A novel blade crack detection method based on diffusion model with acoustic-vibration fusion

Compressors are now widely used in industry and engineering, and blades are one of the most important components in compressors. The performance of the blades directly affects the operating condition and life of the compressor. Currently, the mainstream method for diagnosing and classifying blade faults is based on vibration signal diagnosis. However, traditional methods are limited by the large influence of noise on vibration signals and the singularity of features, and their accuracy and efficiency are relatively low. In addition, as a mainstream diagnostic method, fault diagnosis based on neural networks also suffers from limitations in network structure and data volume, which reduces the generalization of diagnostic methods. Therefore, this paper proposes a new blade fault diagnosis network based on the diffusion model. Specifically, to improve the integrity of the features used for diagnosis, this paper first proposes a learnable weight fusion module and applies it to the fusion process of sound and vibration signals. Secondly, the diffusion model is introduced to generate normal blade signals under corresponding operating conditions when fused features of blades with faults are input. Finally, after obtaining the fused features of normal blades under corresponding operating conditions, the input-output feature difference of the diffusion model is used as the input of the classification network to achieve blade fault diagnosis. In experimental tests, the method proposed in this paper outperforms the current mainstream blade fault diagnosis methods on actual blade fault data. In addition, comparative experiments and ablation experiments also prove the effectiveness of the proposed method.