European J. Industrial Engineering, Vol. 2, No.

1, 2008

17

A study on human-task-related performances in converting conveyor assembly line to cellular manufacturing Ikou Kaku*
Department of Management Science and Engineering Akita Prefectural University 0150055 Tutiya Ebinokuti 844 Yulihonjo City, Japan Fax: +81184272189 E-mail: ikou_kaku@akita-pu.ac.jp *Corresponding author

Yasuhiko Murase
Department of Management Science and Engineering Akita Prefectural University Yulihonjo City, Japan E-mail: ZWU00675@nifty.com

Yong Yin
Department of Economics and Business Management Yamagata University 1412, Kojirakawa-cho Yamagata-shi, 9908560, Japan Fax: (81)236284281 E-mail: yin@human.kj.yamagata-u.ac.jp
Abstract: In this paper, we propose a study to analyse the human-task-related performances in converting a Conveyor Assembly Line (CAL) to cellular manufacturing, which include the possible added operational tasks (which is considered a negative factor for the conversion), the skill level and the cross-training of workers. Three theoretical models (CAL, cellular manufacturing and a joint type, CAL+CM) are constructed involving those constraints respectively. A human-factor-based training approach is also represented for the system performance improvement in cellular manufacturing. Assuming the product mix and the skill level of workers are probability variables, simulation experiments based on the data collected from the previous documents are then used to estimate the marginal impact each factor change had on the estimated performance improvement resulting from the conversion. [Received 18 January 2007; Revised 13 July 2007; Accepted 26 July 2007] Keywords: converting conveyer assembly line to cellular manufacturing; labour efficiency; skill level; cross-training; simulation.

Copyright 2008 Inderscience Enterprises Ltd.

18

I. Kaku, Y. Murase and Y. Yin

Reference to this paper should be made as follows: Kaku, I., Murase, Y. and Yin, Y. (2008) A study on human-task-related performances in converting conveyor assembly line to cellular manufacturing, European J. Industrial Engineering, Vol. 2, No. 1, pp.1734. Biographical notes: Ikou Kaku is a Professor in Industrial Engineering and Management at the Department of Management Science and Engineering, Faculty of System Science and Technology, Akita Prefectural University, Japan. He received a PhD from the Department of Management Science and Engineering at Tokyo Institute of Technology in 1988. He has extensive experience in software design and optimisation methods of manufacturing systems. He worked for two Japanese venture companies ten years ago as technical CEO of Q&D programmes. His teaching and research interests include industrial engineering, human factors and operations management. His current researches include the cognitive process of multisensory information with event-related potential, meta-algorithms for general multilevel lot-sizing problems, and mathematical models and algorithms of cellular manufacturing. Yasuhiko Murase is a freelance Researcher in Japan. He received a PhD from the Department of Management Science and Engineering, Faculty of Engineering Division 1, Tokyo Metropolitan Institute of Technology in 2005. He worked for a Japanese industrial company for 35 years. He has extensive experience in project planning and control. His research interests include project management and operations management. His current researches include comparative studies and mathematical models of cellular manufacturing. Yong Yin is an Associate Professor in the Department of Economics and Business Management, Yamagata University, Japan. He received a PhD (DBA) from the Graduate School of Economics and Management at Tohoku University in 2002. His teaching and research interests include industrial engineering, operations management and management science. His current researches include comparative studies and mathematical models of cellular manufacturing.

Introduction

In recent years, after many Japanese companies had shifted their production organisations to China, those manufacturers left behind in Japan have been changing their production system remarkably. Instead of the conveyor mass production, several manufacturing methods have been developed for strengthening the competitive power of the domestic companies. For example, Tanaka (2005) reported that seven manufacturing methods have been used to deal with varying types and quantities of products in RICOH UNITECHNO Inc., a middle-scale Japanese company that assembles facsimile machine/copy machines/printers. Those methods are as follows: 1 2 One workerOne machine method (The product will be assembled by only one worker, who should do all of the assembly operations.) Two workersOne machine method (There is too much operation work to assemble a larger machine that it cannot be completed by one worker, in which case two workers should be assigned to do this assembly operation.)

A study on human-task-related performances 3 4 Cart-pulling method (Instead of a conveyor line, a cart is used as transport tool, which is pulled among several workers to complete the assembly operations.)

19

Relay method (the form of assembly line exists but the workers assigned to the line do not only those operations for themselves but also the operations not assigned for them, based on their operations ability.) Conveyor Assembly Line (CAL) (The traditional assembly method is also retained for those larger lot-sized products.) One-worker cell (Only one worker does all of the assembly operations of products, usually in a U-shaped layout. The difference with method (1) is that the worker in the cell can do all of the assembly operations of several products, which means he has a higher operations ability.) Division cell (Several workers are assigned to one cell, who may do the assembly operations using methods (3), (4) or (6).)

5 6

All of these methods were based on the philosophy of the Toyota production system, that is, only the products that suit the needs of customers (the kinds of products are changing dynamically) should be manufactured flexibly when they are needed (the production quantities are also variable). These methods and their combinations were successfully used to correspond flexibly to different types (over 400 types of products) and quantities (70% of products are under 100 units/month) of products. The technical innovations described above are not exceptional examples in Japan. Tsuru (1997; 1998) reported that about 25.7% of Japanese enterprises had already adopted the innovations in 1997 and the rate would increase to 53.5% after five years, to overcome the defect in traditional conveyor manufacturing and search for more effective production methods. The most popular among the innovations is Cellular Manufacturing (CM) in Japan, in which one (or multiple) worker(s) carries out all of the operations of an assembly job, usually in a U-shaped layout. CM refers to all of the methods described above except method (5). Since it seems to be able to improve system performance in a changing environment, many Japanese companies have introduced CM into their factories to convert existent CAL. A tremendous achievement of such a conversion was realised at Canon, a famous Japanese electronics company. Takahashi et al. (2003) reported that by introducing CM into Canons factories, by 1995 over 20 000 m of conveyor had been withdrawn and 720 000 square metres of working space from 54 related factories had been emptied. As a result of the innovation, the overall cost rate has decreased from 62% to 50% during the past eight years. Since then converting existent CAL to CM has come into fashion in Japan. A CAL is often converted to several one-worker CMs, in which all of the assembly tasks in CAL will be completed by a worker, according to the workers assigned in the CAL. This is the general case of the conversion in Japanese industries. Johnson (2005) pointed out that depending on the job design, each worker assembles either a portion or all of the subassembly or product produced in the CM. They may also be responsible for dynamically balancing the flow of work as the product mix or demand levels change. It means that the loss of simplification and specialisation caused by the increased number of assembly operations performed by each worker in CM may increase the time required for each operation, which serves to hinder the performance improvement caused by the conversion. Miyake (2006) pointed out that the benefit of converting CAL to CM should be pursued by increasing the production flexibilities and the labour efficiency. The latter

20

I. Kaku, Y. Murase and Y. Yin

relates to cross-training of workers and is a key issue in the conversion. Generally, how to raise labour efficiency is an important decision problem in the current manufacturing environment, not only in job shops but also in CAL. Cellular manufacturing is a considerable choice in both situations but cross-training of workers is becoming a necessary condition, because a worker in cellular manufacturing could do more tasks than before. Several researches have treated the topic of labour efficiency and cross-training over the years. Eckstein and Rohleder (1998) compared the traditional job shop layout and CM when considering human resource issues (number of workers, learning) and claimed that CM outperforms the job shop under the conditions of uneven demand mix, fewer workers and fast learning. Iyer and Askin (1999) discussed the cross-training level in a manufacturing cell and claimed that the complete cross-training (all workers are able to operate all of the machine layouts in a cell) must be managed effectively, keeping in mind the various interactions between operating policies (Work-In-Process, Batches and so on). On the other hand, Slomp et al. (2005) pointed out that balancing workload (shifting work from a heavily loaded worker to a less loaded worker) in a cell can improve the operational performance, which is supported by cross-training decisions. It should be pointed out that the insights based on these researches are helpful when we consider converting from CAL to CM; however, we believe the cross-training in the situation of converting CAL to CM is different from the situation of converting job shop to CM. The number of machines that can be operated by one worker is an important evaluation factor in the situation of converting job shop to CM, while in many cases of converting CAL to CM in Japanese electronic companies, complete cross-training is well done for workers but the skill level of each worker is still different because they only use some simple hand tools, not expensive machines. The purpose of this paper is to indicate the limit of CM utilisation and the precept of the conversion. Two efficiency factors are considered to influence the human-tasks-related performance in the process of converting CAL to CM. The first efficiency factor is the possible added operational tasks, which is a negative factor for the conversion because it may increase the operating time. Consider the case of one worker cell in which not only all of the assembly tasks in CAL will be done, but several added tasks (not existing in CAL) are also essential to support the operating tasks; for example, pulling the assembly cart; searching for a part to assemble; selecting or changing a tool; and confirming something from the manual. Such added tasks waste operating time and decrease system performance, and this is the reason that some companies do not gain performance improvements from converting their assembly lines to cells. However, the negative effects of the added tasks can be reduced by raising workers skill levels through cross-training of workers, which is the second efficiency factor considered in this paper. Instead of discussing how to raise the workers skill level (it is not the main theme of this paper), we discuss the relationships of the efficiency factors in the conversion process and show that the system performance can be improved by using an effective training support system. In this paper, three theoretical models (CAL, CM and a joint type, CAL+CM) are constructed involving those efficiency factors, and a human-factor-based training approach is also represented for the system performance improvement. Assuming the product mix and the skill level of workers are probability variables, simulation experiments based on the data collected from the previous documents are then used to estimate the marginal impact each factor change had on the estimated performance improvement resulting from the conversion.

A study on human-task-related performances

21

The remainder of this paper is organised in the following way: In the next section we give a brief overview of CM development in Japan. Section 3 presents the theoretical models of converting CAL to CM. Simulation experiments are presented in the fourth section, and the results and analysis are given in the fifth section. Concluding remarks are given in the final section.

CM development in Japan

Nowadays, not only Canon, but also most Japanese big companies as Sony, Panasonic, NEC, Fujitsu, Hitachi and many other manufacturers have adopted CM; and most of them have obtained great benefits (Yamada and Kataoka, 2001; Isa and Tsuru, 2002; Iwamuro, 2002; Noguchi, 2003; Kimura and Yoshita, 2004; Kono, 2004; Sony, 2005; Takeuchi, 2006). There are a lot of Japanese documents and academic articles on converting CAL to CM. Early documents were reported by Tsuru (1997; 1998), whose research was based on a large-scale (over 1000) investigation of Japanese companies (in 1997) and a questionnaire of 13 factories and one consultant company (in 1998). The main standpoint of the documents claimed that CM can be recognised as a form of knowledge of the Toyota Production System that has been historically transferred to other industries. It is basically correct because the philosophy of the Toyota Production System is built on how to improve the efficiency of the traditional CAL. As a result, a marked rising tendency of Japanese companies to adopt the CM into their factories has been observed. Basically, the concept of CM is not new. Over the previous decade, a series of research articles have investigated the cellular manufacturing systems and compared the performance improvements with traditional functional layout. Wemmerlv and Hyer (1989), Wemmerlv and Johnson (1997) and Johnson (2005) claimed that a cellular manufacturing system represents a major technological innovation to many manufacturing systems traditionally based on functional and assembly specialisation. In contrast with the abundant literature on converting function layout to cell, few English papers discuss the issues on converting CAL to CM. Sakazume (2005) reported a survey of Japanese literature that included a total of 107 documents (12 academic papers, 18 technical reports and 77 newspaper articles). He tried to explain that the so-called US CM (from job shop to cell) is completely different from Japanese CM (from assembly line to cell), in terms of the implementation changes and mechanisms behind the advantages and disadvantages, even though there are some similarities in cell features and advantages and disadvantages of implementation. Yin et al. (2006) pointed out the economic background of converting CAL to CM in Japan based on a survey of the most recent Japanese literature. Summarily, Japanese manufacturing firms were faced with decreased market demands and increased product variations. To survive in such an extremely tough business environment, the traditional high-volume CALs were no longer adequate. Speedy adjustments were needed to handle transitions in product models and demands. A companys competitiveness was becoming dependent on whether or not it could respond to these transitions. In such an environment, there was a trend in Japanese industries towards converting CALs to more flexible CM. However, converting CAL to CM is not easy because a large number of operation factors should be treated carefully to adapt to

22

I. Kaku, Y. Murase and Y. Yin

their production environment. In fact, several advantages of converting CAL to CM have been reported, but at the same time, almost the same number of disadvantages in the conversion have also been reported (Tsuru, 1998; Isa and Tsuru, 2002; Sakazume, 2005). Distinct from these comparative researches, several researchers focus on what factors determine the performance improvement during such conversions and how these factors operate in favour of system behaviours. Johnson (2005) first proposed a simulation study to the above problem. He investigated three models (named linked line, buffered line and two-person CM) by using multievaluating criteria (set-up time, moving time and in-process waiting time) to estimate the marginal impact each factor change had on the estimated performance improvement resulting from the conversion. He concluded that the two-person CM has the ability to generate significant performance improvement over the assembly lines. Kaku et al. (2006) proposed a comparative study of converting CAL to CM, in which the constructing approach based on human factors is first introduced to adapt to the ability limit of workers. As a result, the performance improvement resulting from the conversion is dependent on the degree to which factors can improve performance and the degree to which their impact can overcome any operation-task-time increases caused by the loss of worker specialisation. Moreover, cross-training of workers is a key issue in the conversion and effective training methods should be considered to improve the system performance.

Theoretical analysis of converting CAL to CM

We consider a production problem as follows: There exists a traditional belt conveyor line with multiple assembly stations. Workers are assigned at each station according to a traditional job design method but they have the ability to do more tasks than that assigned to them. We assume that the workers abilities differ with the stations and products. Multiple products will be manufactured in the conveyor line. Each product has different batch sizes but with a known distribution of demand. When the product is assembled in the conveyor line, the stations and workers who will be used to complete the assembly operations are active. Because workers have different abilities to do those operations (which belong to stations and products), when the batch will be finished depends on the worker who has the slowest speed to do jobs. Thus, workers who have higher abilities should be idle, which may lead to decreasing the motivation of workers. On the other hand, because all of the products should be manufactured on the same conveyor line with a fixed order, some time may be spent on waiting to manufacture so that we cannot respond flexibly to the customers variant demand. Several KAIZEN methods can be considered to improve the system performance of such conveyor lines. For example, we may assume workers will do all of the operations that they can do, even those not assigned to them, since workers who have higher abilities should help other workers in the conveyor line; or we may convert the conveyor line to CM; or we may convert part of the line to cells for workers who have higher abilities and retain part of the conveyor line for workers who have lower abilities. In this paper, we consider three models: pure CM, pure CAL and a joint type of CM+CAL to deal with the above situations. The models are constructed by using the well-known theorem of the random sum of the random variables, which was first proposed by Moodie and Young (1965) for solving the stochastic inventory problem and later developed by Murase et al. (2006) for solving the dynamic line balancing problem.

A study on human-task-related performances

23

We assume a necessary cycle time Sk of station k with a normal distribution, i.e., it consists of an average working time Ek and a standard deviation Vk of a work station (Vk is the variance of Ek):
Sk = Ek + Vk

(1)

where safety factor guarantees a specified confidence limit of station k. For example, suppose equals 1.64; then 95% of the work can be completed in the cycle time. Moreover, such considerations can be exploded to the case of multistations. In that case, the cycle time:
Sk =
iBk

iBk

(2)

is the sum of those stations belong to station k (Bk is a set of stations which belong to station k). Using the cycle time Sk of station k, the system performance can be evaluated by the output of products that can be calculated as below with various product mixes:
Output of product = Shift Time Sk

(3)

where Shift Time is a standard working date.

3.1 Modelling CAL

The cycle time defined above is based on the consideration of standard working time, within which the person performs such work under the standard working conditions by ordinary efforts. Even if the cycle time of workstations is treated as a probability variable, the basic problem is not changed; the results of optimal line balancing are still obtained when all the cycle times are equal. However, the manufacturing environment is changed tremendously, not only in the metabolism of product diversification and the shortened life cycle, but also in the experiences of workers. Therefore we have to take into consideration a probability variable to represent the workers skill level. It is related with the cycle time calculation model incorporating the level of skill of the worker group, which is represented as a random variable and can be represented as the rate of real working time divided by the standard working time. It should be noted that the workers skill level is dependent not only on stations but also on product types. Therefore, two probability variables should be considered in the CAL model. One is the cycle time belonging to stations, the other is the level of skill of the worker group. We use the same approach proposed by Murase et al. (2006) for dealing with this two-probability-variables problem. Summarily, if there are several workers in an assembly line, the average level of skill in the worker group is given by a probability variable with the normal distribution (, 2). is the average level of skill in the worker group and is its standard deviation. When the level of skill is considered in assembly line balancing, the cycle time is reformed as follows:
Sk = Ek + Ek2 2 + Vk .

(4)

24

I. Kaku, Y. Murase and Y. Yin

When the level of skill is not taken into consideration, set = 1 and = 0, then Equation (4) will become the same as Equation (1). Otherwise the cycle time is represented by the co-effect of both workstations and the level of skill of workers. For simple examination in this paper, we assume each workstation includes only one work element and is performed by one worker. Because the workers level of skill is different, the performance of the whole CAL system is determined by the capacity of the worker who has the lowest level of skill. That means the output of products is determined by the longest cycle time in CAL.
Output of products = Shift Time . max{Sk }

(5)

3.2 Modelling CM
Compared to CAL, in which tasks are divided into several stations and workers are fixed at each station, in CM worker(s) should do more tasks and have to move among the assigned stations. Several documents describe that it is one of the contributions of CM because it improves the productivity by reducing the number of workers. However, two practical problems must be considered when converting CAL to CM. First, cross-trained workers are becoming necessary to do multiple tasks assigned to different stations. Complete cross-training (the example is one-worker CM) is desirable but costly. Until now most of the cross-training is limited to learning the tasks of the previous and following stations, so that reconfiguration of the cell may be required to assemble different products, accommodate changes in demand volumes, product types, etc. Second, converting CAL to CM means several work elements should be added naturally. For example, the worker may use a manual cart to move parts between assembly stations; search for and gather parts for assembly; and select and change the tools. All of these added tasks are completed by manual or simple mechanical help. Such added tasks can lead the CM system performance to become worse. Therefore, increasing the level of skill of workers through cross-training to overcome the difficulties brought from the added tasks is a key issue in CM systems. For clarifying the dynamic behaviours of these factors, an empirical approach, which had been discussed by Murase et al. (2006), is used to construct the CM model. We use a criterion named degree of difficulty to represent the effect of cross-training of workers. Suppose the number of work elements (stations) which should be completed by one worker is an important factor to improve the performance of the CM model; the degree of difficulty presents how many elements (stations) should be contained into a cell. It seems to be limited by workers ability to memorise and learn to complete certain tasks. Because it relates to human information processing of workers, how to strengthen the human memory also becomes an important operational factor in the CM model. For a special example, Kimura and Yoshita (2004) reported that a worker of Canon who has the highest level of skill can assemble a product which is composed of 2700 parts in only two hours. However, in common cases the number of elements completed by one worker is smaller. In this paper we use 60 as an upper limit of elements, which had been suggested by several Japanese documents (for example, see Murase et al., 2006). On the other hand, 59 work elements are recommended to be assigned to one station because a worker can memorise these elements well according to the well-known Millers (1956) magic number. Therefore a worker could do tasks with at least five work elements in the case of

A study on human-task-related performances

25

CAL (one station) and more than five work elements in the case of CM. As a result, the degree of difficulty can be represented by the ratio of the number of elements in CM to that of CAL. Obviously, the type of assembly work is also influencing the degree of difficulty. Moreover, the degree of difficulty may be improved by a work support system in which different supports may be supplied by different communication methods; the better the communication method, the lower the degree of difficulty. According to the discussion above, we construct the CM model. Firstly, Hyuga (1985) suggested that a considerable work support system can be carried out as an application of the results of a US Air Force experiment, which investigated the amount of information in human memory after information processing with different communication methods. The conclusion was that, at four levels of communication (Level 1: only language; Level 2: only chart and table; Level 3: language and chart and table; Level 4: more visual information added than in Level 3), humans retain different amounts of information. Because the experimental results were not shown in a formula but in a graph, we apply a regression method to those results and represent such relationships by the following formula:

= 0.4473 e0.81Z
where: Z = 1,2,3,4 = represents the different communication levels; the communication levels can be explained as contents applied at the place of tools and parts as a signboard, a sticker, and highlighting, etc. = represents the workers ability to memorise certain tasks. It can be considered to be in inverse proportion to the degree of difficulty.

(6)

Parameters 0.447 and 0.81 are most appropriate to fit the formula to the graph in our trials (coefficient of determination R2 = 0.97). Then, the degree of difficulty can be defined by the following formula according to the discussion above:

=
where:

N op 1 N op 1 , 5 N op 60, 1 Z 4 = 5 5 0.447 e0.81 Z

(7)

= degree of difficulty Nop = total number of workstations, Nop/5 represents the least number of workstations, Nop has an upper limit of 60 and a lower limit of 5 = a parameter showing the hardness of the total process in consideration; if the process is a complex one, then set > 1, otherwise set < 1.
Finally, we define the cycle time in the CM model. The cycle time can be divided into three parts in this case: The first part is the treatment of the assembly tasks which were assigned to several stations and were completed by several workers in the CAL system. The second part is the treatment of the added tasks which were carried out by doing more tasks in CM. The third part is the treatment of the influence of the degree of difficulty. It should be considered that the workers performance in CM is influenced not only by the factors coming from more assembly tasks and added tasks, but also by the factors coming

26

I. Kaku, Y. Murase and Y. Yin

from the degree of difficulty and the workers skill. In this paper we simply treat the third part as a probability variable Ht and it represents the influence of the factors except the added tasks. Then the cycle time of CM model is represented as follows:
G G G K K K SCM = Sk + Fg + Sk2 + Fg2 2 + Vk + Vg + Ht g =1 g =1 g =1 k =1 k =1 k =1

(8)

where: SCM = cycle time when all of the tasks are done in CM Sk = standard time to do task k, k = 1,2,,K (number of stations) Fg = time to do the added task g, g = 1,2,,G (number of added tasks) Vk = variance of doing task k, k = 1,2,,K (number of stations) Vg = variance of doing the added task g, g = 1,2,,G (number of added tasks) = average level of skill = standard deviation of the level of skill Ht = probability variable representing the influence of the factors except the added tasks, Ht ~ N ( , (0.15 )2). Therefore the output of products by one worker in CM is as follows:
Output of products = Shift Time . SCM

(9)

Note that the output of products of Equation (9) just shows one workers output; the total output of products is the sum of the output of the workers who carry out the production mix. Furthermore, if the CAL is converted into multi-person cells, i.e., each cell holds two or more workers, the cycle time of each cell can be calculated by using the formula of SCM just to divide K and G into different workers, then by using the average cycle time of workers in the cell to calculate the output of products.

3.3 Joint model of CAL + CM

Consider the situation in which CAL has many stations that cannot be converted to one-worker CM. We can design a joint model of CAL+CM as follows. First divide part of the stations of CAL to make one-worker CM (here we consider only one-worker CM), the worker who has the highest level of skill is usually assigned to the CM. The cycle time of the rest stations of CAL is calculated by using the largest Sk from Equation (4). Suppose several stations of CAL are converted to CM; then CM itself has the cycle time from Equation (8). Suppose CM is set before CAL, the output of products by one worker in CM is as follows:
Output of products = Shift Time . SCM + Max{Sk }

(10)

If there are multiple CMs in the joint model, the problem will become very complex because which product should be preassembled in the cells and put to the remaining CAL is a complex scheduling problem. We will discuss this problem in another paper.

A study on human-task-related performances

27

Simulation experiments

Simulation models are designed in this paper to evaluate the performance of different models and to estimate the marginal impact of each factor change discussed on the performance improvement. The CAL model with different stations (20, 30, 40) is investigated. Stations are linked by one worker per station. Once the assembly task at a station has been completed, the unit could be transferred immediately to the next station. There are no inventory buffers between stations. The production plan is based on a fixed product mix, shown in Table 1. There are 11 types of products (denoted by L-M-S, L-M-H, and so on) and a total quantity of 8000 units per month should be completed. The demand rate of each product, average cycle time and necessary number of set-ups are also shown in Table 1.
Table 1 Parameters of product mix Number of necessary set-ups Product L-M-S L-M-H L-L-S L-L-H L-M-B L-P-H L-P-S L-L-B L-P-B X-M-S X-P-S Average cycle time 0.8 0.96 0.96 1.04 0.64 0.96 0.8 0.72 0.64 0.96 0.8 Rate of demand 0.398 0.127 0.092 0.075 0.065 0.057 0.052 0.048 0.044 0.021 0.020 One time per day 1 1 1 1 1 1 1 1 1 1 1 Six times per day 1 2 2 2 2 2 2 2 2 2 2 Four times per day 1 2 2 2 2 2 2 2 2 10 10 Three times per day 1 5 5 5 5 5 5 5 5 5 5

The batch size of each product can be calculated by using the demand rate multiplied by units quantity divided by operating date, and adjusted by a normal variable (1, SD 2). SD is the standard deviation and ranges from (0.2, 0.25, 0.3). The average cycle time of each product is also adjusted by a normal variable (1, SD 2), but the SD ranges from (0.1, 0.15, 0.2). Set-up is assumed 1 time for CM but 3, 4, 6 times for CAL. Each set-up will take a time period of (10 min, 30 min, 50 min). Workers have different skill levels, which depend on a normal distribution of (1, 0.152). For simple examination, we randomised and fixed the average level of skill to all workers and used the top numbers of workers to correspond to different stations of CAL. The workers skill levels are shown in Table 2. CM with a well-known U-type layout is investigated. Three CM models, in which all of the tasks (stations) are completed by one worker or divided and completed by two workers and three workers, are considered as shown in Figure 1. A portable-type parts shelf is arranged in the centre of the U-type layout. Parts are packed and moved by a working table cart. These operations of selecting and moving parts, selecting tools, reading posters or materials and so on are not needed in CAL but necessary in the CM model. These operations are the added tasks of CM. It can be considered that the

28

I. Kaku, Y. Murase and Y. Yin

influence of the added tasks on the cycle time of each product can be divided into two parts: one is a fixed time denoting the time absolutely needed for such tasks, and adjusted by a normal variable (1, SD 2), but SD ranges from (0.1, 0.2, 0.3); the other one is a probability variable (Ht), which represents the degree of difficulty in the CM models.
Table 2 No. of worker 1 2 3 4 5 6 7 8 9 10 Figure 1 Level of skill per worker

0.991 1.082 1.111 1.015 1.176 1.109 1.184 0.885 0.847 0.956

No. of worker 11 12 13 14 15 16 17 18 19 20

0.919 1.022 0.947 0.870 0.783 1.133 0.801 1.269 1.157 0.711

No. of worker 21 22 23 24 25 26 27 28 29 30

1.058 0.919 1.099 1.036 0.875 1.348 0.853 0.883 0.785 1.117

No. of worker 31 32 33 34 35 36 37 38 39 40

1.229 0.929 1.282 0.877 0.891 0.847 0.687 0.680 1.001 1.069

The three types of CM models

1 worker CM

2 workers CM 3 workers CM

For the joint model of CAL+CM, we study two cases based on a 30-station CAL. One case is a ten-station CAL+1 worker CM, the other is two six-station CAL+1 worker CM. The parameters of the joint model are the same as those used in CAL and CM. All simulation models were developed by using the simulation software package of WITNESS of Lanner Group Ltd.

Analysis and discussion of the simulation results

Several simulation results are obtained from our experiments. Here we show some main observations from the standpoint of the superiority of CM and the consideration of efficiency factors. Firstly, the outputs of product L-M-H per worker with different models are represented in Figure 2 under the same condition of normal cycle time (0.96, 0.152). The influences of different cycle times are shown in Figure 3, but only one-worker CM is used to compare the system performance.

A study on human-task-related performances

Figure 2 The performances with different systems

29

350

Output of products

300 250 200 150 100 50 0 20 30 Number of tasks 40

1 worker CM 2 workers CM 3 workers CM CAL

Figure 3

System performance with different cycle times

350
Output of products

300 250 200 150 100 50 0 20 30 Number of tasks 40

1 worker CM (SD = 0.1) 1 worker CM (SD = 0.15) 1 worker CM (SD = 0.2)

CAL (SD = 0.1)
CAL (SD = 0.15) CAL (SD = 0.2)

It is clear that there is a significant difference between CAL and CM. In all cases (20, 30 or 40 tasks), CM over-performs CAL. We can consider that such an advantage is from the fact that CAL is limited to the probable cycle time done by the worker who has the worst level of skill, but this disadvantage can be avoided by CM. From Figure 2 and Figure 3, it can be observed that the system performance of CM is 1.5 or 2 times of that of CAL. Moreover, the influence of changing cycle time in CM can be ignored but that in CAL is significant. That means that CM has more production flexibility than CAL, which has been pointed out as an advantage in several Japanese documents. Since a worker should do not only more tasks but also the added tasks created in CM, the system performance of CM would be decreased and would sometimes go below that of CAL when these negative factors became larger. Figure 4 represents the relationship among system performance, increasing added tasks and the rate of cross-training. As shown in Figure 4, increasing added tasks is represented by a percentage (from 0% to 30%), which means that the added tasks will take up part of the necessary cycle time.

30

I. Kaku, Y. Murase and Y. Yin

Cross-training is represented by a rate from 0.5 to 1.0. The rate of cross-training equal to 1.0 means a complete cross-training has been used. It can be observed from Figure 4 that when the added tasks are increasing, the system performance becomes worse at the same level of cross-training; but it can be improved by increasing the rate of cross-training. Moreover, if not enough cross-training has been provided (i.e., the rate of cross-training is lower), the system performance of CM may become worse than that of CAL.
Figure 4 System performance with increasing added tasks and the level of cross-training

250
Output of products

200 150 100 50 0 0.5 0.6 0.7 0.8 0.9 1.0 The rate of cross-training

1 worker CM (0%) 1 worker CM (10%) 1 worker CM (20%) 1 worker CM (30%) CAL (30 stations)

Again, Figure 5 represents a trend of longer operating time caused by such added tasks, which could decrease the system performance, under different process conditions. From Figure 4 and Figure 5 it can be observed that the CM structurally increases added tasks and is limited to a kind of break-even point at which the system performance has been reversed. However, the break-even point is dependent on the real situation of the manufacturing system and changes with various factors (not only the efficiency factors) which can influence system performance. We infer that converting CAL to CM can be achieved under continuous technical innovation of cross-training, Kaizen, etc.
Figure 5 System performance with increasing operation times

160 140 120 100 80 60 40 20 0 0 3 2 1 4 Average operating time

Output of products

1 worker CM ( = 0.75) 1 worker CM ( = 1.0) 1 worker CM ( = 1.25)

CAL (30 stations)

A study on human-task-related performances

31

Therefore, how to raise the workers level of skill through cross-training is very important in the conversion because the cross-trained workers guarantee the achievement of CM. In this paper, we discussed just one way to improve the workers level of skill, i.e., making a good communication environment to support such training. Figure 6 shows the result that increasing communication levels can improve the system performance.
Figure 6 System performance with different communication levels
200

Output of products

150 100 50 0 1

1 worker CM ( = 0.8) 1 worker CM ( = 1.0) 1 worker CM ( = 1.2) CAL (30 stations)

2
3 4

The communication level

In order to apply the influence of the outside environment to the conversion, two factors are considered for the various product mix and caused set-up times. Firstly, the operation times in various manufacturing systems are compared in Figure 7 under the conditions of product mix according to Table 2. Clearly, the longer the period of set-up the more waste in operations. Therefore the total operation time of manufacturing the same products becomes longer. However, this negative influence could be absorbed through dividing different products to different cells in CM. Moreover, it also can be observed from Figure 7 that the joint models of ten stations CAL+CM and six stations CAL+CM show the best results of system performance. It seems that when the inside and outside manufacturing environments become complex, the proper manufacturing system is not a simple form (as with converting CAL to CM) but a compound form. It has been considered and demonstrated in several companies (an example is Tanaka, 2005) but should be discussed particularly in the future.
Figure 7 Operation times with different set-ups
350 300 3 times setup per day (CAL) 4 times setup per day (CAL) 6 times setup per day (CAL) 1 times setup per day (CM) CAL (10 stations) + CM 10 20 50 Time period of set-up CAL (6 stations) + CM

Operation times

250 200 150 100 50 0

32

I. Kaku, Y. Murase and Y. Yin

Secondly, the inventory rate of products (average inventory amounts divided by output of products) and shortage rate are also compared with the CM and CAL system above. As shown in Table 3, for example, the shortage rate of CM (0.62%) is worse than that of CAL (0% for three times set-up per day) but the inventory rate of products of CM (0.19%) is significantly better than that of CAL (0.54%).
Table 3 Inventory rate and shortage rate with different production systems CM CM+CAL 0.19 0.62 CAL: six times set-up per day 0.288 0.28 CAL: four times set-up per day 0.4 0.19 CAL: three times set-up per day 0.54 0

Production method Inventory rate (month) Shortage rate (%)

Based on the observations above, summarily converting CAL to CM can be considered to produce an effective manufacturing system when the cross-training of workers is achieved and the level of skill of workers attains a considerable class of the operation; but such a conversion should be completed carefully and step by step.

Conclusion

Converting CAL to CM in Japanese industries is a new kind of technical innovation against the traditional conveyor manufacturing system. The largest contribution of the conversion is increasing not only the motivation but also the level of skill of workers to improve system performance in dynamic production environments. However, such a conversion is not easily controlled because the workers in CM need to have higher skill levels and the achievement of the conversion is limited by various inside and outside conditions. In this paper, the human-task-related performances in converting CAL to CM have been investigated standing from this viewpoint. The contributions of this paper are as follows: First, we built the theoretical models to construct the influences of probable added tasks and provided an improvement policy based on human factors, in which cross-training of workers is represented by the level of skill times the degree of difficulty, and proposed that it is effective to strengthen the cross-training through some communication technologies. Second, we also discussed the effectiveness of converting CAL to CM when these efficiency factors are considered using simulation experiments. The insights obtained from our models and simulation experiments can clarify the relationships among these operating factors and give a fundamental indicator of manufacturing system selection. We conclude that a manufacturing system containing a compound form will be appropriate in the current dynamic production environment. Future work is to build a mathematical model involving the compound form manufacturing system and to analyse the behaviour of the system.

A study on human-task-related performances

33

References
Eckstein, A.L. and Rohleder, T.R. (1998) Incorporating human resources in group technology/cellular manufacturing, International Journal of Production Research, Vol. 36, No. 5, pp.11991222. Hyuga, S. (1985) Technology of Presentation (in Japanese), GOMA Press. Isa, K. and Tsuru, T. (2002) Cell production and workplace innovation in Japan: toward a new model for Japanese manufacturing?, Industrial Relations, Vol. 41, No. 4, pp.548578. Iwamuro, H. (2002) Cellular Manufacturing System (in Japanese), Tokyo: Nikkan Kogyo Shinbun. Iyer, A. and Askin, R.G. (1999) Operating manufacturing cells with labor constraints, Handbook of Cellular Manufacturing Systems, John Wiley & Sons, pp.195223. Johnson, D.J. (2005) Converting assembly lines to assembly cells at sheet metal products: insights on performance improvements, International Journal of Production Research, Vol. 43, No. 7, pp.14831509. Kaku, I., Murase, Y. and Yin, Y. (2006) Theoretical analysis of converting assembly line to cellular manufacturing: how to select your manufacturing system?, The Proceedings of LSCM2006, Hong Kong, January. Kimura, T. and Yoshita, M. (2004) Cellular manufacturing runs into trouble when nothing is done (in Japanese), Nikkei Monozukuri, Vol. 7, pp.3861. Kono, H. (2004) The aim of the special issue on Seru manufacturing (in Japanese), IE Review, Vol. 45, pp.45. Miller, G.A. (1956) The magical number seven and plus or minus two: some limits on our capacity for processing information, The Psychological Review, Vol. 63, pp.8197. Miyake, D.I. (2006) The shift from belt conveyor line to work-cell based assembly system to cope with increasing demand variation and fluctuation in the Japanese electronics industries, Report Paper of CIRJE-F-397. Moodie, C.L. and Young, H.H. (1965) A heuristic method of assembly line balancing for assumptions of constant or variable work element time, The Journal of Industrial Engineering, Vol. XVI, No. 1, pp.2329. Murase, Y., Masuda, S. and Fukuda, S. (2006) The assembly line balancing by deployment of inventory theory (in Japanese), Journal of Japan Industrial Management Association, Vol. 57, No. 1, pp.19. Noguchi, H. (2003) Production Involution in Japan (in Japanese), Tokyo: Nikkan Kogyo Shinbun, LTD. Sakazume, Y. (2005) Is Japanese cell manufacturing a new system?: a comparative study between Japanese cell manufacturing and cellular manufacturing, Journal of Japan Industrial Management Association, Vol. 12, pp.8994. Slomp, J., Bokhorst, J.A.C. and Molleman, E. (2005) Cross-training in a cellular manufacturing environment, Computers & Industrial Engineering, Vol. 48, pp.609624. Sony (2005) The manufacturing and production of Sony (in Japanese), Unpublished report. Takahashi, S., Tamiya, H. and Tahoku, H. (2003) A miracle Canon production system: it is not only Toyota (in Japanese), Special Report of Weekly Toyo Economics, 27 November 3 January, pp.123136. Takeuchi, N. (2006) Manufacturing Methods Illustrated: Cell Production System (in Japanese), Tokyo: JMAM. Tanaka, T. (2005) Birth of a cart pulling style production system (in Japanese), IE Review, Vol. 46, No. 4, pp.3137. Tsuru, T. (1997) Innovation of production system and human utility in the 90s (in Japanese), Report of Economic Research Institute, Japan Society for the Promotion of Machine Industry, H8-8, Japan, May.

34

I. Kaku, Y. Murase and Y. Yin

Tsuru, T. (1998) Cell manufacturing and innovation of production system (in Japanese), Report of Economic Research Institute, Japan Society for the Promotion of Machine Industry, H9-9, Japan, September. Wemmerlv, U. and Johnson, D.J. (1997) Cellular manufacturing at 46 user plants: implementation experiences and performance improvements, International Journal of Production Research, Vol. 35, No. 1, pp.2949. Wemmerlv, V.U. and Hyer, N.L. (1989) Cellular manufacturing in the US industry; a survey of users, International Journal of Production Research, Vol. 27, No. 9, pp.15111530. Yamada, H. and Kataoka, T. (2001) Unusual production revolution (in Japanese), NHK. Yin, Y., Kaku, I., Murase, Y. and Yasuda, K. (2006) Convert flow shops to manufacturing cells: lessons from Japan, The Proceedings of LSCM2006, Hong Kong, January.