Executing production schedules in the face of uncertainties: a review and some future directions
Citations
881 citations
786 citations
Cites background from "Executing production schedules in t..."
...Dynamic scheduling has been defined under three categories (Mehta and Uzsoy 1999; Vieira et al. 2000a, 2003; Aytug et al. 2005; Herroelen and Leus 2005): completely reactive scheduling, predictive–reactive scheduling, and robust pro-active scheduling....
[...]
...…priority, changes in job processing time, etc. Dynamic scheduling has been defined under three categories (Mehta and Uzsoy 1999; Vieira et al. 2000a, 2003; Aytug et al. 2005; Herroelen and Leus 2005): completely reactive scheduling, predictive–reactive scheduling, and robust pro-active scheduling....
[...]
...The periodic and hybrid policies have received special attention under the name rolling time horizon (Church and Uzsoy 1992; Ovacik and Uzsoy 1994; Sabuncuoglu and Karabuk 1999; Vieira et al. 2000a; Aytug et al. 2005)....
[...]
353 citations
315 citations
Cites background from "Executing production schedules in t..."
...ease of implementation and their flexibility to incorporate domain knowledge and expertise [69] explain the wide usage of dispatching rules in practice [70] and the ongoing research...
[...]
288 citations
Cites background from "Executing production schedules in t..."
...The baseline schedule serves very important functions (Aytug et al. 2003, Mehta and Uzsoy 1998)....
[...]
References
1,726 citations
"Executing production schedules in t..." refers background in this paper
...The potential impact of such disruptions can be quite high, as evidenced by the well-studied ‘‘bullwhip effect’’ (Chen et al., 2000) in supply chains, that causes variation at downstream nodes in the supply chain to be amplified at upstream stages....
[...]
...The potential impact of such disruptions can be quite high, as evidenced by the well-studied ‘‘bullwhip effect’’ ( Chen et al., 2000 ) in supply chains, that causes variation at downstream nodes in the supply chain to be amplified at upstream stages....
[...]
1,579 citations
"Executing production schedules in t..." refers background in this paper
...An important paper in this area is that of Lawrence and Sewell (1997), who compare the performance of a global scheduling heuristic based on the shifting bottleneck algorithm of Adams et al. (1988) with myopic, completely reactive dispatching rules in the presence of uncertain job processing times. They demonstrate that as processing time uncertainty increases, the difference in performance between the global method and the dispatching rules becomes less significant. They conclude that in systems with high uncertainty, completely reactive algorithms can be used with relative confidence, and question the benefits of global scheduling procedures in general. In contrast, Barua et al. (2001) propose another approach in which a global schedule for the factory is developed using a periodic rescheduling policy. This schedule is not implemented directly, but rather serves to provide a priority index for the jobs as execution unfolds. The jobs are dispatched at the machines based on their start times in the global schedule. Extensive simulation experiments show that under a variety of operating conditions, including processing time uncertainty and machine failures, this approach significantly outperforms myopic dispatching rules. However, as the level of uncertainty becomes high relative to the frequency of rescheduling, performance becomes deteriorates to a level comparable to that of myopic dispatching rules. Honkomp et al. (1999) describe a simulator for semi-continuous and batch processing manufacturing environments that can accept deterministic schedules and simulate both a deterministic and a stochastic realization of the schedule. The stochastic version can also use rescheduling logic. Running two versions of the simulation the authors compare the performance and robustness of the schedules. Two metrics are used for comparison. PB 1⁄4 OF=OFDB is a measure of how well the average objective function value of the stochastic simulation compared to the objective function of the best deterministic schedule. DB 1⁄4 SD=jOFDBj is the standard deviation of the replicas of stochastic version compared to best deterministic objective function. This is used as a measure of robustness. In simulations without rescheduling schedules with the best performance also had the best robustness which is somewhat counter intuitive. In cases with rescheduling, rescheduling strategy with no penalties (i.e., can reschedule anything in the future) or no rescheduling created the best performance. Again those that had the best performance had the best robustness. Matsuura et al. (1993) provide an extensive study of a slightly different rescheduling policy. In their approach, called switching, a predictive schedule is developed on a periodic basis. However, if the realized schedule is deemed to have deviated sufficiently from the predictive one, the system switches to using a dispatching rule for the remainder of the period. This approach is contrasted with using the predictive schedule throughout the period (by right-shifting jobs when delays occur) and dispatching approaches. They focus on three different types of disruptions: rush order arrival, specification changes (which cause new operations to be added to a job, or existing operations to be deleted), and machine failures. Their results are quite insightful: they show that when the frequency of disruptions is low, the predictive/reactive approaches outperform the dispatching. Once the level of disruption reaches a certain level, however, the dispatching begins to perform better than the predictive/reactive approaches. We believe that the answer to this debate lies in the results of Matsuura et al. (1993) and Lawrence and Sewell (1997), and is hinted at in the results of several other papers. In an environment with little uncertainty, predictive/reactive methods based on global information and optimization techniques are highly likely to yield better schedules than completely reactive dispatching procedures. However, once the variability in the system exceeds a certain level, which appears to be system-dependent, the global information on which the predictive/reactive approaches are based becomes invalid, causing them to generate poor schedules due to solving the wrong problem: the problem data they use does not correspond to the problem encountered on the shop floor. Having agreed with Lawrence and Sewell (1997) thus far, however, we do not believe that this insight should push us to disregard work on predictive/reactive scheduling methods....
[...]
...An important paper in this area is that of Lawrence and Sewell (1997), who compare the performance of a global scheduling heuristic based on the shifting bottleneck algorithm of Adams et al. (1988) with myopic, completely reactive dispatching rules in the presence of uncertain job processing times....
[...]
...An important paper in this area is that of Lawrence and Sewell (1997), who compare the performance of a global scheduling heuristic based on the shifting bottleneck algorithm of Adams et al. (1988) with myopic, completely reactive dispatching rules in the presence of uncertain job processing times. They demonstrate that as processing time uncertainty increases, the difference in performance between the global method and the dispatching rules becomes less significant. They conclude that in systems with high uncertainty, completely reactive algorithms can be used with relative confidence, and question the benefits of global scheduling procedures in general. In contrast, Barua et al. (2001) propose another approach in which a global schedule for the factory is developed using a periodic rescheduling policy. This schedule is not implemented directly, but rather serves to provide a priority index for the jobs as execution unfolds. The jobs are dispatched at the machines based on their start times in the global schedule. Extensive simulation experiments show that under a variety of operating conditions, including processing time uncertainty and machine failures, this approach significantly outperforms myopic dispatching rules. However, as the level of uncertainty becomes high relative to the frequency of rescheduling, performance becomes deteriorates to a level comparable to that of myopic dispatching rules. Honkomp et al. (1999) describe a simulator for semi-continuous and batch processing manufacturing environments that can accept deterministic schedules and simulate both a deterministic and a stochastic realization of the schedule....
[...]
...An important paper in this area is that of Lawrence and Sewell (1997), who compare the performance of a global scheduling heuristic based on the shifting bottleneck algorithm of Adams et al. (1988) with myopic, completely reactive dispatching rules in the presence of uncertain job processing times. They demonstrate that as processing time uncertainty increases, the difference in performance between the global method and the dispatching rules becomes less significant. They conclude that in systems with high uncertainty, completely reactive algorithms can be used with relative confidence, and question the benefits of global scheduling procedures in general. In contrast, Barua et al. (2001) propose another approach in which a global schedule for the factory is developed using a periodic rescheduling policy. This schedule is not implemented directly, but rather serves to provide a priority index for the jobs as execution unfolds. The jobs are dispatched at the machines based on their start times in the global schedule. Extensive simulation experiments show that under a variety of operating conditions, including processing time uncertainty and machine failures, this approach significantly outperforms myopic dispatching rules. However, as the level of uncertainty becomes high relative to the frequency of rescheduling, performance becomes deteriorates to a level comparable to that of myopic dispatching rules. Honkomp et al. (1999) describe a simulator for semi-continuous and batch processing manufacturing environments that can accept deterministic schedules and simulate both a deterministic and a stochastic realization of the schedule. The stochastic version can also use rescheduling logic. Running two versions of the simulation the authors compare the performance and robustness of the schedules. Two metrics are used for comparison. PB 1⁄4 OF=OFDB is a measure of how well the average objective function value of the stochastic simulation compared to the objective function of the best deterministic schedule. DB 1⁄4 SD=jOFDBj is the standard deviation of the replicas of stochastic version compared to best deterministic objective function. This is used as a measure of robustness. In simulations without rescheduling schedules with the best performance also had the best robustness which is somewhat counter intuitive. In cases with rescheduling, rescheduling strategy with no penalties (i.e., can reschedule anything in the future) or no rescheduling created the best performance. Again those that had the best performance had the best robustness. Matsuura et al. (1993) provide an extensive study of a slightly different rescheduling policy. In their approach, called switching, a predictive schedule is developed on a periodic basis. However, if the realized schedule is deemed to have deviated sufficiently from the predictive one, the system switches to using a dispatching rule for the remainder of the period. This approach is contrasted with using the predictive schedule throughout the period (by right-shifting jobs when delays occur) and dispatching approaches. They focus on three different types of disruptions: rush order arrival, specification changes (which cause new operations to be added to a job, or existing operations to be deleted), and machine failures. Their results are quite insightful: they show that when the frequency of disruptions is low, the predictive/reactive approaches outperform the dispatching. Once the level of disruption reaches a certain level, however, the dispatching begins to perform better than the predictive/reactive approaches. We believe that the answer to this debate lies in the results of Matsuura et al. (1993) and Lawrence and Sewell (1997), and is hinted at in the results of several other papers....
[...]
...An important paper in this area is that of Lawrence and Sewell (1997), who compare the performance of a global scheduling heuristic based on the shifting bottleneck algorithm of Adams et al. (1988) with myopic, completely reactive dispatching rules in the presence of uncertain job processing times. They demonstrate that as processing time uncertainty increases, the difference in performance between the global method and the dispatching rules becomes less significant. They conclude that in systems with high uncertainty, completely reactive algorithms can be used with relative confidence, and question the benefits of global scheduling procedures in general. In contrast, Barua et al. (2001) propose another approach in which a global schedule for the factory is developed using a periodic rescheduling policy. This schedule is not implemented directly, but rather serves to provide a priority index for the jobs as execution unfolds. The jobs are dispatched at the machines based on their start times in the global schedule. Extensive simulation experiments show that under a variety of operating conditions, including processing time uncertainty and machine failures, this approach significantly outperforms myopic dispatching rules. However, as the level of uncertainty becomes high relative to the frequency of rescheduling, performance becomes deteriorates to a level comparable to that of myopic dispatching rules. Honkomp et al. (1999) describe a simulator for semi-continuous and batch processing manufacturing environments that can accept deterministic schedules and simulate both a deterministic and a stochastic realization of the schedule. The stochastic version can also use rescheduling logic. Running two versions of the simulation the authors compare the performance and robustness of the schedules. Two metrics are used for comparison. PB 1⁄4 OF=OFDB is a measure of how well the average objective function value of the stochastic simulation compared to the objective function of the best deterministic schedule. DB 1⁄4 SD=jOFDBj is the standard deviation of the replicas of stochastic version compared to best deterministic objective function. This is used as a measure of robustness. In simulations without rescheduling schedules with the best performance also had the best robustness which is somewhat counter intuitive. In cases with rescheduling, rescheduling strategy with no penalties (i.e., can reschedule anything in the future) or no rescheduling created the best performance. Again those that had the best performance had the best robustness. Matsuura et al. (1993) provide an extensive study of a slightly different rescheduling policy....
[...]
1,463 citations
615 citations
576 citations
"Executing production schedules in t..." refers methods in this paper
...…Mackulak (1997) provide an alternative formulation of this approach motivated by the controltheoretic models of Gershwin and his co-workers (e.g., Kimemia and Gershwin, 1983) Ashby and Uzsoy (1995) illustrate the performance of a particular under-capacity scheduling scheme in the face of…...
[...]
...Yellig and Mackulak (1997) provide an alternative formulation of this approach motivated by the control-theoretic models of Gershwin and his coworkers (e.g., Kimemia and Gershwin, 1983 ) Ashby and Uzsoy (1995) illustrate the performance of a particular under-capacity scheduling scheme in the face of uncertain arrivals of such orders....
[...]