Research Of Fault Diagnosis Based On Relative Principal Component Analysis

The paper studies mainly about relative methods of fault diagnosis, rela-tive principal component analysis and kernel principal component analysis are researched.The paper studies relative principal component analysis. As each variable has different sensitivity in the principal component space and residual space, the proportion factors are acquired, and they are used as weights in the relative translation. When the weights are added before each variable standardized, the variables are normalized and the diversities of all the variables are enlarged, so the contradiction is solved. The paper studies the method of relative prin-cipal component analysis and the simulation is finished on TE.As the single index can’t reflect all the information change of the varia-bles, the paper combines the daily used indices used in relative principal component analysis. And two combined indices are researched:the combina-tion between T2and SPE and the combination between T2and TH2. Also, the two indices are used in the fault diagnosis of relative principal component, and the simulation is finished on the TE. The simulation shows that the result using the combined indices is better than using single index, and the two combined indices show the same perfect simulation result.As the relative principal component analysis is mainly applied in the lin-ear system, and TE may be nonlinear system, the paper studies kernel princi-pal component analysis, and the simulation is finished on TE. The simulation shows the result is almost perfect.At the last of the paper, the relative principal component analysis and the kernel principal component analysis are combined and relative kernel princi-pal component analysis is put forward. The simulation is finished on TE and it shows the same good result as kernel principal component analysis.