Fault Diagnosis of Rotating Machinery Using Denoising-Integrated Sparse Autoencoder Based Health State Classification

The diagnostic study on single-fault with distinguishing features based on monitoring data analysis is mature and fruitful in recent years. However, the early fault signals collected by practical monitoring systems often possess the following characteristics: 1) Fairly weak signal strength; 2) Submerged in powerful background noise; 3) Coupling of different fault data. These features not only increase the diagnostic difficulty, but also make the existing methods hardly to get the desired results. Consequently, the early compound faults diagnosis commonly in industrial systems is still a thorny and urgent problem. Therefore, in order to solve this problem and provide technical support for the practical industrial machinery fault diagnosis, a denoising-integrated sparse autoencoder (DISAE) model for early compound faults diagnosis is proposed in this paper. The innovation points of this study mainly include: 1) A feature-enhanced and denoising solution based on fault sensitivity degree (FSD) is designed, and the reconstructed diagnostic signals are acquired. 2) A disassociation framework is formulated, and the data coupling is solved. 3) A weight constraint term of SAE is constructed to improve the effectiveness and diversity of feature learning. 4) An adaptive loss function and a DISAE model is formed, and the early compound faults diagnosis is achieved. Finally, different trials and comparison results display the effectiveness and superiority of the designed DISAE based scheme.