On-Line Monotonic Fault Diagnostics via Recursive Constrained Quadratic Programming

In this paper, we present a model-based approach for estimating fault conditions in a turbine. We formulate fault estimation as a convex optimization problem, where estimates are obtained by solving a quadratic program (QP). A moving horizon framework is used to enable recursive implementation of the QP of fixed size. The estimation scheme takes into account a priori known onotonicity constraints on the faults. Monotonicity implies that the fault conditions can only deteriorate with time. We validate the proposed estimation scheme on a sample of real data from a turbine. An excellent performance of the developed approach is demonstrated by detecting a developing failure while the unit is running at high normalized loads.