of labour flexibility on a longer term, such that it could be considered a constraining factor for active worker allocation.

The above mentioned simplification of the capacity representation at the case company was partly caused by the complex relation between worker- and machine capacity, with three main situations:

1) A worker keeps several machines running at the same time;

2) A worker keeps a machine running for 24 hours with only 8 hours of attendance; 3) A worker keeps a machine running only when he is actively processing this machine.

(24)

23

6 D

ESIGN FOR FLEXIBLE CAPACITY REPRESENTATION

The two newly identified factors are not covered in literature, which is why this section presents a design of the capacity representation problem, to facilitate a more active worker allocation policy. The factor about responsibilities is mainly based on a company’s strategic decision and will therefore not be covered in a separate design section.

6.1 Capacity in literature

In both the scheduling- and cross-training literature, capacity is of key relevance. The findings of papers about the productivity differences between workers (Brusco & Johns, 1998) and the effects of learning and forgetting (Kim & Nembhard, 2013; Nembhard, 2001), have for example direct effects on the output numbers companies can realise. Most studies are however limited in listing factors that influence companies’ production rates (Bokhorst, et al., 2004; Felan & Fry, 2001; Sayin & Karabati, 2007), while they lack propositions on how these factors should be translated to practical capacity representations. Even papers that incorporate some aspects of variability, such as heterogeneous processing times (Bokhorst & Gaalman, 2009), do not consider the practical implications for the planning process, while section 5.7 in this paper has shown that variability is a key complexity in capacity dimensioning.

To fill this gap, this section will present alternatives to incorporate four cross-training related aspects in companies’ capacity representation. The lack of two aspects, the “connection between workers and machines” and “efficiency differences”, were found to be the main reasons why the capacity representation was found to be a constraining factor on the adoption of an active allocation policy. The third aspect, “distinction of skill requirements”, was included because interviews with operational workers unveiled the highly order-dependent skill requirements for the operations of machines. Finally, a simulation in the case company’s planning system to investigate the effect of automatically generated worker transfers introduced a problem of uncontrolled random allocations, resulting in the fourth aspect: “worker-task priorities”. To design the integration of these aspects in the capacity representation, such that it facilitates both short- and long-term worker transfers, unstructured interviews were held with planning employees, while the researcher simulated the worker allocation process based on historical data in the case company’s planning system.

(25)

24

6.2 Connecting workers and machines

A key reason of the constraining effect of “capacity representation”, was the missing connection between workers and machines. This is however of high relevance, since in Dual Resource Constrained environments both labour- and machine capacity can be a bottleneck (Felan & Fry, 2001). A machine’s capacity is dependent on the attendance of a skilled worker, which can have different effects on CNC-machines than on conventional machines, as section 5.5 about asymmetry has shown. To exploit the labour flexibility that comes with partly unmanned machines, the relation between worker attendance and workstation capacity should be covered in a factor. Such a factor should represent the linearity of this relation, such that a factor 1 corresponds with a conventional machine, where a workstation’s capacity equals a worker’s attendance, while a factor of for example 0.5 could correspond with a partly unmanned operating machine that only needs to be occupied half of the time. The use of such a factor should be combined with two important constraints, being the availability of a machine per day and the duration of a working day.

The factor introduced above gives a rough indication of a company’s longer term capacity, which is particularly useful during order acceptance. On a medium-term, the factors should be more accurate, by incorporating the interval at which a worker is needed at a partly unmanned machine. At some machines a worker is needed every 30 minutes, while other machines only require a worker at the beginning and end of an order, which has different consequences on the allocation of a worker to different tasks within an order.

However, on the short-term, a superficial factor might not be sufficiently accurate, which is why manual decisions regarding the worker allocation are still needed. On this term, workers’ feelings and efficiency consequences due to learning and forgetting (Kim & Nembhard, 2013) should be taken into account in considerations about the number and duration of transfers. Furthermore it is important to review the feasibility of suggested allocations of worker to multiple partly unmanned machines, since this can be constrained by the physical distance between machines.

6.3 Efficiency differences

(26)

25 that it can be difficult to capture efficiency in a number, this data should be available in most companies’ ERP-systems. Since these numbers affect both the worker allocation decisions and companies’ achievable output, they should be incorporated in the capacity representation.

Section 5.7 has demonstrated that the manual adjustment of workstations’ availability to incorporate efficiency differences, does not facilitate active cross-training. It reduces the availability of a machine, instead of increasing the processing time to perform a task, which makes it an unrealistic interpretation of efficiency differences. A more realistic and flexible solution would thus be to absorb efficiency differences in the processing times of orders. This means that as soon as work calculations for an order are made, which is often before an order is accepted, the efficiency numbers can be used in the planning system and used in worker allocation decisions. One way to implement this, would be to calculate different processing times for all workers that are capable to perform a task, which is an accurate, but time consuming solution. A less time consuming solution would be to use a general factor per worker-task combination, to compensate the standard processing times for efficiency differences, such that the required processing time for a less efficient worker is more than for a 100% efficient worker.

Incorporating efficiency in this way, enables automated planning algorithms to take efficiency into account, while the factors are useful on both the long- and short-term, since the information they are based on does usually not change from the moment of order acceptance. For the shorter term, efficiency levels could however be manually reviewed for orders that are expected to show significant deviations from the standard efficiency differences between workers.

6.4 Distinction of skill requirements

(27)

26 however much extra complexity to the planning system, reducing the system’s accessibility. Besides, when different activities on an order at one machine are performed by different workers, this increases the risk of mistakes due to extra communication demands. An option without these difficulties, is to separate more skill-demanding orders from less skill-demanding orders, such that high skill-demanding orders can only be performed by highly skilled workers, while other orders can also be executed by secondary workers. The case company made a distinction between new orders and repeated orders, which could be a useful distinction for worker allocation as well, since repeated orders do usually not require advanced research and programming anymore and can be performed by secondary workers. This distinction mainly assists from the moment of order acceptance up to the short-term planning, after which manual intervention is needed to consider more advanced order specific details.

6.5 Worker-task priorities

During the researcher’s test with the case company’s planning system, an effect of random worker allocation occurred, where secondary workers were allocated to tasks, while primary workers were also available. For most allocation decisions this effect is prevented by workers’ efficiency differences, but for simple tasks efficiency levels are often equal, while many workers are capable to perform these tasks. Also for advanced workstations this effect can occur, since efficiency differences are off less importance when workloads are low. Random allocation is however considered to be undesirable, since it creates unnecessary movement and leads to dissatisfied workers.

Table 6-1: Priority ranking on machine- and worker level (efficiency within brackets)

To prevent this randomness, a priority rule could be used to distinguish the desirability of worker-task combinations. Bokhorst (2004) suggests a who-rule based on priority, such that workers with the least workload on their unique skill are allocated. This rule does however insufficiently cover the effect of randomness for tasks where multiple workers without unique skills are available. To incorporate this, a priority ranking could be used for machines and workers, as can be seen in table 6-1. Such a priority ranking enables a planning system to make controlled short- and long-term allocation decisions, even when no objective trade-offs can be

Machine 1 Worker 1

1 Worker 1 (1.0) 1 Machine 1 (1.0)

2 Worker 2 (0.8) 2 Machine 2 (1.0)

3 Worker 3 (0.8) 3 Machine 3 (1.0)

(28)

27 made. It thereby contributes to a better performance by using cross-training only when all primary workers are occupied (Yang, 2007).

6.6 Design discussion

The presented solutions to incorporate the four cross-training related aspects in the capacity representation, support planning employees in short- and long-term worker allocation decisions. By covering efficiency and worker-task relations in estimated factors, automated planning algorithms can also use labour flexibility, which is particularly interesting from a long-term perspective. Furthermore, the design to separate skill requirements and include worker-task priorities leads to more robust allocation decisions. For shorter term planning decisions, the factors might however be too rough, since they are based on generalisations. On this term, more details about order characteristics and worker demands could be processed manually, to improve allocation decisions. To conclude this section, it is important to note that the suggested four aspects are generally applicable in most job shops, but additional aspects might be desired in different cases.

7 D

ISCUSSION AND CONCLUSION

This paper started by introducing the different applications and rules of cross-training in order to gain capacity flexibility, while five factors were presented that can assist or complicate worker allocation decisions. A case study in a typical job shop with different asymmetry aspects, has shown that the five factors can also limit the exploitation of a cross-trained workforce. Two additional factors were found, being the responsibility for capacity and the capacity representation. The study specifically demonstrated the consequences of different asymmetry aspects, while the integration of four cross-training related variables in the capacity representation factor was discussed, which improved this factor to better suit an active worker allocation process.

7.1 Managerial and theoretical implications

(29)

28 in- and between departments, makes the application of the where- and who-rule difficult, due to the limited amount of capable workers. A concrete consequence is that it hinders companies’ abilities to balance workloads through the entire production process. To improve this, the amount of undesired asymmetry in the cross-training patterns should be limited by avoiding asymmetry on crucial operations, such as frequent bottlenecks. Furthermore, more situation-focused rules are needed, enabling opportunities to shift workers within a job and considering the fact that worker attendance can have different effects on a workstation’s capacity.

The remaining six factors mentioned in this paper led to a number of limitations for the adoption of an active allocation policy. An evaluation of the limiting circumstances resulted in a number of prerequisites a company should fulfil prior to introducing a successful allocation policy:

1) To prevent learning effects to become an unbridgeable issue and to maintain workers’ secondary skills on a useful level, supervisors and planning employees should arrange worker transfers for longer periods once in a while, by avoiding the frequent demand for a worker to move back to his primary workstation. This requires a sufficiently covered skill matrix with a pattern of chaining;

2) To maintain productivity levels high and decrease the influence of job preference, there should be a plan for when the workload of a workstation becomes too low. This could for example be a worker transfer or a training trajectory;

3) In order to create awareness of the opportunities for worker transfers, companies should adopt a controlled order release system, or implement a system that visualises the moment when workers finished their planned work and start on lower priority orders;

4) Especially when cross-training workers is costly or time consuming, a strategy incorporating chaining or more selective flexibility (Felan & Fry, 2001), increases a company’s opportunities to utilise flexibility;

5) When there is not always a surplus of machines, compared to the amount of workers, it is important to know precisely which specific tasks at a workstation (i.e. tool preparation) can be performed by an extra worker and which capacity gains are expected;

(30)

29 requirements and 4) worker-task priorities. The proposed design helps planning employees in the manual process of worker allocation on the shorter term, while automated planning algorithms can also exploit labour flexibility on a longer term.

The listed prerequisites will help companies to facilitate their organisation in such a way that opportunities for beneficial worker transfers are better detected, training decisions are more focused and trade-offs regarding pro-active worker transfers are easier to make.

7.2 Limitations and future research

The fact that this research has been performed by a single case study, brings some limitations regarding the generalisability. The case company was clearly in a pre-mature state of exploiting their cross-trained workforce, which is why the factors from the conceptual model were investigated from a constraining point of view, instead of the guiding role they can have. The fact that the case company worked with highly educated planning employees and could still not use the factors to facilitate the allocation process, makes it however questionable to which extent this facilitating role can be realised in practice. Furthermore the study is focused on a company with many complexities, such as extreme skill demands, asymmetry characteristics in the cross-training pattern, different labour- and machine constraining factors and order-dependent skill demands and capacities. This helped highlighting the difficulties to translate theoretical policies into practice, but also created a gap to such a degree that single theoretical propositions were hard to isolate. To improve this research, it would be useful to test the factors in a case with a more advanced level of cross-training, in order to research the guiding side of the factors. In addition, a case with only one type of asymmetry (i.e. asymmetric cross-training patterns) would gain more specific results.

(31)

30

R

EFERENCES

Bobrowski, P. M. & Park, P. S., 1993. An evaluation of labor assignment rules when workers are not perfectly interchangeable. Journal of Operations Management, 11(1), pp. 257-268.

Bokhorst, J. A. C., 2011. The impact of the amount of work in process on the use of cross-training. International Journal of Production Research, 49(11), pp. 3171-3190.

Bokhorst, J. A. C. & Gaalman, G. J. C., 2009. Cross-training workers in Dual Resource Constrained systems with heterogeneous processing times. International Journal of Production Research, 47(22), pp. 6333-6356.

Bokhorst, J. A. C., Slomp, J. & Gaalman, G. J. C., 2004. On the who-rule in Dual Resource Constrained (DRC) manufacturing systems. International Journal of Production Research, 42(23), pp. 5049-5074.

Bokhorst, J. A. C., Slomp, J. & Molleman, E., 2004. Development and evaluation of cross-training policies for manufacturing teams. IEE Transactions, 36(1), pp. 969-984.

Brusco, M. J., 2008. An exact algorithm for a workforce allocation problem with application to an analysis of cross-training policies. IIE Transactions, 40(1), pp. 495-508.

Brusco, M. J. & Johns, T. R., 1998. Staffing a multiskilled workforce with varying levels of productivity: An analysis of cross-training policies. Decision Sciences, 29(2), pp. 499-515.

Campbell, G. M., 1999. Cross-utilization of workers whose capabilities differ. Management Science, 45(5), pp. 722-732.

Das, A. & Narasimhan, R., 2001. Process-technology fit and its implications for manufacturing performance. Journal of Operations Management, 19(5), pp. 521-540.

Eisenhardt, K. M., 1989. Building theories from case study research. The Academy of Management Review, 14(4), pp. 532-550.

Felan, J. T. & Fry, T. D., 2001. Multi-level heterogeneous worker flexibility in a Dual Resource Constrained (DRC) job-shop. International Journal of Production Research, 39(14), pp. 3041-3059.

(32)

31 Francas, D., Löhndorf, N. & Minner, S., 2011. Machine and labor flexibility in manufacturing networks. International Journal of Production Economics, 131(1), pp. 165-174.

Fredendall, L. D. & Melnyk, S. A., 1998. Assessing the impact of reducingd emand variance through improved planning on the performance of a Dual Resource Constrained job shop. International Journal of Production Research, 33(6), pp. 1521-1534.

Hottenstein, M. P. & Bowman, S. A., 1998. Cross-training and worker flexibility: a review of DRC system research. The Journal of High Technology Management Research, 9(2), pp. 157-174.

Kher, H. V. & Malhotra, M. K., 1994. Acquiring and Operationalizing Worker Flexibility in Dual Resource Constrained Job Shops with Worker Transfer Delays and Learning Losses. Omega, 22(5), pp. 521-533.

Kim, S. & Nembhard, D. A., 2013. Rule mining for scheduling cross training with a heterogeneous workforce. International Journal of Production Research, 51(8), pp. 2281-2300.

Land, M., Stevenson, M. & Thürer, M., 2014. Integrating load-based order release and priority dispatching. International Journal of Production Research, 52(4), pp. 1059-1073.

Liu, T. & Li, M., 2012. Structural flexibility indices with shrinking capacities in cross production. International Journal of Production Research, 50(2), pp. 393-407.

Nelson, R. T., 1967. Labor and machine limited production systems. Management Science, 9(1), pp. 648-671.

Nembhard, D. A., 2001. Heuristic approach for assigning workers to tasks based on individual learning rates. International Journal of Production Research, 39(9), pp. 1955-1968.

Nembhard, D. A., 2007. Workforce cross training. 1st ed. Boca Raton: CRC Press.

Nembhard, D. A., 2014. Cross training efficiency and flexibility with process change. International Journal of Operations & Production Management, 34(11), pp. 1417-1439.

Nembhard, D. A. & Osothsilp, N., 2005. Learning and forgetting-based worker selection for tasks of varying complexity. Journal of the Operational Research Society, 56(5), pp. 576-587.

(33)

32 Oosterman, B. A., Land, M. J. & Gaalman, G. J. C., 2000. Influence of shop characteristics on workload control. International Journal of Production Economics, 68(1), pp. 107-126.

Philipoom, P. R. & Fry, T. D., 1999. Order review/release in the absence of adherence to formal scheduling policies. Journal of Operations Management, 17(3), pp. 327-342.

Sayin, S. & Karabati, S., 2007. Assigning cross-trained workers to departments: A two-stage optimization model to maximize utility and skill improvement. European Journal of Operational Research, 176(1), pp. 1643-1658.

Schultz, K. L., McClain, J. O. & Thomas L J, 2003. Overcoming the dark side of worker flexibility. Journal of Operations Management, 21(1), pp. 81-92.

Slomp, J. & Molleman, E., 2002. Cross-training policies and team performance. International Journal of Production Research, 40(5), pp. 1193-1219.

Stevenson, M., 2006. Refining a workload control (WLC) concept: A case study. International Journal of Production Research, 44(4), pp. 767-790.

Thürer, M., Stevenson, M. & Silva, C., 2011. Three decades of workload control research: a systematic review of the literature. International Journal of Production Research, 49(23), pp. 6905-6935.

Thürer, M., Stevenson, M., Silva, C. & Land, M., 2013. Towards an integrated workload control (WLC) concept: The performance of due date setting rules in job shops with contingent orders. International Journal of Production Research, 51(15), pp. 4502-4516.

Treleven, M., 1987. The timing of labor transfers in Dual Resource Constrained systems: 'push' vs. 'pull' rules. Decision Sciences, 18(1), pp. 73-88.

Voss, C., Tsikriktsis, N. & Frohlich, M., 2002. Case research in operations management. International Journal of Operations & Production Management, 22(2), pp. 195-219.

Xie, J. L. & Johns, G., 1995. Job scope and stress: can job scope be too high?. Academy of Management Journal, 38(5), pp. 1288-1309.

(34)

33 Yin, R. K., 2013. Case study research: Design and methods. 5th ed. London: SAGE Publications.

(35)

34

A

PPENDICES

Appendix I: Skill matrix

Wo rk er /ma ch in e DRG DRK DRK _ NC DRK P LT DRM MORIN L RS L RS LT HSL DRV FEL MAK INO S OD ICK VON DMU5 0 DMUBEM DMUP RO F DE F HU NT_ ON B NT1 0 0 0 NTP RO YAS YA S 7 0 YA S P RO M ALG M F AG M F TS M WE RTH M ZE IS S KON CBH CLRM ON M ON 1 1 1 2 1 1 2 2 2 1 1 2 1 2 2 3 1 2 1 2 2 4 1 1 1 1 2 5 2 1 1 2 2 6 1 1 2 2 7 2 1 1 1 2 2 8 2 1 1 2 2 9 2 1 2 1 2 2 10 2 1 1 1 2 11 1 1 2 2 12 1 2 2 1 1 2 2 13 1 1 1 2 2 1 2 2 14 1 1 1 2 2 2 2 1 1 1 1 2 2 15 1 1 1 2 2 16 1 1 1 1 2 2 17 2 2 2 2 2 1 2 2 18 2 2 2 1 1 1 1 1 1 1 2 2 19 2 2 2 2 1 1 1 1 2 2 20 2 2 1 1 2 1 2 2 21 1 2 2 1 1 1 2 2 22 1 2 2 1 1 2 1 2 2 23 2 1 1 1 2 2 24 2 1 1 1 2 2 25 2 1 1 1 2 2 26 1 1 1 1 2 2 27 1 1 1 2 2 28 1 2 1 1 29 1 2 2 2 1 30 1 2 2 1 1 31 2 1 2 2 1 32 1 2 2 1 33 1 2 2 1 34 1 2 2 1 1 35 1 1 1 2 2 1 36 1 1 1 2 37 1 1 1 2 1 2 38 1 1 1 2 1 2 39 1 1 1 2 40 1 1 1 1 2 41 1 2 1 1 2 42 1 1 1 2 1 2

ALLOCATION OF CROSS

A

-

-

-

:

M

’

Table of contents

A

1

I

2

T

3

M

4

C

5

C

6

D

7

D

R

A