An event-driven hybrid rescheduling approach for integrated process planning and scheduling considering stochastic...

Area of Science:

• Operations Research
• Manufacturing Systems Engineering
• Artificial Intelligence

Background:

• Static integrated process planning and scheduling (IPPS) is theoretically established but lacks adaptability in dynamic manufacturing.
• Unpredictable environments require dynamic solutions that account for factors like stochastic rework.

Purpose of the Study:

• To propose and address the dynamic integrated process planning and scheduling with stochastic rework (IPPS-SR) problem.
• To optimize both product quality and scheduling performance under uncertainty and reprocessing requirements.

Main Methods:

• Formulated a mathematical optimization model for IPPS-SR to minimize makespan and schedule instability.
• Developed an event-driven hybrid rescheduling approach integrating right-shift scheduling with a multi-objective reinforcement learning-guided adaptive large neighborhood search (MORL-ALNS).
• Designed problem-specific destroy and repair operators to enhance MORL-ALNS search efficacy.

Main Results:

• The MORL-ALNS approach effectively generates high-quality rescheduling schemes for IPPS-SR.
• The RL-guided mechanism improved the number of non-dominated solutions by over 80% compared to baseline ALNS.
• MORL-ALNS demonstrated superior performance, achieving better Hypervolume (HV) and Inverted Generational Distance (IGD) values against other multi-objective algorithms.

Conclusions:

• The proposed hybrid rescheduling approach offers an effective trade-off between computational efficiency and solution quality for dynamic manufacturing environments.
• The integration of reinforcement learning significantly enhances the search for optimal solutions in complex rescheduling problems.
• The MORL-ALNS framework provides a robust and superior method for addressing the challenges of dynamic IPPS-SR.