| Rotating machinery is an important power equipment in modern industrial production,and the good or bad operating status will directly affect the running accuracy and stability of the equipment.Composite fault,as the most common form of failure in rotating machinery,is of great significance for the effective diagnosis of equipment to ensure safe and stable operation.This article focuses on the problem of difficult acquisition of composite fault samples and difficult fault identification under noise interference in engineering sites.It carries out composite fault diagnosis research on the key components of RV reducer and bearings in rotating machinery under the conditions of non-complete sample space and noise interference.The main research work of the paper is summarized as follows:（1）A detailed introduction is provided on the classification and recognition principles of convolutional neural networks and capsule networks,and a comparison is made between the two networks from four aspects:the basic composition structure,classification principles,label output methods,and training and learning mechanisms.（2）To address the problem of incomplete collection of composite fault sample data in industrial scenes,a study of intelligent diagnosis of composite faults under non-complete sample space constraints was conducted,and a composite fault diagnosis method based on an Improved Convolutional Capsule Network was proposed.The method uses a one-dimensional convolutional neural network as a feature learner to learn discriminative deep features from the vibration signals of a single fault.Subsequently,a feature classification recognizer is constructed based on capsule network,which uses dynamic routing algorithm to classify and gather the fault features in composite faults.And theL2 norm is used to independently output the fault probabilities,achieving the goal of compound fault classification.Experiments were conducted in the homemade RV reducer and bearing test bench to verify the effectiveness of the proposed method and to lay the foundation for the subsequent compound fault diagnosis research.（3）Aiming at the problem that vibration signals collected in industrial sites are often accompanied by strong noise interference,it is difficult to carry out effective fault diagnosis,and a composite fault diagnosis method based on wavelet convolutional capsule network is proposed.A compound fault diagnosis method based on wavelet convolution capsule network is proposed.The method incorporates discrete wavelet transform into the first layer of convolutional neural network,and adaptively optimizes the multi-scale wavelet convolutional filter through gradient descent method to enhance the feature extraction ability of the network under strong noise interference.Capsule network is also combined to improve the accuracy of composite fault diagnosis.The effectiveness of the proposed method was verified on the bearing test bench.（4）To address the problem that intelligent diagnostic algorithms are difficult to deploy on site,we develop intelligent diagnostic nodes based on the Jetson platform,taking into account the different data collection needs of industrial sites.Two different design schemes of intelligent diagnosis nodes are proposed,and the functions of vibration data acquisition,data storage,data release,packet spectrum analysis,time-frequency analysis and other intelligent diagnosis are designed and implemented.The proposed intelligent diagnosis algorithm is deployed to the Jetson platform,and the effectiveness of the proposed method and nodes is verified with full-life experiments on rolling bearings.The research results of this paper provide new ideas and techniques for intelligent fault diagnosis of machinery and equipment,which are of great significance in promoting intelligent fault diagnosis technology to engineering applications. |
