Research On Fault Diagnosis Of Rotating Machinery Based On Compressed Sensing And Deep Learning

Rotating machinery is the core equipment in modern industry.Because of its complex structure and changeable operation conditions,it is very difficult to monitor and diagnose the faults of its key components.In recent years,intelligent diagnosis technology based on deep learning algorithm has developed rapidly,and has been developed in the field of rotating machinery fault.Based on the theory of compressed sensing and deep learning,two new methods of fault feature extraction and diagnosis are proposed in this paper.Firstly,the application status and prospect of traditional compressed sensing in the field of vibration signal analysis of rotating machinery are studied.Combining its theoretical knowledge with fault signal analysis of rotating machinery,and applying some algorithms of sparse representation to fault signal feature extraction to reduce the feature dimension of the original vibration signal,greatly reducing the complexity of fault classification.Secondly,aiming at the problem that there are many parameters in deep belief network model training and it is difficult to adjust parameters when it is applied to fault feature extraction and diagnosis of rotating machinery,an iterative error method based on loss threshold is proposed to prevent over fitting of training and optimize the model training process.On the premise of high accuracy,the adaptive setting of iteration times is realized,which reduces the difficulty of setting and adjusting the parameters of deep belief network.When it is applied to the condition monitoring of rotating machinery,the accuracy and efficiency of fault identification are also improved.Then,aiming at the problem that the vibration signal of gear fault caused by the complex working condition and structure is difficult to be accurately classified by the traditional feature extraction and diagnosis method,a gear fault diagnosis algorithm combining wavelet packet energy entropy and multi-scale arrangement entropy and using improved depth belief network is proposed.Firstly,the vibration data of various fault types under multiple working conditions are collected,then the wavelet packet energy entropy and multi-scale permutation entropy distribution are calculated to form the combined feature matrix,and then the improved threshold adaptive deep belief network is used to further extract the fault signal features and classify them.When the method is used to diagnose the gear vibration data of multi-platform and multi working condition,a high and stable diagnostic accuracy is obtained,which verifies the feasibility of the method.Finally,aiming at the high dimension of rolling bearing vibration data,a neural network fault diagnosis method of sparse auto-encoder based on compressed sensing and wavelet packet energy entropy is proposed.Firstly,the low dimension observation matrix of the original vibration signal is extracted by using the compressed sensing algorithm,and then the wavelet packet energy entropy distribution is obtained to form the feature matrix,which is used as the training sample and test sample of the sparse auto-encoder neural network.Through this method,the bearing vibration data collected in the laboratory and published in Western Reserve University are respectively diagnosed,and it is proved that it can be used in the bearing fault classification It can get better diagnosis results in a short time,which verifies the feasibility of the method.