A genetic algorithm for reactive scheduling based on real-time manufacturing information

In real manufacturing system of moulds production, many unexpected events (e.g., processing time variability, job revision, urgent jobs arrival, etc.) often lead to numerous schedule disruptions. The random dynamic characteristics of the scheduling environment render the baseline schedule made off-line infeasible when applied to practical problems. Based on real-time manufacturing information getting from the manufacturing executing system (MES), we would quickly revise the baseline schedule that has suffered from disruptions during schedule execution. In this paper, we restrict ourselves to the reactive scheduling problem encountered in the manufacturing moulds. First, we model the job shop scheduling problem as a dynamic constraint satisfaction problem, where the due date constraint is considered as tightest condition, since a late job completion will incur a large penalty payments to the customers. After a distribution to the schedule, instead of total rescheduling, we present a reactive local repair approach, based on genetic algorithm to make the revised schedule deviate from original schedule as little as possible. The object is to minimize the distance between two schedules, defined as the weighted sum of the absolute deviations between the planned and realized operation start and finish times. In experiment studies, we consider the main uncertainty: process duration variability, and suppose that it is subjected to a uniform distribution. The experimental results suggest that the proposed approach can increase the on-time delivery rate for real situations.