Research On Intelligent Fault Diagnosis Of Rotating Machinery Based On Capsule Network

Rotating machinery,as the most extensively used component in the industrial area,plays an important role in ensuring the safety and stability of contemporary businesses.The shallow approach based on feature extraction has been unable to fulfill the requirements of industrial big data due to the fast rise of data acquired by sensors.The residual neural network,which is based on vibration signals from spinning equipment,is at the heart of the deep learning algorithm,ensuring the information integrity of the capsule neural network and improving the interpretability of the attention mechanism.An end-to-end intelligent diagnostic model is built,resulting in a novel defect diagnosis concept.The following are the key research findings of this paper:The general framework of the intelligent fault diagnosis method for rotating machinery is first introduced,followed by a summary of the research status of the diagnosis method based on the vibration signal of rotating machinery as an input object both at home and abroad,as well as the benefits of vibration diagnosis over other state information.Second,defect detection and diagnostic techniques based on neural networks,support vector machines,and other machine learning algorithms are identified and presented from various angles.Second,under the situation of difficult feature extraction and noise submerged vibration signals,a fault diagnosis technique integrating 2-D grayscale picture and Transformer is presented to overcome the problem of low fault identification accuracy in rolling bearing fault detection.The 1-D time series is first converted to a 2-D grayscale image,which is then sliced and fed into the Transformer model,which is made up of a fully connected network and an attention mechanism.To enhance the attention of sparse matrix,the threshold shear layer and Sparsemax activation function were used in coding to extract the characteristics of the classification by Softmax,and lastly,rolling bearing data sets of tests were carried out to verify the proposed model.The research results indicate that the classification accuracy of2D-Transformer model on rolling bearing data sets reaches 99.35%.The upgraded model’s attention has a better sparse matrix thanks to the matrix distribution of visual attention.Experiments indicate that the proposed model is reliable in detecting rolling bearing faults.Thirdly,There was over-reliance on experts’ knowledge when extracting time domain signals by the traditional fault diagnosis method of rolling bearings,and the fault information was expressed inadequately by features.Aiming at the problems,an intelligent fault diagnosis model based on residual network and capsule network was proposed.Firstly,raw vibration signal was used as input,and the one-dimensional convolution neural network was used to extract global features from the time domain signal,and then the residual network was used to extract the low-level features of the data,and they were sent to the capsule network to vectorize the low-level features,after that the low-level features were combined into advanced features and classified through dynamic routing process which was improved by fuzzy clustering.Finally,in order to verify the effectiveness of this method,the proposed method was tested through the rolling bearing data sets.The research results indicate that the residual cap-sule network reaches 99.95% in classification accuracy The t-distributed stochastic neighbor embedding(t-sne)visible analysis further verifies that the network model has the ability to self-adaptively mine high-level features.The residual capsule network possesses good accuracy and generalization in the fault diagnosis of rolling bearings.Finally,a new model based on end-to-end fault diagnosis of attention mechanism and capsule network is proposed in response to the diagnosis of rotating machinery and how to extract and express significant aspects from complicated fault signals.As an input,the model takes a one-dimensional original vibration signal.The attention mechanism is used in the feature extraction step to obtain a feature graph with varying weights.The defect feature is vectorized by the capsule network to improve the expression ability,and the dynamic routing technique is improved based on the idea of G-K clustering to adapt to the complicated data set during the feature fusing phase.The rolling bearing and gear data set is used to verify the test and is compared to various depth learning techniques.The results show that: The average accuracy of the proposed model is 99.37% on the rolling bearing data set,99.91% on the gear data set,and 98.69% on the gear and bearing combined data set.Experimental results show that the average accuracy and stability of the proposed model are better than other deep learning models.