Skip to Main Content
This paper describes an application of state-space parametric identification techniques for determining the presence and nature of un-modeled stochastic components that are present in test data but not in the assumed (baseline) model (i.e., model faults). The technique uses residuals from a data processing filter based on the assumed model as inputs. This information is combined with information about the structure of the assumed model to build a second-stage identification filter which is used as the core of the parametric identification procedure. The approach uses the small-dimensioned second-stage filter to capture the disburbance (unmodeled stochastic components) dynamics and noise structure. The use of a small, independent second-stage filter, where alternative fault structures and locations can be experimented with and compared, allows the freedom to perform off-line model structure verification. Alternative model structures are evaluated statistically through the use of likelihood ratios and chi-square test statistics. Simulation results are given for a 76-state guidance system model.