Reproduce “Generative Model Predictive Control in Manufacturing Processes: A Review” on OpenResearch

About this paper

Suk Ki Lee, Ronnie F. P. Stone, Max Gao, Wenlong Zhang, Zhenghui Sha, Hyunwoong Ko

A comprehensive review analyzes the integration of Generative Machine Learning (ML) methods into Model Predictive Control (MPC) frameworks for manufacturing processes, synthesizing how five key generative ML techniques enhance predictive modeling, state estimation, and optimization. The review identifies current limitations and outlines future research directions for robust, real-time industrial deployment.

Key insights

• Generative ML fundamentally differs from conventional ML by learning the underlying data distributions, which allows it to capture hidden patterns, generate diverse plausible outcomes, and inherently manage uncertainty probabilistically.
• Generative models systematically enhance MPC by serving as high-fidelity probabilistic surrogates for complex system dynamics, enabling more accurate and robust state estimation through latent variable inference, and improving optimization via multi-solution exploration and constraint-aware generation.
• Despite its transformative potential, current generative ML-driven MPC applications face critical challenges including limited direct application to manufacturing, inadequate robustness to distribution shifts, and computational latency that conflicts with real-time control requirements.

Summary by alphaXiv.