An Approach to Detect and Isolate Faults for Nonlinear Systems with Periodic Input

To improve the reliability of engineering systems, fault diagnosis techniques are often incorporated into the operating system. However, most existing techniques are derived for systems with non-periodic input, even though periodic inputs are also common. For this reason, the aim of this paper is to derive a diagnosis technique for nonlinear systems with periodic input. The proposed technique is derived assuming the frequency of the input is available. However, if it is not available, it will be estimated first and its estimate is used for fault detection and isolation. The proposed fault diagnose technique involves a statistics, the mean of which becomes non-zero when a fault occurs. To test the mean of this statistics, a hypothesis test is derived based on the asymptotic local approach, which yields a χ2-test. The implementation of the proposed method is presented, and the performance and robustness illustrated by a simulation involving a nonlinear system with sinusoidal and square input.