Research On Fault Diagnosis Of Rotating Machinery Based On Sparse Filtering And Deep Learning

Rotating machinery is widely used in modern industrial production.Once a failure occurs,it will have a serious impact on the safe production of rotating machinery and equipment.In actual production,the collected equipment fault signals are often mixed with background noise or incomplete data,etc.These problems bring serious challenges to effectively extracting the representative features of the signals.At present,although deep learning has achieved fruitful results in the field of rotating machinery fault diagnosis,the parameter selection and training of deep learning methods are still difficult,and sparse filtering only needs to adjust the output dimension by optimizing the feature distribution without selecting too many parameters,and easier to train.Therefore,this paper takes the fault detection and identification of the key components of rotating machinery-bearings and gears as the goal,and conducts research on the fault diagnosis method of rotating machinery based on sparse filtering and deep learning.The main research contents include:(1)On the basis of analyzing the research status at home and abroad,aiming at the problem of gradient disappearance and explosion in the feature extraction of time series signals by recurrent neural network(RNN),a stacked long-term and short-term memory artificial neural network(SLSTM)is constructed,and the appropriate network parameters are selected through comparative experiments.It is shown that the stacked LSTM can fully mine and memorize the feature information in the time series signal.(2)A fault diagnosis method based on sparse filtering and stacked LSTM is proposed in order to mine the features with rich fault information from the original vibration signal of noise background and improve the accuracy of diagnosis.First,convert the time-domain signal into a frequency-domain signal.Then,through a sparse filtering feature learning network,initially learn and reconstruct features from the frequency-domain signal,and perform feature dimension reduction.Finally,input the learned features to the stack LSTM network for further learning and classification of features.The effectiveness of the proposed method is verified by comparing and analyzing the data collected by the rolling bearing and the gear test bench of the firststage gearbox with the traditional feature extraction method and deep learning method.(3)In order to further improve the generalization performance of the diagnostic model and accurately diagnose faults in different data states,a rotating machinery fault diagnosis model based on feature-level information fusion with sparse filtering and stacked LSTM is proposed.A sparse filter network is constructed to extract timedomain and frequency-domain features in parallel,and the complementarity of features is increased by fusing parallel features,thereby improving the performance of fault diagnosis.The effectiveness of the proposed method is verified by the rolling bearing fault dataset.The paper concludes with a summary and outlook of the full text.