Manufacturing control systems autonomously manage AI models for safer production
Autonomous Model Lifecycle Management for Digital Twin-Based Manufacturing Control
Machine Learning
Summary
Machines in car factories need to keep working well even as conditions change, like raw materials or equipment wearing out. The authors created a system that acts like a digital twin—a computer model that mirrors the real process—and uses it to train smart control policies. These policies are carefully tested, and only the best ones that operators trust get used to run the factory. This approach helped make manufacturing more stable and safe in several factories without any accidents.
What this means in practice
- •For manufacturing engineers: Run adaptive control systems that autonomously retrain and select AI models for safer, more stable production lines in automotive factories.
- •For industrial automation teams: Manage and orchestrate digital twin models and reinforcement learning controllers to maintain continuous process stability under changing conditions.
Authors
Zhengyang, Gu, Thomas Cook, Fredaljohn Rohrbaugh, Joseph E. Hernandez, Chris Couch
Abstract
Manufacturing AI systems must autonomously adapt to continuous distributional shift from raw-material variability, ambient changes, and equipment aging, under strict safeguard and operator-trust requirements where model failures risk physical damage. This paper presents a closed-loop Cyber-Physical System (CPS) for autonomous model lifecycle management in automotive manufacturing, deployed since 2023. The system manages product-specialized model pairs: a sequence-to-sequence physics model (LPP) serving as a digital twin, and a deep Reinforcement Learning (RL) control policy (LCP) trained against it. Per retraining cycle, multiple model variants spanning architecture families and RL algorithms compete; only the best-scoring candidate advances. A Conductor orchestrator autonomously manages plant-wide model inventories with dependency-aware retraining and Proportional-Integral-Derivative (PID) fallback. Reflecting the principle of Human-Centric Intelligence, the LCP composite score embeds an operator-trust gate penalizing policies deviating from established practice; without it, 23% of policies are rejected by operators despite passing accuracy thresholds. Across multiple facilities, LCP-controlled processes achieve process stability improvements of 28-45% over uncontrolled baselines with zero safety incidents.