Multi-Objective Flexible Job Shop Scheduling with Tool Switching: A Hybrid NSGA-II-SA-VNS Approach

With the widespread application of small-batch, multi-variety production in the automotive equipment manufacturing industry, frequent tool switching significantly affects machining efficiency. It is essential to consider tool life when treating tool switching as a necessary constraint for the multi-objective flexible job-shop scheduling problem (MOFJSP). To address this problem, a non-dominated sorting genetic algorithm II with simulated annealing and variable neighborhood search (NSGA-II-SA-VNS) is proposed. The optimization objectives are to maximize average machine utilization and minimize makespan and tool switch times. Tool remaining useful life (RUL) is predicted based on the multi-scale convolutional neural network and bidirectional long short-term memory (MSCNN-BiLSTM) to guide tool selection during scheduling. A three-stage encoding method is employed to represent the assignment of operations, machines, and tools. A clustering-based mating restriction (CMR) strategy is proposed to balance local search (LS) and population diversity. A VNS method incorporating an SA-based acceptance criterion is developed to enhance exploration of the solution space. A case study is conducted to verify the effectiveness of the proposed algorithm by comparing it with three other algorithms. The results demonstrate the superiority of the proposed algorithm in terms of solution quality.