The poorer effect of vibration signal collection from robotic arm motors affects the time-frequency character analysis process and leads to lower fault diagnosis effectiveness and accuracy. Therefore， a time-frequency scale diagnosis method for robotic arm motor faults based on deep learning and laser Doppler vibration measurement technology is proposed. Using laser Doppler vibration measurement technology and wavelet threshold de-noising algorithm， we develop a mechanical arm motor vibration signal collection system to obtain and reconstruct fault signals； we extract the time-domain， frequency-domain and other scale features of the motor vibration signal， introduce an artificial neural network to establish a fault diagnosis model with learning ability， input the extracted time-domain， frequency-domain and other scale features of the mechanical arm motor fault into the diagnosis model， and output the classification diagnosis results， which can complete the time-frequency scale diagnosis of the mechanical arm motor fault. The experimental results show that， when this method for motor fault diagnosis is adopted， the deviation between the detection results and the actual motor fault type is small， and the diagnostic results are good with higher accuracy.