Browsing by Author "Philip, Deepu"

Now showing 1 - 2 of 2

Robust schedules and disruption management for job shops
(Montana State University - Bozeman, College of Engineering, 2008) Philip, Deepu; Chairperson, Graduate Committee: Edward L. Mooney
This dissertation documents the results of research evaluating policies to schedule for unanticipated disruptions in job shops. The disruptions studied in this research are of two types - machine failure and job release-time. The study was conducted using modified classical job shop problems with the minimize maximum completion time (Cmax) objective. Best random non-delay schedules (BRS) provided job sequences for each machine. Different slack policies based on frequency, duration and location of slacks in the schedule were used to strategically insert slack in the BRS schedule. The resulting robust schedules proactively managed machine or job releasetime disruptions. The change in Cmax quantified schedule robustness. Simulation was used to generate and modify the BRS schedules with and without slacks and disruptions, simulate disruption events and evaluate schedule performance. It was observed that policies that equally distribute slack to all tasks on heavily utilized machines and those that equally distribute slack to all tasks on all machines performed best. When the number of jobs to process was more than the number of machines and "big jobs" with long processing times on heavily utilized machines were present, the policy distributing slack tasks equally on heavily utilized machines exhibited superior performance. For systems with less variability both policies performed equally well. By comparing the average, minimum and maximum schedule deviations across all policies, the study concluded that strategically distributing slack to tasks on heavily utilized machines results in good robust schedules that can absorb the effects of disruptions.
Scheduling reentrant flexible job shops with sequence dependent setup times
(Montana State University - Bozeman, College of Engineering, 2005) Philip, Deepu; Chairperson, Graduate Committee: Edward L. Mooney
This study presents a new simulation-based local search approach for solving shop scheduling problems. Results for classical problems from the literature demonstrate the effectiveness and quality of the approach. Application is also shown for reentrant flexible job shop with sequence dependent setup times (RFJSSDS), a new, very general, class of problems. RFJSSDS is a generalization of the classical job shop, reentrant flow shop and flexible job shop problems. Multiple products (routes), sequence dependent setup times at the work centers and reentrancy of the jobs make RFJSSDS one of the more general and difficult shop scheduling problems. Examples of this type of problem include semiconductor wafer fabrication facilities and flexible machining systems. The solution methodology developed in this study features a new Simulation Based Local Improvement with Multi Start (SBLIMS) algorithm. The local search procedure modifies an initial feasible solution provided by the simulation module to generate promising neighbor solutions. A generated solution is considered to be better if there is a reduction in the total completion time or makespan. A unique filtering strategy is used to select and rank moves, using both task and resource views of a schedule. Multiple random starting points are generated in multistart fashion as part of the solution process. New theorems are presented that form the basis for SBLIMS. The SBLIMS algorithm was evaluated using test instances for several shop scheduling problems as well as RFJSSDS. A set of synthetic problems was generated to study RFJSSDS, because there were no RFJSSDS instances available from the literature. The SBLIMS algorithm was compared with various dispatch rules in the RFJSSDS domain and its performance was found to be better in most cases. SBLIMS was also tested with well known special cases of RFJSSDS: the classical job shop, reentrant flow shop and flexible job shop problems. The SBLIMS algorithm provided excellent results compared with those provided in the literature, establishing the generality of the approach for solving a broad class of shop scheduling problems.