Research On Fault Diagnosis Method Based On Principal Component Analysis

At present,as one of the hotspots in the research of complex processes,data based technology has been developing rapidly and has been widely applied in many industrial sectors.The core of the data based approach is to make full use of a large amount of available process data for the purpose of obtaining internal useful information.Compared with the traditional model based approach,the data based approach provides an effective alternative for different industrial models under different operations.In the past twenty years,modern large-scale processes have experienced great challenges and are likely to become more complex in the industrial sector.With the remarkable improvement of automation,the sensors and actuators installed in modern industrial processes will generate a lot of process data during the operation of the system.Therefore,it has become impractical for the prior knowledge or the physical model in the traditional model-based approach to acquire requirements,especially for complex large-scale industries.The development of modern industry not only improves the complexity of the system model,but also increases the calculation burden of the traditional solution.As the complexity of the large-scale structure is becoming more and more complex and a large amount of industrial data is generated,the new method of data based development is further developed.By using the data based multivariate statistical analysis method,the early fault diagnosis technology is studied,and the validity of the data technology is revealed.With the continuous development of data technology,the application scope has been extended to the dynamic process modeling and monitoring fields.In the past ten years,with the rapid development of modern industry,the amount of data has been increasing,which has greatly accelerated the progress and breakthrough in the related fields.Process data are now used to improve the entire human society,especially the industrial contribution.The latest achievements in the application of a series of new technologies,such as communication system,advanced data acquisition,storage and transmission equipment,have promoted the realization of data utilization.On this basis,information science,electrical and electronics,biology and economics will further benefit from data based technology in order to make more effective use of large amounts of data.Data based technology has been applied to excellent operation and prediction analysis to create further competitive advantages for contemporary forward-looking business.Based on the theory of PCA in multivariate statistical analysis,this paper makes a systematic and in-depth study of some important aspects of the PCA method.The main research contents and results are as follows:1)the research status of fault diagnosis is studied,including research methods,research contents and development trend.The development and current situation of fault diagnosis based on multivariate statistical methods are reviewed in detail.the main mathematical tools of fault diagnosis based on multivariate statistics are introduced,especially the principal component analysis(PCA).2)The sample data collected from a large industrial system usually contains noise,and how to denoise is an important research method in this paper.This paper selects the method of wavelet denoising,wavelet denoising method has some characteristics,such as multi-resolution properties,non-stationary signal denoising can depict good wavelet,such as edge,spike,breakpoint,diverse wavelet base selection,flexible choice of wavelet bases can be different,according to the signal characteristics and choosing the appropriate wavelet denoising,achieve better denoising effect.3)Usually,there is uncertain relationship between raw data.How to determine whether the original data is linear or nonlinear correlation is very important for selecting proper methods for principal component analysis.Therefore,this paper will combine the multivariate statistical regression method to determine the correlation between data and further determine the selected methods.4)For kernel functions,different kernel functions have different properties.How to combine the advantages of different kernel functions to make up for each other's weaknesses?In this paper,we study the construction properties of kernel functions in detail,so as to achieve better results.