Research On Fault Monitoring Of Complex Industrial Process Based On Multivariate Statistics

With the development of science and technology and the continuous improvement of market requirement,the development direction of modern industry tends to be large-scale,complex,integrated and refined.Once the process fault occurs,it is not only difficult to investigate,affect the quality and production of products,but also lead to the occurrence of large casualties.Therefore,it is particularly important to timely monitor the occurrence of faults and diagnose faults.In recent years,with the development of sensor technology and computer technology,both data acquisition ability and data processing ability have been greatly improved.Statistical process monitoring technology based on multivariate statistical method has been widely studied and applied.By analyzing the measured industrial process data,it calculates and compares monitoring statistics and statistical control limits,and monitors whether the current process is abnormal.In view of the various characteristics of complex industrial processes,this paper combines the traditional principal component analysis(PCA)and local preserving projections(LPP)and other process monitoring and data processing methods,studies new methods,and improves the fault monitoring ability of the method for complex industrial processes The specific contents are as follows:(1)In view of the traditional PCA method only retains global information and the nonlinear characteristics of industrial process data,the global local preserving projections(GLPP)method and kernel PCA(KPCA)method are introduced.By combining the two methods,a new GLPP-KPCA method and a new index of fusion monitoring statistics are proposed.(2)Aiming at the problem of noise in the measurement data of the sensor,by using a sliding window wavelet denoising method,and combined with the GLPP method,a SWWD-GLPP method is studied,which can effectively solve the problem of data noise and improve the detection accuracy.(3)In this paper,a dynamic GLPP algorithm is studied.By using the method of extension matrix,the ability of the algorithm to monitor industrial processes with dynamic characteristics is increased.A new method of determining the contribution diagram is used to accurately diagnose the major variables of the fault.The simulation results show that although the traditional PCA and other methods can detect most industrial process faults,the detection effect for some complex industrial process faults is not ideal.The GLPP proposed by combining PCA and LPP can preserve the global and local characteristics of dimension reduction data.By combining GLPP with KPCA,the monitoring problem of nonlinear process can be solved.The combination of wavelet denoising and GLPP can improve the detection rate of some faults.The data matrix is extended to improve the ability of monitoring dynamic process.