Generative Model Predictive Control in Manufacturing Processes: A Review

TL;DR

This paper addresses control of dynamic, uncertain manufacturing processes by examining how generative ML can augment Model Predictive Control (MPC). It argues that generative models—including GANs, normalizing flows, VAEs, diffusion models, and LLMs—enable distribution learning, latent-factor modeling, and multi-modal trajectory generation that can be integrated into MPC across predictive modeling, state estimation, and optimization. The authors provide a taxonomy of methods, summarize integration patterns (predictive surrogates, soft sensors, and learned controllers), and identify key gaps—domain-specific knowledge integration, robustness to distribution shifts, and real-time feasibility—along with future directions exemplified by a multi-robot manufacturing case. The work highlights the potential of generative ML to move MPC from a reactive, model-driven framework toward a proactive, uncertainty-aware paradigm for next-generation manufacturing systems.

Abstract

Manufacturing processes are inherently dynamic and uncertain, with varying parameters and nonlinear behaviors, making robust control essential for maintaining quality and reliability. Traditional control methods often fail under these conditions due to their reactive nature. Model Predictive Control (MPC) has emerged as a more advanced framework, leveraging process models to predict future states and optimize control actions. However, MPC relies on simplified models that often fail to capture complex dynamics, and it struggles with accurate state estimation and handling the propagation of uncertainty in manufacturing environments. Machine learning (ML) has been introduced to enhance MPC by modeling nonlinear dynamics and learning latent representations that support predictive modeling, state estimation, and optimization. Yet existing ML-driven MPC approaches remain deterministic and correlation-focused, motivating the exploration of generative. Generative ML offers new opportunities by learning data distributions, capturing hidden patterns, and inherently managing uncertainty, thereby complementing MPC. This review highlights five representative methods and examines how each has been integrated into MPC components, including predictive modeling, state estimation, and optimization. By synthesizing these cases, we outline the common ways generative ML can systematically enhance MPC and provide a framework for understanding its potential in diverse manufacturing processes. We identify key research gaps, propose future directions, and use a representative case to illustrate how generative ML-driven MPC can extend broadly across manufacturing. Taken together, this review positions generative ML not as an incremental add-on but as a transformative approach to reshape predictive control for next-generation manufacturing systems.

Figures (8)

• Figure 1: Block diagram of a control loop with MPC, modified from Schwenzer21. In the MPC block, a general optimization loop produces a single control input for the current time step, though the predicted trajectory may extend several time steps into the future.
• Figure 2: Reconstructed framework of the ML-driven MPC based on figures from Knaak et al.knaak2021improving. This illustration combines elements from their original figures to highlight how ML components operate within the overall MPC framework: a Random Forest (RF, ML 1) serves as a surrogate predictive model for layer-to-layer quality dynamics (predictive modeling), a Convolutional Neural Network (CNN, ML 2) estimates surface roughness from images (state estimation), and reinforcement learning (RL, ML 3) replaces the traditional solver to optimize process parameters (optimization).
• Figure 3: Integration of Machine Learning into Model Predictive Control: Representative Studies in Manufacturing
• Figure 4: Taxonomy of generative ML methods adapted from Goodfellow et al. goodfellow2016deep. The taxonomy distinguishes between explicit and implicit density methods based on whether the data distribution is modeled through an explicit likelihood function. Explicit methods include tractable approaches and approximate approaches, which are further divided into variational and Markov chain-based methods. Implicit methods, by contrast, do not define a likelihood function explicitly but instead learn to generate samples directly. The bottom layer highlights the key generative ML methods reviewed in this paper, indicating where each method is positioned within the taxonomy.
• Figure 5: Reviewed Studies on the Integration of GANs, Normalizing Flows, and VAEs with MPC