A scheduling rule discovery and parallel learning system
Abstract
Priority dispatching rules are heuristic methods for scheduling problems that have been studied for several decades. Their popularity is due to their ability to rapidly generate reasonably good solutions in practical settings. However, when the industrial applications become more complex, customized priority rules are often required. A user wishing to develop customized scheduling rules must first develop a simulation model of the facility in question and implement different scheduling rules to evaluate their performance under realistic operating scenarios. While most commercial factory simulation software include a limited set of predefined rules that can be easily implemented, evaluating customized rules generally involves coding on the part of the user, which is often time-consuming, especially since many scheduling personnel are not skilled programmers. In this research, we develop a flexible, general representation for a wide range of dispatching rules. This in itself can permit automatic generation of code for the rules, significantly simplifying the task of coding alternative dispatching rules. However, the main research contribution is that this representation now allows us to view the problem of obtaining the best dispatching rule for a given production environment as that of searching the space of rules defined by this representation. We use this representation to develop an autonomous learning mechanism that is capable of discovering effective dispatching rules for a given production environment. It possesses the ability to identify significant relationships between system attributes and problem structure. This is a significant step beyond current applications of artificial intelligence (AI) to production scheduling, which are mainly based on learning how to select a given rule from among a number of candidates, rather than identifying new and potentially more effective rules. We evaluate the performance of the proposed learning system in two different production scheduling environments under a range of shop conditions. The learning system discovers scheduling rules that are competitive with those in the literature, which are the results of decades of research.
Degree
Ph.D.
Advisors
Uzsoy, Purdue University.
Subject Area
Industrial engineering|Systems design
Off-Campus Purdue Users:
To access this dissertation, please log in to our
proxy server.