The standardization of commodities makes the price competition fiercer, and thediversification of product demands brings about unprecedented challenges to manufacturingindustry. Developed gradually in such industrial environment, mixed-model assembly line is aspecial flow-line production system where several models are produced on the line with littlefacility changes, low production cost and fast response speed. The diversity of automotiveproduct demands necessitates mixed-model assembly line to produce different productsconcurrently efficiently and competitively. Therefore, mixed-model assembly line balancing hasbecome the key point in manufacturing development.In this work, mixed-model assembly line balancing problem with sequence-dependent taskswas considered. According to the characteristics of sequence-dependent relations, this problemwas tackled with the combination of traditional genetic algorithm and heuristic factors.The characteristics of mixed-model assembly line balancing problem andsequence-dependent relations were analyzed respectively, and the integration of mixed-modelassembly line balancing problem and sequence-dependent relation provided a theoretical basisfor practical production.The mathematical model of mixed-model assembly line balancing problem withsequence-dependent tasks was formulated and the traditional genetic algorithm was improvedfrom three aspects. First, sequence-dependent connections and precedence relations among taskswere both integrated into combined precedence graph, transforming mixed-model assembly linebalancing into single-model version. Second, the process of initial population was complementedby introducing three heuristic factors: processing times, the number of immediate successors andthe number of updated tasks. Third, considering sequence-dependent tasks, novel logic stringswere designed to ensure the feasibility of chromosomes during crossover and mutationoperations.Numerical experiments based on mixed-model assembly line balancing problem withsequence-dependent tasks were studied with hybrid genetic algorithm. The solutions haveverified that the whole population after heuristic initialization is of better quality, the bestsolutions converge to near-optimality (even optimality) with less computational efforts, and mostcases outperform the current best solutions. The relevant experiments have demonstrated that theproposed hybrid genetic algorithm functions effectively and efficiently. |