Selective cooperative disassembly planning based on multi-objective discrete artificial bee colony algorithm

Abstract

Disassembly sequencing has significant effects on the performance of remanufacturing and recycling of used or discarded products. Studies on disassembly sequence optimization have largely focused on sequential disassembly. However, for large or complex products sequential disassembly takes long time to complete and is rather inefficient since it removes only one part or subassembly at a time with only one operator assigned to disassemble a product. This work studies selective cooperative disassembly sequence planning (SCDSP) problem which is essential to disassemble large or complex products in an efficient way. Similar to sequential disassembly planning, SCDSP aims at finding the optimal disassembly task sequence, but is more complicated. SCDSP is a nonlinear NP-complete combinatorial optimization problem, and evolutionary algorithms can be adopted to solve it. In this paper exclusive and cooperative relationships are introduced as additional constraints besides the common precedence relationship. A novel procedure to generate feasible cooperative disassembly sequences (GFCDS) is proposed. A mathematical programming model of SCDSP is developed based on the parallel disassembly characteristics with two optimization objectives i.e. disassembly time and profit, considered. A multi-objective evolutionary algorithm (MOEA), i.e., multi-objective discrete artificial bee colony optimization (MODABC), is adopted to solve the problem to create the Pareto frontier. This approach is applied to real-world disassembly processes of two products (a small product and a medium/large one) to verify its feasibility and effectiveness. Also, the proposed method is compared with the well-known NSGA-II. For our comparative study, the nondominated solutions of the two MOEAs are compared in both cases, and two quantitative metrics, i.e., inverted generational distance (IGD) and spacing (SP), are adopted to further measure the algorithm performance. Results indicate that the set of nondominated solutions from MODABC are better for each instance tested, and the Pareto front is overall superior to that from NSGA-II. For both cases, IGD and SP are decreased by up to 81.5% and 62.2%, respectively.