An iterative principal component test for fault detection and isolation

In most safety critical systems, control and protection operations are based on signals of various detectors measuring different variables. This signal data should accurately represent the true variations of variables. But in practice, the data from the detectors are not deterministic, since they contain random variations arising from various causes. On the other hand, faults also may occasionally take place in detectors. Random errors and faults need to be removed or reduced to a minimum possible level, so that efficient control and protection operations can be performed. In this paper, a new technique for Fault Detection and Isolation (FDI), called Iterative Principal Component (IPC) test, which is derived from the principal component test, has been proposed. Its effectiveness is examined when applied on two sets of the neutronic detector data from nuclear reactors. This test is performed on the measurement residuals obtained from the steady-state Data Reconciliation (DR) technique, which is powered by a constraint model developed from Principal Component Analysis (PCA). The DR coupled with the IPC test is aimed at minimizing the effects of random errors and faults in the data. For evaluating the performance of these coupled techniques, comparison with other FDI techniques such as Generalized Likelihood Ratio (GLR) test and Iterative Measurement Test (IMT) is presented.