Research On Fast Incremental Learning Based Fault Diagnosis Method For Rotating Machinery

The degree of intelligence and complexity of rotating mechanical equipment poses new challenges for its efficient and accurate fault diagnosis.With the development of sensor technology,fault diagnosis methods with artificial intelligence technology as the core have been widely used,and relevant algorithms using pattern recognition and machine learning are able to determine whether a rotating mechanical component is malfunctioning and identify its fault type.However,in practical engineering applications,due to the complex working environment of mechanical equipment,as time passes,the wear and tear of components intensifies and new types of faults may arise;general methods such as support vector machines and neural networks use batch learning to train models,so it becomes difficult to update fault diagnosis models when new data appear.Therefore,there is a need to study suitable incremental learning methods to achieve the purpose of updating models efficiently with new data and avoid the disadvantage of batch learning which requires a lot of time and resources to update fault diagnosis models.In this paper,we take rolling bearings,a key component of rotating machinery and equipment,as the research object to study the fast incremental learning method in rolling bearing fault diagnosis.The main research content of the thesis is divided into the following parts:（1）The incremental learning in rolling bearing fault diagnosis is divided into two main steps: feature extraction and model building.Firstly,based on the characteristics of rolling bearing fault signals,various feature extraction methods of rolling bearings are studied,including feature extraction method based on convolutional neural network,feature extraction method based on multivariate multi-scale entropy,and multi-domain feature extraction method based on statistical analysis.Then,in the incremental learning scenario,these three different methods are used to conduct feature extraction experiments,which lays the foundation for the construction of subsequent incremental learning models.（2）Regarding the model construction problem in the incremental learning process,two neural network-based category incremental learning methods are used:the first method is a class incremental learning method based on the theory of autoencoder,which constructs a support function using reconstruction error based on the principle that autoencoders can reconstruct signals,and combines autoencoders belonging to multiple classes into a multi-fault classification model for rolling bearings.This method has high accuracy,but it takes a lot of time to build the model.In order to solve the above problem,the autoencoder is optimized using an extreme learning machine,and a second type of fast class incremental learning method based on an extreme learning machine autoencoder is proposed,which can construct the model efficiently and fastly,but the fault diagnosis accuracy is slightly reduced.（3）Finally,in order to achieve high fault recognition accuracy while shortening the model construction time,the support vector data description theory is studied,a category incremental learning method based on the combination of Bayesian optimal decision theory and support vector data description is proposed,and the training process of the support vector data description model is optimized,and then the proposed method is applied to the bearing fault diagnosis.It is demonstrated that the class incremental learning method based on support vector data description can shorten the construction time of the fault diagnosis model while ensuring a high recognition accuracy.