Curriculum-Based Soft Actor-Critic for Multi-Section R2R Tension Control

A research team from multiple institutions has published a paper detailing a novel AI-driven approach to a persistent industrial challenge: precise tension control in roll-to-roll (R2R) manufacturing. The paper, "Curriculum-Based Soft Actor-Critic for Multi-Section R2R Tension Control," introduces a controller built on the Soft Actor-Critic (SAC) reinforcement learning algorithm. The core innovation is a curriculum-based training strategy, where the AI policy is trained in three distinct phases, each with progressively wider reference tension ranges. This method starts with a narrow 27 to 33 Newton range and systematically expands to handle the system's full operational envelope of 20 to 40 N. The goal is to create a controller that generalizes effectively across both nominal operating conditions and significant, unpredictable disturbances.

The technical results demonstrate the controller's robustness. When tested on a standard three-section R2R benchmark system, the single, universally trained policy successfully managed accurate tension tracking during normal operation. Crucially, it also handled extreme disturbances—including massive step changes from 20 N to 40 N—without requiring any scenario-specific retuning or adjustments. This performance indicates that the curriculum-trained SAC controller is a viable and practical alternative to traditional, complex model-based control systems, especially in environments where system parameters vary and process uncertainty is high. The work signifies a meaningful step toward deploying more adaptive, learning-based AI agents in real-world industrial control applications, potentially reducing engineering overhead and improving resilience in continuous manufacturing processes like web handling for films, textiles, and electronics.