Key variable identification using nonlinear variable weighting

Fault identification is an important topic in statistic process monitoring. It aims to identify key observation variables most relevant to diagnosing the specific fault in order to focus the plant operators and engineerspsila attention on the subsystems. A new fault identification approach with nonlinear variable weighting is proposed. The variable weighting finds out the weight vector of each fault by maximizing separation between the normal and each fault data. Instead of Rayleigh coefficient in kernel Fisher discriminant analysis (KFDA), the kernel target alignment is selected as the variable weighting criteria. With continuous non-negative values, each element of the weight vector represents the corresponding variablepsilas contribution to a special fault. Applying the proposed method to the Tennessee Eastman process, the results show more reliable fault identification than the contribution plot based on the fault direction in pair-wise FDA. The nonlinear variable weighting is a promising technique for nonlinear key variable identification.