1D-DCTN: 1-D Deformable Convolutional Transformer Network for Multi-Signal Fault Diagnosis

The early detection and accurate classification of motor faults ensure the safe operation of the motor. At present, the fault diagnosis method based on deep learning has been widely used in the industrial field because of its strong feature extraction ability and accurate recognition. However, most methods only utilize a single type of signal, while neglecting the robustness of combining multiple signals. To address this issue, a novel approach called the one-dimensional deformable convolutional transformer network is proposed. The proposed model utilizes current and vibration signals to diagnose induction motors, leveraging continuous wavelet transform to capture frequency domain features. Furthermore, the integration of one-dimensional deformable convolutional layers replaces the explicit positional encoding commonly employed in traditional Transformer models. This clever substitution not only captures spatiotemporal features but also ensures the reliability of the diagnosis. The experimental results show that the proposed method has obtained an outstanding diagnostic accuracy of 99.83%. Moreover, compared to traditional single-signal fault diagnosis methods, the proposed model performs enhanced robustness and accuracy.