Model-Based Monitoring Technologies in Large-Scale Plants

Abstract

The hot strip mill is a typical large‐scale plant consisting of numerous units, where equipment failures may lead to significant downtime and product quality deterioration. To support stable operation, monitoring and diagnostic technologies with high interpretability are increasingly required. This paper proposes a monitoring and diagnostic approach that emphasizes explainability by introducing model‐based monitoring. The proposed method focuses on specific abnormal phenomena that are difficult to interpret using conventional statistical approaches. For rolls and motors, frequency‐domain analysis using Fast Fourier Transform (FFT) monitors harmonics based on the fundamental rotational frequency to detect abnormal vibration patterns. For actuators such as the hydraulic cylinder, an Auto‐Regressive with eXogenous input (ARX) model is employed to identify time constants and damping ratios, enabling precise monitoring of response deterioration. Application results from an actual hot strip mill demonstrate that the proposed approach identifies abnormal phenomena and localizes their occurrence. The results confirm the effectiveness of model‐based monitoring in improving the interpretability of diagnostic results and supporting practical operation in large‐scale plants.