Research On Fault Diagnosis Of Rotor Based On Deep Learning

Due to the high-speed rotation of the equipment and the huge kinetic energy generated by the large-scale rotating machinery,the failure of the rotor components poses a great hidden danger to the safe operation of the equipment.In the fault diagnosis of rotating machinery,rotor imbalance,misalignment,and friction between moving and static parts are several common fault forms.With the development of modern science and technology,industrial manufacturing is gradually moving towards an automated,intelligent,and integrated development model,and the field of mechanical fault diagnosis has begun to emerge in the field of machine learning-based intelligent diagnosis methods.In terms of fault diagnosis,deep learning technology shows strong feature recognition and classification capabilities.In the fault diagnosis of the rotor system,due to the diversity and nonlinearity of the fault characteristic data,this paper focuses on these problems,takes the rotor as the research object,analyzes its fault mechanism and fault characteristics,and uses the deep learning theory to test the test bench.The data is diagnosed and analyzed to realize the transformation of experimental research results to engineering applications.This article first introduces the characteristics and mechanisms of several common types of rotor failures,and provides a theoretical basis for subsequent failure diagnosis.Then the vibration data is collected through the rotor failure simulation test bench of the Mechanical Engineering Laboratory of Three Gorges University,and the use of one-dimensional depth is verified through comparative experiments.The convolutional network(1D-DCNN)model can achieve better accuracy and accuracy in the fault diagnosis of the vibration signal of the rotor.Later,based on the advantages of the existing neural network in the field of image recognition,a Gram angle field(GAF)method is introduced to convert the rotor vibration signal from one-dimensional to two-dimensional.It was discovered that the GAF method was used to convert the vibration signal and feature extraction,and then combined the advantages of machine vision with deep learning.The existing CNN was used to optimize and diagnose the rotor fault.It can get better results than 1D-DCNN.At the same time,it is verified Using the features of 4-channel data fusion as input can effectively improve the training speed and recognition rate,and the recognition rate reaches 95.9%±0.2%.Because the GAF-based feature extraction and fusion method can effectively improve the test effect,the GAF is further combined with other neural networks in deep learning and used SDAE’s advantages in feature extraction to introduce SDAE to the feature extraction of rotor fault vibration data to verify that the GAF-SDAE-CNN network model has reached the recognition rate on the rotor vibration fault data set.96.4%±0.2%.Finally,the features extracted from the last hidden layer of GAF-CNN and GAF-SDAE-CNN,two diagnostic models used in this paper,are visualized by using data T-SNE technology,which verifies that the enhancement of feature expression ability is beneficial to improve the accuracy of fault diagnosis.