Electric motors are core power units in industrial, transportation, aerospace, and other fields. Early detection and prevention of motor faults are critically important for ensuring operational safety and optimizing economic efficiency. Traditional maintenance strategies rely on model-based monitoring and signal analysis, which suffer from difficulties in modeling and are susceptible to noise, etc. Deep learning methods, with their strong nonlinear mapping capabilities, powerful data processing, and adaptability, offer significant advantages. This paper reviews and analyzes the application status and challenges of electric motor fault diagnosis methods based on deep learning. In light of current research issues, such as high costs of data acquisition and labeling, poor model interpretability, and weak generalization across operating conditions, this paper proposes that future research should focus on intelligent data augmentation, explainable deep learning, model lightweighting, and multi-modal data fusion. This work provides theoretical basis and practical guidance for the further development of intelligent fault diagnosis technology for electric motors.