Dynamic Contrast Analyzer for Generating Health Indicators in Machinery Monitoring Under Time-varying Speed Conditions

Machinery health monitoring under time-varying speed conditions is an ongoing challenge, where the significant variations in rotational speed frequently result in false alarms and missed defect detections. In this paper, a simple method, i.e., dynamic contrast analyzer (DC-Analyzer), is proposed to generate health indicators for machinery self-adaptive monitoring. Contrastive learning is used to adjust the representational features of vibration signals with various speeds, which facilitates the self-extraction of manifold trend over speeds. A minor network is proposed to perform feature regression between rotational speeds and vibrational representations for condition alignment. The final health indicators of rotating machinery can be obtained by calculating the feature similarity of real-time vibrational representations and corresponding approximate features from the regressor. The effectiveness of DC-Analyzer is verified on a rolling bearing test rig. The results show that the proposed method outperforms those representative approaches in health indicator generating, which provides more potential for the issue of dynamic and time-varying conditions.