An Intelligent Disassembly Sequence Optimisation Framework for End-of-Life EV Batteries Using Adaptive Operator...

End-of-life (EoL) electric vehicle (EV) batteries comprise numerous interconnected components with complex topological and precedence relationships. These constraints significantly increase the difficulty of disassembly sequence planning (DSP), as feasible sequences must satisfy multiple dependency requirements. Moreover, the large number of possible disassembly alternatives creates a vast search space, making sequence optimisation susceptible to combinatorial explosion and convergence to local optima. Therefore, effective DSP requires both robust constraint-handling mechanisms to ensure sequence feasibility and efficient optimisation strategies to identify high-quality solutions. To address these challenges, this paper proposes a disassembly sequence optimisation method that integrates a hard-constraint rule base, the linear upper confidence bound (LinUCB) algorithm, and the Bees Algorithm (BA). First, a disassembly-oriented hard-constraint rule base is developed to standardise the identification of component topological relationships and precedence constraints, thereby ensuring the generation of feasible disassembly sequences. A LinUCB-based contextual adaptive operator-selection mechanism is subsequently introduced to dynamically select neighbourhood operators according to the current search state. A weighted multi-criteria evaluation function incorporating disassembly time, payment cost, and human-robot utility is integrated into the BA. Two representative EoL-EV battery case studies with different levels of structural complexity are used for validation. Across 50 independent runs, LinUCB-BA reduced the mean normalised weighted objective value by 39.01% and 28.12% relative to simplified swarm optimisation (SSO) and teaching-learning-based optimisation (TLBO), respectively, in the 27-component case, and by 7.19% and 2.33% in the 16-component case. Compared with the enhanced discrete Bees Algorithm (EDBA) ablation baseline, further reductions of 1.98% and 0.43% were achieved, together with lower run-to-run variability. These results indicate that the proposed framework is effective for the two investigated battery disassembly scenarios, while broader validation across additional battery architectures and operating conditions remains necessary.