<p>Efficient scheduling in discrete manufacturing environment can improve productivity through reduced work-in-progress and finished good inventories. This research examined the feasibility and advantages of using genetic algorithms in manufacturing scheduling. All literature to date indicate that there is a need for better scheduling algorithms which can provide better and faster solutions. This thesis focuses on the improvement of the scheduling performance through: a) the application of genetic algorithms to the schedule optimization problem, b) the consideration of batch splitting and c) dynamic scheduling.</p> <p>First, genetic algorithms are developed for two types of scheduling situations: a) scheduling in the presence of single process plans without any routing flexibility and b) scheduling where routing flexibility and multiple process plans exist. Analytical techniques and heuristics-based methods have been developed by other researchers to solve such scheduling problems which are useful for limited size problems. Since the schedule population size is fixed in genetic algorithms, there is a tremendous reduction in the memory requirements. In every generation bad schedules are replaced by better ones. Also, since genetic algorithms performs a multi-point search at a time, they are relatively faster than other techniques used for scheduling which are usually based on single point search. Genetic algorithms models were formulated and tested using nineteen example problems for both flexible and fixed routing scenarios. Numerical results show that the genetic algorithms approach perform significantly better than other approaches both in terms of funding the optimal/near optimal solution and the computation time. The saving in computation time was up to 50% in some cases. Because of the vast number of schedules available in the case of multiple process plans, dispatching rules have been employed in the genetic algorithms to obtain a satisfactory schedule. It is often observed that in the case of alternate routings and while using dispatching rules, researchers tended to use a single job in each order which is rarely the case in practice. In this thesis, both alternate routings and order sizes greater than one are simultaneously considered along with dispatching rules while scheduling with genetic algorithms. Twelve different dispatching rules and seven performance criteria are used and the performance of each rule has been studied extensively with respect to each of the performance criteria.</p> <p>Second, three new batch splitting policies were introduced in this research to split the batches into smaller sub-batches. A process plan-based method was developed, where the information available from the process plans is used for batch splitting. This is different from other batching policies where batching is done based on inventory models from queueing theory for which cost function consisting of ordering, inventory holding and work-in-process carrying costs are required. The new batch splitting policies were compared with existing ones and the results show that in several cases the newly developed policies performed significantly better than existing ones. The improvement in the performance was observed to be in the range of 35% to 45%.</p> <p>Next, the dynamic scheduling aspects are considered and rescheduling algorithms were developed for four kinds of uncertainties. These include unexpected machine breakdown, rush order arrival, order cancellation and increased order priority. These algorithms reschedule only interrupted tasks, hence focusing on local rescheduling. These algorithms were tested using three different performance criteria namely, mean flow time, mean tardiness, and average machine utilization. The developed rescheduling algorithms proved successful in dealing with the shop floor uncertainties and the results indicate that the proposed algorithms are effective and could easily be used to dynamically schedule the manufacturing systems in the presence of disturbances.</p> <p>Finally, as part of the system implementation, a user interface has been developed in 'C' to input required scheduling data used for schedule optimization. The scheduler outputs a list of several good schedules. These schedules are then analyzed by the output analyzer which outputs a list of different performance criteria values, machine utilizations, ready and completion times of each order and the Gantt chart of the best schedule.</p> / Doctor of Philosophy (PhD)
| Identifer | oai:union.ndltd.org:mcmaster.ca/oai:macsphere.mcmaster.ca:11375/6982 |
| Date | 03 1900 |
| Creators | Jain, Kumar Ajay |
| Contributors | ElMaraghy, Hoda A., Mechanical Engineering |
| Source Sets | McMaster University |
| Detected Language | English |
| Type | thesis |
Page generated in 0.0134 seconds