A Curvature-Guided Shape DTW for Anomaly Detection of Industrial Robot Joints Under Variable Operating Conditions

As a core component of intelligent manufacturing equipment, it is particularly important to detect the abnormality of industrial robot. However, the variable motion conditions of robot joints lead to strong time-varying and discontinuous monitoring signals, which pose a challenge for anomaly detection. To address this, this study proposes an anomaly detection method based on current time-frequency ridge alignment. Firstly, the current time-frequency ridge is preprocessed based on peak-frequency normalization, which effectively mitigates the speed variation and ridge extraction error interference. Subsequently, an improved ridge alignment algorithm named curvature-guided shape dynamic time warping (CSDTW) is proposed to further eliminate the impact of variable operating conditions. This algorithm innovatively integrates local curvature into the shape DTW, this advancement solves the overfitting and feature destruction problems of conventional DTW. Finally, the aligned ridges are transported to the convolutional neural network (CNN) for discriminative feature learning and fault classification. Experimental validation on an industrial robot joint dataset demonstrates that CSDTW has the highest ridge alignment accuracy under variable operating conditions compared to the traditional DTW method, improving the anomaly detection accuracy of the CNN for multiple fault types. The method provides a reliable solution for detecting anomalies in industrial robots under variable operating conditions.