Research On The Application Of Improved Convolutional Neural Network In Equipment Fault Diagnosis

Rotating machinery is the focus of mechanical equipment condition monitoring and fault diagnosis,rolling bearing is one of the vulnerable parts of mechanical equipment,it is also one of the most commonly used parts.The fault diagnosis of mechanical equipment has been receiving increasing attention.In the current field of fault diagnosis,the vibration signal of the equipment can be analyzed to understand the operating state information of mechanical equipment.It is of great significance for fault diagnosis to collect enough signal samples and reflect the state signal of mechanical equipment operation objectively.At present,there are two main directions in the fault diagnosis method of mechanical equipment.One is the fault diagnosis method based on the traditional model.The vibration signal of mechanical equipment is collected and analyzed in time and frequency domain.However,it is mostly difficult to extract fault features under the interference of complex factors such as strong noise and variable loads.The other is the fault diagnosis method based on deep learning.The sample signal is preprocessed and input to the neural network for training classification,which can automatically complete feature extraction and fault pattern recognition.With the increasing complexity of mechanical equipment manufacturing processes,traditional shallow neural networks can no longer meet the diagnosis requirements,and the improvement of fault diagnosis efficiency is limited.This proposed architecture significantly improves and optimizes the shallow convolutional neural networks.The main contents are summarized as follows.In this paper,an improved convolutional neural network is proposed to solve the problem of low recognition accuracy of conventional convolutional neural network.For the MaFaulDa bearing database,the convolution group of the identity map is embedded in the two convolution layers.The computation of the network model is reduced and the capability of feature extraction is enhanced.The Adam optimizer has been improved and the power exponential learning rate was used to control the iteration direction and step size.The learning rate of the previous stage and the gradient relationship between the previous stage and the current stage are selected for adaptive adjustment.The 5-fold cross-validation is used to compare the classification accuracy of different methods.For the Case Western Reserve University database,a new data processing method is introduced.The original vibration time domain signal data of the rolling bearing is converted into a two-dimensional matrix.The variable convolution kernel is considered for feature extraction and mapping,the mode of the traditional network fixed convolution kernel is changed,and the high-precision recognition of the vibration signal of the rolling bearing is realized.The improvement of the general convolutional neural network can improve the training accuracy and stability of the model.In the detection of rolling bearings,the coexistence of many faults is common.Considering the compound faults of bearings and the time-series of bearing vibration data,an improved ResNet-LSTM compound fault diagnosis model of locomotive bearings is proposed.The two-dimensional signal of bearing vibration is input into the residual network and the local feature is extracted by embedding a residual layer.In addition,the bearing feature information is loaded into a long-term memory unit and the gate mechanism is introduced to extract the global features of the time-series data.Finally,the probability distribution of predicted values is obtained by softmax function to realize multiple fault classification.Experiments show that the proposed fault diagnosis methods have good diagnostic effects on different experimental data,including single bearing faults and compound faults,which have certain practical application value.