Abstract: To address the problems of limited fault diagnosis accuracy of single-signal (current or vibration) methods and difficulty in identifying multiple coexisting faults under complex operating conditions of underground coal mine motors, a motor fault diagnosis method based on the combination of electrical and vibration signals was proposed, leveraging the electromechanical coupling characteristics and the complementarity of multi-sensor information. Fault information was captured from the motor current and vibration signals in the time domain, frequency domain, and time–frequency domain, respectively. The features were fused in the channel dimension to generate feature color images containing multi-domain information, thereby enriching the fault characterization. A Dual-Channel Residual Network (DCResNet) model embedded with an Improved Convolutional Block Attention Module (ICBAM), namely ICBAM-DCResNet, was constructed. Through multiple residual blocks and the attention mechanism of ICBAM, deep features of image samples were extracted. Finally, feature fusion and classification were performed to achieve fault diagnosis based on the combination of electrical and vibration signals. The comparative experimental results showed that multi-domain fusion achieved higher diagnosis accuracy than single-domain analysis, and the ICBAM-DCResNet model outperformed the Residual Network (ResNet) model, demonstrating stronger feature extraction capability for the signal samples. The experiment results on the public dataset demonstrated that the proposed motor fault diagnosis method based on the combination of electrical and vibration signals achieved an accuracy of 99.8%, with good identification performance for rotor and bearing faults and strong generalization ability.