Deadlock-free genetic scheduling for flexible manufacturing systems using Petri nets and deadlock controllers

Abstract

In this paper, a new deadlock-free scheduling method based on genetic algorithm and Petri net models of flexible manufacturing systems is proposed. The optimisation criterion is to minimise the makespan. In the proposed genetic scheduling algorithm, a candidate schedule is represented by a chromosome that consists of two sections: route selection and operation sequence. With the support of a deadlock controller, a repairing algorithm is proposed to check the feasibility of each chromosome and fix infeasible chromosomes to feasible ones. A feasible chromosome can be easily decoded to a deadlock-free schedule, which is a sequence of transitions without deadlocks. Different kinds of crossover and mutation operations are performed on two sections of the chromosome, respectively, to improve the performance of the presented algorithm. Computational results show that the proposed algorithm can get better schedules. Furthermore, the proposed scheduling method provides a new approach to evaluate the performance of different deadlock controllers.

Deadlock-Free Scheduling Method for Flexible Manufacturing Systems Based on Timed Colored Petri Nets and Anytime Heuristic Search

“…The first example has been described in Section II, the second is an FMS taken from a real automated manufacturing workstation [24], [25], while the last example is a well studied FMS problem with flexible part routing and multiple resource capacity [7]- [9], [19], [31]. Five different cases of DL-prone situations are considered to investigate the impact of resource competitions and limited buffer capacity on FMS scheduling.…”

“…Case 5 is the third FMS example [7]- [9], [19], [31] that consists of four machines M 1 , M 2 , M 3 , and M 4 and three robots R 1 , R 2 , and R 3 . Each machine can hold two parts at the same time while the robots can move only one part at a time.…”

“…The second approach [6], [7], [31], [32], [40], [41] introduces DL control policies [8]- [11], [19] into DL-prone PNs to make them live (or DL-free) before scheduling is carried out. A live PN ensures that no DL state is reached during the system evolution.…”

“…Since the computation procedure of optimal control policies is NP-hard [7], [31], [42], these methods [7], [31] are system-specific and more suitable to a particular set of FMS problems where the capacity of the same kind of resource and lot sizes of parts are larger. References [8]- [11] and [42] have focused on how to efficiently compute siphons and design a live PN.…”

“…Also, the controllers are conservative and impose constraints on the system by considering a limited number of alternatives, which in turn affects the system's performance. Genetic search algorithms [7], [31], [32] and reachability graph methods [6], [44] have been used to search for near-optimal DL-free schedules. Another approach [27] based on AA uses Ramadge-Wonham's [45] supervisory control theory to synthesize a DL-free controller for FMS models, and A * heuristic search for scheduling.…”