Nonlinear multimode process fault detection based on KNN-KICA

In order to detect faults in nonlinear multimode industrial process, a new fault detection method is proposed based on k nearest neighbor-kernel independent component analysis (KNN-KICA). Firstly, process data are standardized with its k nearest neighbors to eliminate multimode difference. Then, in consideration of the nonlinear dependency among data variables, the algorithm maps the data in original nonlinear space into linear space by kernel function technique. Finally, independent component analysis (ICA) is applied to construct monitoring statistics for fault detection. Simulation results on a continuous stirred tank reactor (CSTR) system show that KNN-KICA can obtain better performance in process monitoring than traditional ICA.