Planning for contiual jit

Planning for contiual jit - IJOPM 13,6 4 Received February...

This preview shows page 1 - 19 out of 19 pages.

Image of page 1
Image of page 2
Image of page 3
Image of page 4
Image of page 5
Image of page 6
Image of page 7
Image of page 8
Image of page 9
Image of page 10
Image of page 11
Image of page 12
Image of page 13
Image of page 14
Image of page 15
Image of page 16
Image of page 17
Image of page 18
Image of page 19

You've reached the end of your free preview.

Want to read all 19 pages?

Unformatted text preview: IJOPM 13,6 4 Received February 1991 Revised October 1991 International Journal of Operations 8: Production Management. Vol. 13 No. 6. 1993. pp 422. o MCB Universxty Press. 0144-3571 © Emerald Backfiles 2007 Planning for Continual Improvement in a Just-in-Time Environment Scott R. Swenseth University of Nebraska-Lincoln, Lincoln, USA Krishnamurty Muralidhar Florida International University, Miami, Florida, USA, and Rick L. Wilson Oklahoma State (adversity, Oklahoma City, Okla/201220, USA Introduction A major component of the Just-in-Time UIT) production philosophy is the continual improvement which results from employee meetings or other programmes, like quality circles. It can be expected that processing time variation, like all other aspects of pull production processes, will improve (reduce) continually over time and, further, it is reasonable that the initial improvements will be somewhat drastic and subsequently will be made at a marginally decreasing rate over time. The relationship between processing time variation and the performance of pull production systems has been extensively addressed in the literature. While previous research has successfully identified general relationships, it has omitted a significant implementation aspect of JIT, namely continual improvement over time. This study addresses this dynamic nature of the JIT implementation process. Learning curves are incorporated to analyse the impact of continual reduction of processing time variation over time. The results provide information on the relationship between the level of processing time variation, the output rate of the system, and inventory between work processes. The methodology used in this study incorporates an additional dimension in analysing JIT. It also provides a general and effective tool for decision makers facing the complex issue of implementing pull production processes. Inventory accumulates between workstations in traditional manufacturing settings because: (1) processing times for the sequence of workstations are not balanced; (2) set-up costs for differing processes require large batch sizes and; (3) processing time variability within the workstation creates the need for buffer stock between stations to keep the production line flowing smoothly. In a pull orJIT production process, it is generally assumed that line imbalances have been or can be removed and that set-up costs have Planning for been reduced such that batch sizes approaching the objective of one are Continual realistic[1,2]. It cannot, however, be assumed that processing time Improvement variability can be completely eliminated. Because of this inability to eliminate processing time variability several studies have addressed the issue of providing a buffer stock in pull systems by 5 increasing the number of kanbans between work processes to some (so called) optimal level[3-13]. These studies have addressed the variability in processing time in a pull production system using simulation analysis and Markov processes. Results indicate that, if processing time variability cannot be eliminated, the theoretical objective of a single kanban may not be appropriate. When this occurs, it may be necessary to increase the number of kanbans in order to achieve production rates consistent with the production plan. While these studies generally test multiple levels of variation, they fail to address the issue of their changing levels over time. Rees et al.[10] presented a model which adjusted the number of kanbans for different time-frames based on estimates of lead times for the given period, but failed to incorporate the continually improving nature of JIT. In one example, however, they do suggest that training programmes will indeed have an effect on the variability of processing times. This study incorporates the dynamic nature of JIT by considering the reduction of processing time variation over time. This time variation is incorporated through the use of learning curves. Simulation experiments investigate the relationship between the output rate, the variation of processing time, and the number of kanbans between work centres in an environment where processing time variation is continually reduced. Two experiments are performed. First, a pull production process is implemented with a single kanban between operations, thus representing the scenario of instant or crash conversion. Here the continual reduction of processing time variability leads to increasing output rates. The results of this experiment provide information about the impact of such a conversion on the output rate of the production process. The second experiment represents a gradual or phased implementation of the kanban system. Here a minimum acceptable level of output is specified; then the number of kanbans between work processes is determined, such that this minimum output level is achieved. Inventory levels (or the number of kanbans) between work processes are reduced over time, as the level of processing time variation is reduced. Learning curves have been applied to several comparable decision situations; yet up to this point they have not been incorporated in analysing the implementation of pull production processes. Person[14], in a comprehensive review of the analytical and simulation studies of JIT, noted that one of the missing areas in the literature is the lack of research addressing the concept of learning in JIT environments. This article specifically addresses this issue. Note that the learning effects are applied only to the variability of processing time and not to the standard time required to produce an item. While standard © Emerald Backfiles 2007 IJOPM 13,6 © Emerald Backfiles 2007 time could also be reduced over time, we contend that, for many production processes, it may not be feasible or practical to reduce it. Further, such a reduction in standard time may hinder the ability to reduce processing time variability and, in fact, variability may actually increase in an attempt to speed up a process. Hence, in this study, standard processing time was based on realistic levels and remained constant. The article is organized into the following sections: Description of the Simulation Experiments, Simulation Results, Implications for Decision Makers, and Conclusions. Description of the Simulation Experiments The objective of these experiments was to model the implementation of a pull production process which incorporated continual improvement by reducing processing time variability over time. The simulation package used to model this scenario was SIMAN[15]. The structure of the pull production process modelled in this study was the same as that used by Huang et al.[5], i.e. a three- line, four-stage production process. This structure is displayed in Figure 1. It should be noted that the procedure described in this study can be used for any product structure. That described in Figure 1 was chosen only as an example. A second aspect of the production process which was specified for the simulation experiments was the distribution used to describe processing time. In general, the characteristics of processing times can be described as being that: (1) they exist only for non-negative values; (2) as variability decreases, the form of the distribution changes from: O monotonic decreasing; to O unimodal distributions heavily skewed to the right; to 0 normal type distributions, truncated at zero. In addition, because these experiments address the dynamic, continually improving JIT environment, it is also necessary that the distribution selected to describe processing times should be capable of describing continuous, alternative levels of variation. Several distributions, such as the Gamma, truncated normal, and Weibull, satisfy all these requirements. A preliminary study indicated that the Gamma distribution was best suited to describe processing time in pull production processes because of its flexibility, computational efficiency, and ease of use. Several assumptions were made in the simulation model which depict aspects of the JIT philosophy. For example, it was assumed that the production line was balanced. Further, in keeping with the JIT philosophy, it was assumed that raw materials were purchased under JIT arrangements from qualified suppliers, and, therefore, no stock-out of raw materials occurred. It was also assumed that final assembly was completed according to a frozen master production schedule, i.e. the demand was deterministic. This master Planning for Continual Improvement 7 ———‘ Raw materials Figure 1. A Multiline, Multi-stage Production Process production schedule was divided into time-buckets. The standard processing time for each process was one and each time-bucket contained 500 processing time-units. Hence maximum capacity for each time-bucket called for 500 units to be placed in finished goods inventory. A time-bucket could refer to a shift, a day, a week, or whatever time-frame is appropriate for a given situation. The decision maker can determine at what point learning takes place and apply the time-bucket concept on that basis. Experiment I The parameters of the Gamma distribution used to describe processing time were specified to provide a standard processing time of one and a standard deviation of one, resulting in an initial coefficient of variation of one. Each time- bucket consisted of 500 processing time-units. The number of kanbans between work stages was specified as one and the learning rate was specified. After reaching steady state, the pull production process was simulated for one time- bucket, starting with time zero and ending with time 500. The number of units produced was recorded. Next the learning rate was used to determine the reduction in coefficient of variation for the succeeding time-bucket. Owing to learning, the processing time variation in each succeeding time-bucket was lower that that of its preceding time-bucket. The degree of reduction was marginally reduced over time in accordance with standard learning curve logic. The parameters of the © Emerald Backfiles 2007 IJOPM 13,6 Figure 2. Flow Chart Description of Experiment I © Emerald Backfiles 2007 Gamma distribution were modified to reflect the reduction in processing time variability. The pull production process was then simulated using these modified Gamma parameters. This process was repeated for 365 time-buckets. The experiment described above was replicated 100 times and the output rate was determined for each of the 365 time-buckets based on the average production output of the 100 replications. The entire experiment was than repeated for alternative learning rates of 95, 90, 85, 80, 75, 70, 65 and 60. Note that lower learning rates indicate faster learning. Figure 2 provides a flow chart which describes the first simulation experiment. Experiment [I The objective of the second simulation experiment was to investigate the impact of learning on the number of kanbans required between processes in order to maintain a specified output rate. As in Experiment I, it was initially Initialize kanban (K = 1) Set learning rate 95 90 85 80 75 70 65 60 Initialize replicate (Ft = 1) Initialize CV and time-bucket (CV: 1.0 andT: 1) Execute simulation and record Increment trme- end of day production bucket (T=T+ 1) and update CV as per learning rate increment . replicate replicate _ equal to (R ' R + 1) 100? Calculate average production for each time-bucket all learning rates been assumed that the number of kanbans was one. It was further assumed, for Planning for example purposes only, that an average output rate of 90 per cent of maximum Continual capacity was acceptable. Therefore, in each time-bucket, the number of Improvement kanbans was increased until the average output rate reached a specified level of 90 per cent. Owing to learning and the consequential reduction in processing time variability, it was expected that the number of kanbans would reduce over time until no measurable improvements in processing time variability existed. 9 In the first step of this simulation experiment, the parameters of the Gamma distribution were once again set to provide a mean of one and a standard deviation of one. As in Experiment I, after reaching steady state, the pull production process was simulated for 500 processing time-units or one time- bucket and the output achieved was recorded. Next, as described in Experiment I, the learning rate was used to determine the reduction in coefficient of variation for the succeeding time-bucket. The parameters of the Gamma distribution were modified to reflect the reduction in processing time variability and the simulation was repeated for that time- bucket. This process was repeated for 365 time-buckets. The simulation was replicated 100 times. The average output rate of the system was determined for each time-bucket as the average output from the 100 replications. For each time-bucket, if the average output rate was at or above the desired production level, the number of kanbans required to provide the desired level of production had been achieved. For each time-bucket in which the average production level was less than the desired level, the number of kanbans was incremented by one and the simulation repeated. This process was repeated until the desired production level was reached for each of the 365 time- buckets. The entire simulation was then repeated for multiple learning rates ranging from 95 to 60, as in Experiment I. Figure 3 provides a flow chart which describes the second simulation experiment. This experiment was also conducted for several alternative levels of minimum output rate. While the specific numbers achieved from these additional simulations varied, the results and conclusions were not affected. As a result, only those results obtained with a desired output rate of 90 per cent of maximum capacity are presented here. Simulation Results The results of Experiment I, where the consequences of reducing the variability of processing time over time were measured in terms of the output rates, are presented in Table I and Figure 4. Table 1 provides the average output rate achieved during several selected time-buckets for the eight learning rates considered in the study. These results are graphically displayed in Figure 4. Here the average output level for learning rates of 90, 75 and 60 are presented for all 365 time-buckets. Table I and Figure 4 indicate that learning, and the resulting reduction in processing time variation, have a positive impact on the output rate. It is important to note that, in both simulation experiments, the standard processing © Emerald Backfiles 2007 IJOPM 136 Set learning rate 95 90 85 80 75 70 65 60 Initialize kanban (K = 1) 10 Set time check (TCHK = 365) Initialize replicate (R = 1) Initialize CV and time-bucket CV = 1.0 andT: l) Execute simulation and record end of day production Increment time bucket (T = T + 1) and update CV as . Does per Ieaming rate T = TCHK? Yes Increment replicate (R = R + 1) Decrement time check (TCHK = TCHK-1) Calculate avera e production tortime-buc at check Increase , kanban average productio (K=K+1) >=90percent? Record kanbans Does time-bucket check TCHK = 1? and learning rate Have all learning rates Figure 3. been run? Flow Chart Description of Experiment II time was held constant at one. Hence the maximum capacity in each time- bucket also remained constant at 500 units per time-bucket. Because the standard processing time remains constant, yet the average production per time-bucket increases over time, as the processing time variation decreases, it is necessary to differentiate the standard processing time per unit from the average per unit processing time. The average time required to produce a single unit will decrease in accordance with the learning curve applied to processing time variation, even though the standard time is not changed. As a result, any © Emerald Backfiles 2007 Planning for Learning rate Continual Per cent Time-bucket 95 90 80 75 70 Improvement 11 1 2 3 4 5 6 7 8 Table I. Experiment 1. Per cent of Maximum Capacity Attained with One Kanban between Workstations © Emerald Backfiles 2007 IJOPM 13,6 12 Figure 4. Increase in Average Production over Time with the Gamma Distribution and K=1 © Emerald Backfiles 2007 OLeaming rate = 90 per cent ALearning rate = 75 per cent El Learning rate = 60 per cent >. .2: u a o. «a o c .9 .. u :1 '0 9 a. E a .E x a: E ._ o .. c a) o i. a) Q. 100 150 200 250 300 350 Production time period increase in the actual (realized) output rate can be attributed directly to the reduction in the standard deviation of processing time. Table I and Figure 4 also show that the increase in the output rate is a function of the rate of learning. These results indicate that, if an organization wishes to obtain a specified output rate within a specified time~period, management must implement programmes to assist the employees in the learning process. For example, assume it is desired to achieve a level of 90 per cent of maximum capacity. If the initial standard processing time and standard Planning for deviation are both one and the learning rate is 95 per cent, it would not be Continual practically possible to achieve this capacity at any point in the future with a Improvement single kanban. On the other hand, for the same system parameters, but with a 60 per cent learning rate, 90 per cent of maximum capacity is achieved in less than 30 time-buckets, and an output rate of 95 per cent achieved in less than 80 time-buckets with a single kanban. 13 The results of Experiment II, where the consequences of reducing the — variability of processing time over time were measured in terms of the potential for reducing the number of kanbans between work stages, while maintaining a specified output rate, are presented in Table II and Figure 5. Table 11 provides, for the eight learning rates considered in this study, the number of kanbans required to provide an output rate of 90 per cent for several selected time- buckets. Figure 5 graphically displays the information for learning rates of 90, 75 and 60 for all 365 time-buckets. These results show that, if an organization wishes to maintain a specified output rate, it may not be possible to start with a single kanban, but that the number of kanbans can be gradually reduced over time, as processing time variability decreases. The eventual goal would naturally be to achieve the ideal of a single kanban between stages. For example, if the initial standard processing time and standard deviation are both one and the learning rate is 95 per cent, it would be practically possible to reduce the number of kanbans to one, even after 365 time-buckets. Such a system would require at least eight kanbans to maintain the specified level of output. On the other hand, for the same system parameters, but with a 60 per cent learning rate, the number of kanbans can be reduced to one in just 26 time~buckets. In other words, after 26 applications of learning, only one kanban would be required to maintain an output rate of 90 per cent. These results, like the results from Experiment I, are extremely important to decision makers, considering the type of employee involvement programmes to implement and to analyse the cost of the alternative programmes. The results provided in Table II and Figure 5 also show the incremental impact of higher rates of learning. For example, consider two firms implementing pull production systems, both of which have a standard processing time of 1.0 and a standard deviation of 1.0. The results provided in this study indicate that, at the time of implementation, 16 kanbans will achieve an output rate of 90 per cent. Assume that the learning rate of the first organization is 95 per cent and that of the second organization is 75 per cent. In ten time-periods, the number of kanbans required to maintain the output rate of...
View Full Document

What students are saying

  • Left Quote Icon

    As a current student on this bumpy collegiate pathway, I stumbled upon Course Hero, where I can find study resources for nearly all my courses, get online help from tutors 24/7, and even share my old projects, papers, and lecture notes with other students.

    Student Picture

    Kiran Temple University Fox School of Business ‘17, Course Hero Intern

  • Left Quote Icon

    I cannot even describe how much Course Hero helped me this summer. It’s truly become something I can always rely on and help me. In the end, I was not only able to survive summer classes, but I was able to thrive thanks to Course Hero.

    Student Picture

    Dana University of Pennsylvania ‘17, Course Hero Intern

  • Left Quote Icon

    The ability to access any university’s resources through Course Hero proved invaluable in my case. I was behind on Tulane coursework and actually used UCLA’s materials to help me move forward and get everything together on time.

    Student Picture

    Jill Tulane University ‘16, Course Hero Intern