Research Of Fault Diagnosis Method Based On Kernel Partial Least Squares

With the increasing requierment on safety, reliabiliy and effectiveness of automation devices, stuy on the problem of fault diagnosis has received great attention and been one of the most active research topics during the past three decades. Multivariate statistical analysis has been considered as an important method of fault diagnosis due to its independence of process model and the availability of mass data in industrial process. Based on fault diagnosis method of multivariate statistical anaysis, the main purpose of this article is to make further stuy on the fault diagnosis method based on PLS and KPLS and develops their application to the Tennessee Eastman process. The main contents of the paper are as follows,Study the basic principle of the partial least squares (PLS) and construct the model of fault diagnosis in use of the partial least squares; According to the principle that the partial derivative of a function with respect to a specific dimension can indicate the relative influence of the corresponding variable on that function, the Variable contribution plot of the partial derivatives based on PLS is proposed and the simulation result indicates that the method of the contribution of the partial derivative of varible applied to the fault identification of PLS is feasible and effective.On the basis of the fault diagnosis method of partial least squares, study the fault diagnosis method of kernel partial least squares (KPLS) further. The fault diagnosis of the kernel partial least squares is difficult because of unknown mapping function. The paper makes use of an indirect method to obtain partial derivative of variable in kernel function. At the end, the paper proposes a fault identification method based on kernel partial least squares-variable contribution plot of the partial derivatives.The simulation is researched in use of Tennessee Eastman process data. The result indicates that the method of the fault detection based on kernel partial least squares is good and the result of fault diagnosis is accurate. At the same time, compared with the fault diagnosis method based on partial least squares, the fault diagnosis method based on kernel partial least squares has better result in nonlinear domain.In a word, the kernel partial least squares is applied to fault dignosis, meanwhile, the article proposes a method of fault identification based on KPLS- variable contribution plot of the partial derivatives, and solve the problem of nonlinear identification difficultly.