Research On Mechanical Fault Intelligent Diagnosis Based On Deep Capsule Network

Rotating machinery,as an indispensable and important equipment in industrial production,plays a pivotal role in all walks of life;Once the rotating machinery fails,production will be brought to a standstill,which can cause economic losses at a light level and casualties at a heavy level.Therefore,it is of great practical significance to research and develop effective fault diagnosis techniques for rotating machinery to ensure safe,reliable and stable operation of rotating machinery and avoid accidents.Currently,intelligent fault diagnosis methods based on deep learning have the advantages of not relying on expert experience,strong feature extraction ability,and become a research hotspot in the field of rotating machinery fault diagnosis.However,due to the complex operating conditions of rotating machinery and the characteristics of non-stationary and nonlinear vibration response signals,it is difficult to extract fault feature information.Moreover,the equipment is mainly operating in a healthy state for a long time,making it difficult to obtain a large number of sample data of different fault types.Therefore,in response to the above issues,this paper uses the powerful representation ability of capsule networks for complex data features to carry out research on intelligent fault diagnosis methods for rotating machinery based on deep capsule networks.The main research contents are as follows:(1)Aiming at the complex characteristics of rotating machinery and the difficulty of comprehensively extracting fault features from vibration signals,a fault diagnosis method for rotating machinery based on attention dual scale feature fusion capsule network was proposed.Firstly,two convolution layers with different kernel sizes are used to extract dual scale features from grayscale images;Secondly,an attention based dual branch network is designed to calculate feature weights at different scales and perform feature fusion;Finally,the fused features are input into the capsule layer,and the classification loss and reconstruction loss optimization model are used to achieve the classification and recognition of rotating machinery faults.The proposed method is analyzed using XJTUSY rolling bearing fault dataset and SQI motor fault dataset,and the results show that the proposed method has good diagnostic performance for rolling bearing and motor faults.(2)Aiming at the small sample learning problem caused by insufficient fault data samples in engineering practice,a siamese capsule network based small sample fault diagnosis method for rotating machinery was proposed.Firstly,the fault samples are cross matched and recombined to improve the usage rounds of training samples to expand the training set;At the same time,the capsule network is used as the sub network of the siamese network to extract the fault characteristics of the sample pairs;Then,Euclidean distance is used to calculate the similarity between fault features and input the fault features into a classifier to complete fault diagnosis.Finally,the classification loss and similarity loss are calculated to iteratively update the model.The performance of the model is analyzed using the CWRU rolling bearing experimental dataset and the SQI motor fault dataset.The results show that the proposed method can achieve high diagnostic test accuracy under small sample conditions...