Dynamic scheduling for flexible job shop based on MachineRank algorithm and reinforcement learning.

Area of Science:

• Operations Research
• Artificial Intelligence
• Manufacturing Systems Engineering

Background:

• The Dynamic Flexible Job Shop Scheduling Problem (DFJSP) presents significant challenges due to unpredictable events like job insertions, machine breakdowns, and processing time variations.
• Existing scheduling methods often struggle to adapt to the continuous and dynamic nature of modern manufacturing environments, especially when incorporating Automated Guided Vehicles (AGVs).

Purpose of the Study:

• To develop an intelligent scheduling system capable of dynamically adapting to disruptions in the DFJSP.
• To minimize the maximum completion time (makespan) and improve the rate of on-time completion in dynamic job shop environments.

Main Methods:

• A Dueling Double Deep Q Network (D3QN) was employed to learn optimal scheduling rules from continuous production states.
• A MachineRank (MR) algorithm was proposed to enhance solution quality, leading to seven composite scheduling rules.
• Eight general state features were defined to represent the scheduling status for the D3QN input.

Main Results:

• The D3QN demonstrated superior performance compared to various composite rules, advanced scheduling rules, and standard Q-learning agents across numerous test instances.
• The proposed dynamic scheduling trigger rules were validated for their effectiveness and rationality in real-time decision-making.
• The D3QN approach showed strong generality across different production configurations.

Conclusions:

• The D3QN-based approach provides an effective and robust solution for the DFJSP, outperforming traditional methods.
• The integration of continuous state features and advanced reinforcement learning techniques offers a promising direction for dynamic scheduling optimization.
• The study validates the practical applicability of the developed model in complex, disruptive manufacturing settings.