Scheduling flexible manufacturing cell with no-idle flow-lines and job-shop via Q-learning-based genetic algorithm

Authors：Cheng, Lixin; Tang, Qiuhua^*; Zhang, Liping; Yu, Chunlong

Source：COMPUTERS & INDUSTRIAL ENGINEERING, 2022, 169: 108293.

DOI：10.1016/j.cie.2022.108293

• Flexible manufacturing cell
• Cell scheduling
• Flow production
• Job production
• Genetic algorithm
• Q-learning

Owing to enormous varieties in product quantities and types in customer orders, two production methods, flow production and job production, should be simultaneously carried out in the flexible manufacturing cell so that standardized products in high volume and individualized products in low volume can be compatibly manufactured. Thus, this work schedules such a realistic manufacturing cell with these two production methods. A mixed integer linear programming model is developed to determine cell formation and machine assignment, and particularly to locally sequence standardized products within each product family and globally schedule all individualized products and product families. Meanwhile, a Q-learning-based genetic algorithm (Q-GA) is proposed to solve large-scaled cases effectively and efficiently. In Q-GA, a rule-based initialization improves the quality of initial solutions; a forward and no-idle backward decoding mechanism satisfies the job availability, machine availability and no-idle time constraints; a Q-learning-based approach adaptively controls the crossover/mutation rates to lower relative percentage deviation. Experimental results demonstrate that for the flexible manufacturing cell scheduling problem with two production methods, the proposed Q-GA obtains the lower bounds of small-scaled cases and outperforms the comparison algorithms for large-scaled cases in the solution effectiveness and robustness.