Optimizing State Monitoring With Domain Degradation Knowledge

Monitoring the degradation of industrial equipment is vital for maintaining efficient production. Existing deep learning-based monitoring techniques often focus on isolated target characteristics, failing to capture the device’s condition throughout its degradation process comprehensively. We propose a novel approach to constructing a unified feature space incorporating degradation-based features from multiple degradation characteristics. The proposed method enhances the predictive accuracy of degradation parameters, improves model interpretability, and minimizes overfitting by avoiding reliance on unrealistic characteristic patterns. We introduce a new model architecture, Industrial Health Index Extraction, designed to implement this approach effectively. Our methodology demonstrates state-of-the-art performance on self-supervised and supervised tasks using NASA’s CMAPSS and milling datasets.