Neural network-based robust fault detection for nonlinear jump systems

The observer-based robust fault detection (RFD) design problems are studied for nonlinear Markov jump systems (MJSs). Initially, multi-layer neural networks (MNN) are constructed as an alternative to approximate the nonlinear terms. Subsequently, the linear difference inclusion (LDI) representation is established for this class of approximating MNN. Then, attention is focused on constructing the residual generator based on observer. What is more, in order to take into account the robustness against disturbances and sensitivity to faults simultaneously, the H1 filtering problem is formulated to minimize the influences of the unknown input and another new performance index is introduced to enhance the sensitivity to faults. Based on this, the RFD observer design problem is finally formulated as a two-objective optimization and the linear matrix inequality (LMI) approach is developed. An illustrative example demonstrates that the proposed RFD observer can detect the faults shortly after the occurrences without any false alarm.