A robust hybrid metaheuristic framework for two-sided assembly line balancing with adaptive genetic search and post-optimization workload balancing

Abstract The Two-Sided Assembly Line Balancing Problem (TSALBP) is a challenging combinatorial optimization problem. Although genetic algorithms (GAs) have been widely applied to TSALBP, many existing approaches suffer from premature convergence, limited scalability, and insufficient statistical validation. This paper proposes an Enhanced Genetic Algorithm (EGA) for TSALBP-1. The approach combines heuristic population initialization based on ranked positional weight and longest-path rules, adaptive crossover and mutation operators, earliest–latest station feasibility bounds, and a single-tier station assignment strategy. To improve workload distribution, a hybrid Tabu Search–Variable Neighborhood Search (TS/VNS) procedure is applied as a post-optimization stage. In addition, an AI-generated heuristic was developed, and its performance was compared with the proposed EGA and conventional algorithms. Performance was evaluated using solution-quality metrics and rigorous statistical analyses, including Friedman ranking, Wilcoxon signed-rank tests, and Dolan–Moré performance profiles. Experimental results show that the EGA achieved an average gap of only 0.34% from the lower bound and achieved the highest overall ranking among all competing heuristics. Five independent runs across all 42 benchmark configurations showed low run-to-run variability, supporting the robustness of the proposed EGA. Besides, the proposed approach consistently outperformed the AI-generated heuristic, especially for medium- and large-scale benchmark instances. Finally, the TS/VNS post-optimization phase improved workload balance significantly, reducing idle-time variance by up to 99.7% in large-scale instances. The results demonstrate that the proposed adaptive hybrid framework provides an effective and scalable solution for TSALBP while achieving superior solution quality, robustness, and workload distribution. While hybrid initialization, adaptive operators, and TS/VNS refinement are individually established techniques, the novelty of this work lies in their coordinated integration within a TSALBP-specific framework, their controlled component-wise evaluation, and the rigorous statistical validation of the overall framework. The main contributions of this work are the development of an enhanced adaptive genetic algorithm with heuristic-based population initialization and dynamic search operators, together with a hybrid TS/VNS post-optimization framework for smoothing workload without increasing the number of stations. In addition, the proposed approach is comprehensively validated through rigorous statistical analyses and benchmarked against both state-of-the-art algorithms and an AI-generated heuristic to assess the potential of large language models for generating heuristic solution procedures for TSALBP.