In order to detect the vibration motor of mobile phone, neural network was introduced to detect the waveform of the motor. LabVIEW was used to collect the original signal when the motor was rotating, the original signal was decomposed by using twolayer wavelet packet, the decomposed lowfrequency signal was reconstructed, the original signal minus the reconstructed signal image was intercepted for preprocessing, and the preprocessed image was served as the data set.The model was trained by Tensor Flow. The five types of bad brush, abnormal waveform, waveform drop, bad magnetic field and good product were classified, and the accuracy rate of the improved convolutional network model had reached 98.76%. The results show that the improved neural network has better diagnosis effect and can improve the accuracy of fault diagnosis.