Abstract:Traditional shallow model is adopted in most fault diagnosis methods for rotorshaft systems, but it is difficult to deal with a large sample.In order to solve this problem, a fault diagnosis method based on the improved Stacked Denoising Auto Encoder (SDAE) depth model was proposed, and it was used to diagnose the typical faults of the rotorshaft system.Using a mechanical fault comprehensive simulation platform, combined with the signal acquisition system developed based on LabVIEW, the 10 types of single fault signal and 7 types of composite failure signal of the rotorshaft system were simulated and collected. Dropout mechanism was introduced to improve the performance of SDAE model, and Softmax classifier was combined for network training and diagnosis. Compared with traditional BP network, automatic encoder (AE), SDAE without Dropout mechanism and convolutional neural network (CNN), the results show that the improved SDAE method has the highest correct identification rate for rotorshaft system faults, especially the diagnosis effect to composite faults is better than the other models, which fully verifies the superiority of the improved SDAE depth model