Abstract: Considered the limitation of wavelet analysis in signal processing, a wavelet analysis method based on wavelet packet energy spectrum and Back Propagation (BP) neural network is proposed to detect the slurry quality in bellows. Ultrasonic detection method is adopted to receive the echo signal of the bellows model, and the energy in every sub-frequency band after the wavelet packet decomposition is taken as the detection feature. When the concrete slurry inside the bellows falls off, the detection features change. Finally, the features are input into the BP neural network for classification and identification. The experimental results show that this method can be used to diagnose the internal defects of bellows and provides a technical support for the non-destructive testing of bellows.