Aiming at the difficult blade in daily maintenance of large-scale wind turbine, the sound signal of wind turbine blade can be detected by analyzing the non-contact fault of the blade. In this paper, a sound acquisition and transmission device is designed derived on stm32f103c8t6 single chip microcomputer. The device is arranged below the fan blade to collect the fan blade sound and transmit it to the PC through GPRS. The data is restored to the WAV format audio file on the PC. then LabVIEW software pre-processes the sound signal in the audio file and 1 / 6 octave processing to extract the acoustic characteristics of the blade, Finally, the BP neural network is trained and predicted by using the acoustic characteristics of fan blades in different states. The experimental results show that the training of BP neural network can be completed after 27 iterations, and the recognition rate of fault sound blades is more than 88%.