Fault Diagnosis Approach Based On Principle Component Regression Algorithm

With the wide application of data-driven approaches in large-scale industrial equipment and systems,data-driven quality-related process monitoring technologiesare also attracting more attention in recent years.Compared with the former data-driven monitoring technologies,quality-related data-driven monitoring approaches can obtain the relationship between the product quality and faults,which can divide the faults into quality-related part and quality-unrelated part.This partition can reduce the unnecessary fault alarming and repairing that can reduce the costs of maintenance and production.As the uniqueness of such methods,quality-related process monitoring technologies have been a hot topic in engineering and academia area.However,due to its short time of development,there are someexisted problems to be resolved.For instance,this technique cannot handle the dynamic problem in the data sets which are collected by the industrial process systems.In this paper,the principle component regression algorithm(PCR)of multivariate statistical analysis techniques is adopted with quality-related fault diagnosis methods to get a better improvement for the dynamic and static process systems.Furthermore,in this paper,the research body of contents are as follows:(1)In process systems,the former multivariate statistical analysis technologiesare static,and the sampled data in the actual process usually have dynamic characteristics.Based on the basis of the PCR algorithm,a new method called dynamic principal component regression(DTPCR)is proposed which is combined with auto-regressive moving average exogenous(AXMAX)and PCR algorithm.The novel method adapts the dynamic characteristics of the process data and provide the dynamic relationship beweent the input data and output data.Further,DTPCR can separate the process variables into quality-related subspace and quality-unrelated subspace to achieve a better results of fault detection and quality prediction.The Tennessee Eastman Process(TEP)and the numerical examples are applied to the comparison of PLS and DTPCR in the simulation of quality-related fault detection and quality prediction.Compared with the standard PLS,DTPCR has the better abilities of fault detection and quality prediction.(2)This paperalso designed a diagnosis method calledcontribution plots based improvedPCR(IPCR)to improve the inaccurate diagnosis problem existed in the previous methods.This approach mainly employs the contribution plots method and the detection results of IPCR algorithm,which calculates the contribution of T~2 statistics one by one to diagnosis the reason behind the fault.Meanwhile,a novel fault diagnosis strategy is proposedin this paperbased on proposed diagnosis method.Compared with contribution plots based PLSalgorithmin the experiment of TEP benchmark,it is clearly concluded that the designed method can provide more accurate results and more explicitlyfault diagnosis strategy which adapts the quality-related fault diagnosis methods.