In this study, we aim to develop an automatic identification and diagnostic system for motor vibrations based on image processing technology. This technology converts motor time-domain vibration signals, acquired from a single-axis accelerometer, into original frequency-domain signals and performs multilevel smoothing processing to generate energy trend spectral graphs. These graphs are used to reduce the computational load for image recognition and improve its accuracy. Subsequently, you only look once version seven (YOLOv7) image processing model is used for establishment and training. The established system automatically identifies mechanical vibration modes. Digital band-pass filtering technology is also used to select a single mechanical vibration mode for analysis. Next, the cumulative enhanced energy operator (CEEO) is used to demodulate the original signal, generating CEEO signals. By analyzing the CEEO spectrum and motor bearing damage characteristics, the health status of the motor is determined. By using AI in this system, the labor and long-term training costs of personnel can be reduced, ensuring the appropriate functioning of critical components.