Fault Diagnosis Approach Based On Improved Partial Least Squares And Contributions Plots

With the continuous progress of modern science and technology in our country,the whole industrial production process has become more and more complicated.If the equipment fault in the production process operation of the enterprise can not be found in time,then it may directly lead to the whole system into an abnormal state,and then cause the system to shut down or production safety accidents.Compared with the traditional data-based process monitoring technology,the quality-related data-based process monitoring technology can better establish a close relationship between faults and product quality,and in the industrial process of failure is divided into two kinds,that is,quality related failure and quality independent failure.This classification method will not only make part of the fault alarm greatly reduced,in the meantime,can also reduce part of the unnecessary stop maintenance,so that the production and maintenance costs are significantly reduced,directly make the enterprises significantly improve their economic benefits.Because of the uniqueness of the technology,the quality-related data-based process monitoring technology has become the focus of engineering and academic research.This paper focuses on the quality-related fault diagnosis as follows:(1)In view of the partial least squares(PLS)inadequate of separation of variables,and it is unable to provide accurate results,this paper puts forward a kind of improved partial least squares(IPLS)contribution figure to diagnose the fault diagnosis strategy of relevant variables,and through the Tennessee Eastman(TEP)experiments to verify the effectiveness of the IPLS model.From the experimental results of TEP model,it can be seen that IPLS successfully divides faults into quality-related subspaces and quality-independent subspaces,and the contribution graph of IPLS can distinguish quality-related variables.Compared with PLS,the contribution graph based on IPLS has better fault diagnosis capability.(2)In view of the existing contribution figure technology decomposition is not correct,it can not accurately provide quality related fault diagnosis results,in this paper,a method based on Least Squares(LS)and contribution graph is proposed to realize fault diagnosis.this method can obtain mass dependent subspaces and mass independent subspaces by decomposing the variable space of the orthogonal decomposition process.Then,the variable contribution of each subspace Q statistic is calculated respectively,and the relative contribution graph of each variable is obtained.The occurrence of the fault can be directly detected,and influence of the failure on the final quality and the main cause of the failure can be detected according to the defined control scope.Finally,the proposed method is validated by the Tennessee Eastman process(TEP),the proposed method based on least square method and contribution graph can be applied to the detection and diagnosis of quality-related sensor faults,that is,the occurrence of faults and the effects of faults on quality can be detected.