An Online Recognition Method for Identifying the Motion Characteristics of AC Contactor Based on SDP Image Features and EfficientNetV2

The continuous on-and-off over an extended period can lead a contactor to gradually enter a phase of inferior operation. When its status is monitored in real time, potential faults can be promptly detected, thereby enhancing the stability of the control system. In light of this, it is proposed to identify the characteristics of motion based on the image features of vibration-acoustic signals fusion and lightweight deep learning model. Firstly, considering the impact of the electrical wear state of the contacts on identification results, in order to eliminate this interfering factor, this paper simulates the contacts colliding process, which reveals the mechanisms of vibration and acoustic signals generation. Also, simulation integrates real experiments to investigate the impact of contact surface roughness on signal frequency characteristics. The measured signals are then decomposed, and the appended frequency components induced by the rough contact surface are eliminated through spectral analysis of the modes. The modes that are sensitive to movement characteristics are retained through correlation coefficient. This effectively mitigates the interference of irrelevant factors. Secondly, the selected effective modes of vibration signal and acoustic signal are fused into symmetrized dot pattern (SDP) to visualize the subtle changes in signals in different states. Moreover, the joint analysis of two signals can enrich the information on the characteristics of motion. Finally, EfficientNetV2-S model is constructed to capture multi-scale image features using few computational resources at various levels. The results show that EfficientNetV2-S achieves a 14% higher accuracy rate compared to RestNet50, while exhibiting improved operational efficiency.