مقایسه کاهش واریانس با مدیریت واریانس سیستم در یک تولید کارگاهی

Variance within the manufacturing system leads to uneven shop loads, long manufacturing lead times, and unreliable customer service. This study compares techniques that reduce system variance to techniques that manage system variance. The study is placed in a dual resource constrained job shop. Results indicate that reducing system variance improves flow time and customer service performance measures, such as mean tardiness and percent tardy jobs more than techniques that react to system variance.

Schmenner and Swink (1998) argue that the key to improving productivity is to improve the continuity of work flow and decrease its variability. They suggest that this can be done by cross-training the work force, conducting faster equipment changeovers and instituting other policies that will reduce variance. This theory is partially supported by Schmenner (2001) using historical evidence. But, there has been no review of the existing literature to determine whether prior research findings support their theory. As an initial step in testing their theory, this paper first examines and classifies types of management decisions in terms of their influence on variance. This classification suggests that while there are several ‘causes’ of variability, the primary (important) sources of variability are the manner in which orders (demand) arrive into the shop, and the amount of work that is associated these orders. Management may treat this natural variance as a given and simply react to it, or it may choose to modify or influence the manner in which this variance is realized by the shop. The commonly used methods of coping with system variance, such as scheduling and flexibility (e.g. labor flexibility) are also studied to understand how they influence shop flow. The simulation experiment is placed in a dual resource constrained (DRC) job shop to examine whether it is better to eliminate variance or to react to variance. In a DRC shop both the machines and the workers can become the constraint. A DRC environment is chosen for this study, since it is representative of the job shops commonly found in actual practice (Wisner & Siferd, 1995).

A major purpose of our paper was theory testing. Specifically, our work sought to verify whether the theory of ‘continuity of work and decreased variability’ which is implicit in the ‘theory of swift and even flow’ can be accomplished via reactive variance reduction strategies alone, or whether proactive strategies were more effective in accomplishing this. So, the study compared the two reactive variance control strategies (i.e. worker flexibility and dispatching rules) to the use of two proactive variance elimination strategies (i.e. reduced variance in job arrivals and processing times and the setting of realistic, internal due dates). This paper is a step towards a fuller understanding of how management decisions in each of the eight categories of manufacturing strategy increase or decrease system variance. This study was placed in a DRC job shop, but it has implications for larger systems. For example, Lee, Padmanabhan, and Whang (1997) examined how management decision in a supply chain can increase demand variance upstream in the chain. The results did not entirely support the model shown in Fig. 1. The two dispatch rules tested here (FCFS and EDD) and the due date rules had little influence on flow time. However, reducing variance in order arrivals and job processing times did significantly reduce flow times. At the same time worker flexibility also reduced flow time. However, when the arrival and/or processing variance was low, flexibility did not decrease flow time further. This partially supported the model in Fig. 1. Earlier results (Park, 1991) had suggested that the effectiveness of dispatching rules decreases with increased levels of worker flexibility, and that dispatching rules are effective when the shop does not utilize worker flexibly (i.e. Flex=1). The present study confirms these results, and suggests that when processing variance is limited dispatching rules have no effect. So, in that environment simple rules such as FCFS perform as well as the EDD rule. As shown in Table 2, in many cases the improvement in customer service from controlling processing time variance is greater than that possible through the introduction of worker flexibility. This result generally holds true for both dispatching rules, and at both utilization levels. As an example, consider a DRC shop with exponential arrivals and processing times, which uses the EDD rule is used for dispatching. Increasing worker flexibility in this shop reduces mean tardiness from 0.62 to 0.16 h, and percent tardy jobs from 37.05 to 9.07%. In comparison, the DRC shop with exponential arrivals, uniform processing times, and EDD dispatching rule yields a mean tardiness of 0.16 h and percent tardy jobs of 7.65% without the use of incremental flexibility. Several previous studies have documented the beneficial impact of worker flexibility on customer service measures. While the results of this study confirm their findings, it adds to their results by showing that greater improvements in customer service performance can be obtained by controlling job processing times in a DRC environment. These findings are displayed in Fig. 3a and b for the FCFS and EDD rules, respectively. The results for the mean tardiness performance measure are very similar to those for the mean flow time. When both the arrival and processing variances were small, there was very little difference in mean tardiness between treatments. The arrival variance did not influence mean tardiness as much as processing variance, but that is understandable. Since the due date is set after the arrival, there would be some adjustment. So the setting of the due date actually reduces variance in tardiness due to the arrival variance. As expected, the EDD dispatch rule had lower mean tardiness than FCFS. A portion of the model in Fig. 1 was supported. The dispatch rule did interact with processing variance. Worker flexibility did interact with both arrival and processing variance. They helped reduce mean tardiness at both levels of arrival variance, but only reduced the mean tardiness at high levels of processing variance. Dispatching rules, due date setting methods, and order arrival distributions also had a statistically significant impact on customer service performance. The direction of results for customer service with respect to dispatching rules is similar to that reported in the previous literature. EDD results in lower values of mean tardiness and percent tardy jobs as compared to FCFS. However, the performance of the dispatching rules is also affected by the processing time variability. As an example, mean tardiness and percent tardy jobs for the FCFS rule in the environment with exponential arrivals, Flex=1, and uniform processing time are 0.39 h and 18.70%, respectively. These values are lower than those for EDD in the shop with exponential arrivals and no inter-departmental worker flexibility (Flex=1), but where processing times are highly variable (mean tardiness=0.62 h, and percent tardy=37.05%). The performance of dispatching rules for customer service measures is affected to a lesser degree by due dates setting methods, and order arrival distributions. In general, Table 2 shows that internally set due dates and less variable order arrivals allow for reductions in mean tardiness and percent tardy jobs under both utilization levels. However, the reductions in customer service measures due to these two factors are less than those achieved by controlling processing time variance, and introducing incremental worker flexibility. Table 2 also shows that the difference between the performance of dispatching rules diminishes in DRC shops where the variance with respect to order arrivals, due date setting methods and job processing times can be controlled. In these shops, variance control facilitates for a near-perfect delivery performance even in the absence of inter-departmental flexibility. The main findings of this study can be summarized as follows: 1. Of the three proactive variance control strategies considered in this study, controlling variance associated with processing times appeared to offer the greatest benefits. 2. Incremental worker flexibility, operationalized by requiring workers to process jobs in two different departments in the shop has the greatest impact on inventory performance. The benefits of incremental worker flexibility on inventory performance appear to be greatest when system variance cannot be controlled. Conversely, the benefits of incremental flexibility are reduced when processing time variability is reduced on the shop floor. 3. Dispatching rules appear to be more effective only when job processing times are highly variable. However, if the shop is able to reduce the variability associated with job processing times, the simple FCFS rule performs as well as the EDD rule for mean flow time. 4. The effect of reducing processing time variance on mean tardiness and percent of tardy jobs is greater than that of incremental worker flexibility. Recent surveys indicate that a large number of firms are emphasizing customer service as a competitive strategy. The present study showed that such an objective can be supported through proactive strategies aimed at controlling system variance. Empirical research is needed to determine the variety of DRC environments that exist and their actual operating conditions. Important issues include (1) how to quantify the costs of training workers to increase their flexibility and (2) how to quantify the benefits of reduced flow times. This empirical data will allow researchers to determine which variables are the most valuable to examine. Given that variance reduction is the most important influence on both flow times and tardiness measures, it is important to understand which techniques used by managers effectively reduce either arrival or processing time variance. These four findings do provide partial support for Schmenner and Swink's (1998) proposition that work flow continuity and variance reduction improve shop performance. While all of the techniques proposed for managing variance were not effective, it is clear that reducing variance improved shop performance. Further, research is needed to test this theory and to determine the conditions where it is applicable and where it may fail.