Non-linear adaptive fault detection filter

Publisher Website
Google Scholar
Abstract
A novel non-linear adaptive fault detection filter (NAFDF) is proposed. It can be used to detect on-line and isolate the faults of a class of non-linear systems arising from accidental jumps of the process parameters. The extended Kalman filter and weighted sum-squared residual method are first combined to delect the faults rapidly. A non-linear filter is then proposed and used for joint state and parameter estimation of the system, resulting in a series of parameters. Based on them, Bayes' decision algorithm is modified and used to isolate and classify the faults. An alternate initialization method is also presented, which makes it possible to detect and isolate the faults repeatedly. Finally, the effectiveness of the NAFDF is demonstrated by a simulation study.