Model predictive estimation of evolving faults

In this work we present an optimization based statistical estimation approach for diagnostics in large scale systems. The fault estimation scheme relies on prediction residuals generated by detailed prediction models of the system under consideration. The system dynamics are generally nonlinear. We linearize the system around its nominal operation and estimate deviations (faults) from the nominal behavior. The statistical estimation approach is based on numerical optimization of a log-likelihood function. It allows us to estimate time varying fault parameters in an online setting, and can accommodate the loss of some sensor measurements during system operation. The proposed estimation approach is explained through examples from aerospace applications