Fault Diagnosis Of Rotating Machinery Based On Neural Network

Rotating machinery is an important system in industry production.With the increasing requirements for security and reliability,fault diagnosis of rotating machinery becomes extremely important.However,due to the complicated and varying working conditions,representation difficulties of fault types and magnitudes by external signal,fault diagnosis in rotating machinery is relatively more difficult.Motivated by the above mentioned problems,this thesis focuses on the fault diagnosis of rotating machinery based on neural network with advantages in signal processing and classification.The main contents are as follows:First,based on the analysis of past neural network based mechanical fault diagnosis methods and general structure and characteristics of rotating machinery,an innovative algorithm based on sensitive feature selection and recursive least square back propagation(RLS-BP)neural network is proposed.Accurate representation of fault information can be solved using this diagnostic method by selecting features sensitive to classification from both time domain and frequency domain characteristic parameters of multi-channel signals.The RLS-BP neural network algorithm provides an effective approach for feature sample classification with its dynamic adaptation and strong pattern identification abilities.Second,in view of the shortcomings of conventional neural networks with no deep information mining capability,a recurrent convolutional neural network(R-CNN)diagnostic model based on deep learning is proposed.The convolutional neural network(CNN)is used to analyze the internal relationship of time-domain sampled data to obtain the main features,and the recurrent neural network(RNN)is used to summarize and compare the feature extraction results of multiple adjacent time periods to obtain deeper information representation.Compared with the sensitive feature selection method above,the R-CNN based fault diagnosis method can achieve high classification accuracy and avoid tedious steps such as feature extraction,selection and fusion.Finally,the proposed methods are tested on data sets of different mechanical components of rotating machinery to verify the effectiveness.The experimental results show that strong feature extraction ability can be observed from both methods which can effectively improve the fault identification rate and solve the diagnosis problem of multi-type faults under multiple operating conditions.