ICA-based fault-relevant reconstruction

In this study, an ICA-based fault relevant reconstruction method is proposed for fault detection and diagnoses. ICA-base method makes it possible to analyze sample data with non-Gaussian quality. Further fault reason identification is based on the extracted independent components which involves higher-order statistics. According to the fault relevant reconstruction method, the fault relevant direction is found in independent component subspace. Along this fault direction, reconstruction will eliminate the fault cause and bring faulty statistic under the control limits. When all kinds of possible faults are analyzed, and their fault directions are identified, the new fault data will be diagnosed with which kind of fault it belongs to. In this paper, ICA-based fault relevant reconstruction method is applied to two examples, simple liner process and penicillin fermentation process. The simulate results show the capability of this new method to diagnose sample data having non-Gaussian.