A Generative DRL Framework for Adaptive Quality Control in Industrial Cyber-Physical Production Systems

The manufacturing sector faces escalating pressures to enhance production efficiency, product yield, and energy utilization, making Intelligent Quality Inspection (IQI) a cornerstone of product reliability. To address this need, we propose a data-driven decision-making framework for IQI, grounded in generative artificial intelligence and deep reinforcement learning. This study focuses on the component assembly process, a critical stage in manufacturing. We first develop a generative AI-driven daptive framework that constructs a high-fidelity assembly line. This environment realistically incorporates stochastic factors such as component defective rates, inspection mechanisms, and handling strategies for non-conforming products. To provide a rigorous foundation for algorithmic optimization, we formally model this dynamic ecosystem as a Markov Decision Process (MDP), defining its state space and probabilistic transition rules. We then introduce the integrated Rainbow Deep Q-Network (Rainbow DQN) algorithm to solve the resulting multi-step sequential decision-making problem. Our approach leverages the synergistic advantages of multiple DQN extensions to master dynamic quality control policies. Extensive experimental evaluations, including ablation studies and comparisons against benchmark algorithms such as traditional Monte Carlo methods, confirm the superiority of our proposed framework. The results demonstrate significant advantages in convergence speed, decision-making accuracy, and computational efficiency, underscoring its potential to advance the capabilities of intelligent quality inspection in modern industrial cyber-physical systems.