Research On Small Sample Data Feature Extraction And Classification Model Of Industrial Equipment

With the gradual improvement of industrial intelligence and scale,the connection of all processing units in the industrial process becomes closer and closer.Once a certain equipment failure will lead to the stagnation of a wide range of production links,resulting in huge economic losses,so it is necessary to ensure the stable and reliable operation of industrial equipment effective methods.At present,the fault monitoring technology of industrial equipment is mostly based on large sample data sets.However,in the actual industrial process,the amount of fault data collected is usually small,which is a small sample fault.The rest of the data are normal operation data,which makes some fault monitoring methods which need a large number of fault samples for training invalid.Today,there are two key steps in the method of fault monitoring for industrial equipment: feature extraction and establishment of classification models.In the feature extraction method based on signal processing,there is a problem that the signal decomposition effect is not ideal,and the signal contains similar fault characteristics and noise interference,and the fault feature extraction accuracy is poor.As for the research on classification model,due to the various failure categories of industrial equipment,the direct use of binary classification model for classification may lead to an increase in calculation and training time,which has certain limitations.Therefore,it is of great significance to carry out feature extraction and classification model research of small sample data of industrial equipment.In this paper,aiming at improving the accuracy of signal processing feature extraction and establishing a more suitable multi-category and multi-classification model,a fault monitoring method for industrial equipment is proposed.The main work is divided into two aspects:(1)Aiming at the fault feature extraction method of small sample in signal processing,a feature extraction method based on Decorrelation Multiple-Frequency Empirical Mode Decomposition is proposed.By adding masking signals,the correlation coefficient processing operation is embedded in the decomposition process to reduce the interaction between normal feature information and fault feature information and the aliasing phenomenon in decomposing mixed signals is suppressed so the capability of decomposing signals in empirical mode is finally improved.Then,the Sample Entropy algorithm is combined to highlight the fault features in the signal and further improve the accuracy of fault feature extraction.(2)Analyzing the multi-classification model of Relevance Vector Machine for small sample fault data,propose a Multi-Classification Relevance Vector Machine model based on Cuckoo Search algorithm optimization.This method can not only give full play to the classification performance of Relevance Vector Machine,realize the recognition of different faults at the same time,but also optimize the kernel parameters by Cuckoo Search algorithm,which can fully improve the Multi-Classification Relevance Vector Machine.The self-adaptability greatly improves the classification performance,and can effectively improve the efficiency of fault monitoring.The simulation results show that the proposed feature extraction method can significantly suppress the modal aliasing phenomenon in the decomposition of different mixed signals,and effectively improve the extraction of fault features from small samples.When dealing with small samples,the Multi-Classification Relevance Vector Machine model can identify multiple fault states with high classification accuracy,and can achieve high-efficiency and high-precision fault monitoring.In order to extend to practical application.