Labeling Degraded Full Operation Cycle Data with 2nd-Level Relaxation Clustering

The paper proposes a convenient unsupervised degradation data labeling method based on 2nd-level relaxation clustering. The proposed method serves as a practical auxiliary tool to help researchers segment a full operation cycle data of mechanical equipment into different pieces of healthy, degradation, or failure states. The proposed method is fully automotive and by applying 2nd-level relaxation – a structure specially designed for segmenting operation cycle data with degradation – can outperform existing clustering methods. Experiments on both simulation data and industrial pump data show the proposed method achieves accuracy matching the human expert baseline, making it a simple and effective tool.