Research on motor fault recognition based on multi-sensor fusion

With the rapid development of new energy and power technology, motors are widely used in daily life. The fault recognition of motor can effectively reduce the economic loss and the threat to personnel safety. In recent years, motor fault detection based on deep learning has made remarkable achievements. But these methods only use one modality, such as voltage or current signals. However, multi-modal information fusion can make full use of the complementarity between different modes to effectively improve performance. To this end, this paper proposes a new deep network to leverage multi-modal fusion for motor fault recognition. Specifically, we use different sensors to simultaneously collect the sequence signals, including voltage, current and vibration signals. To explore the relationship of intra-modality, we design a Transformer-based deep model by exploiting the multi-head attention mechanism. To mine the inter-modality relationships, we use the cross-attention mechanism. All the experimental results show that the performance of the proposed deep model is better than other deep sequence models in motor fault detection.