Reinforcement Learning-Based Multiobjective Evolutionary Algorithm for Mixed-Model Multimanned Assembly Line...

Area of Science:

• Operations Research
• Industrial Engineering
• Manufacturing Systems

Background:

• Customization and rush orders create uncertain demand in assembly enterprises.
• This necessitates assembly line configurations that improve both production efficiency and robustness.
• Balancing assembly lines effectively is crucial for managing these challenges.

Purpose of the Study:

• To address the cost-oriented mixed-model, multimanned assembly line balancing problem under uncertain demand.
• To develop a robust mixed-integer linear programming model for minimizing production and penalty costs.
• To design a novel reinforcement learning-based multiobjective evolutionary algorithm (MOEA) for solving the problem.

Main Methods:

• A new robust mixed-integer linear programming model was formulated.
• A reinforcement learning-based multiobjective evolutionary algorithm (MOEA) was developed, featuring priority-based solution representation and a task-worker-sequence decoding strategy.
• The MOEA incorporates Q-learning for operator selection and a probability-adaptive strategy for operator coordination.

Main Results:

• The proposed MOEA demonstrated superior performance compared to 11 competitive MOEAs and a prior single-objective method across 269 benchmark instances.
• The algorithm effectively balances production and penalty costs while considering robustness and reducing idle time.
• Experimental results validate the model's and algorithm's effectiveness in uncertain demand environments.

Conclusions:

• The developed robust model and MOEA provide an effective solution for assembly line balancing under uncertain demand.
• The findings offer practical managerial insights for improving assembly line efficiency and robustness.
• Further research can explore limitations and extend the algorithm's applicability.