A B-spline neural network based actuator fault diagnosis in nonlinear systems

For a number of industrial systems, actuators are characterized not only by the general nonlinear (saturation) function of the control input, but also by the derivatives of the input. For these type of actuators, their fault detection and diagnosis can be performed by assuming that the nonlinear functions can be approximated using a B-spline neural network expansion. As a result, the fault diagnosis task is to estimate the expansion weights of the B-spline approximations once all the basis functions are fixed. In the paper, two actuator fault diagnosis algorithms are developed for general known nonlinear systems. First, a nonlinear observer is proposed which detects and diagnoses the actuator faults via adaptive tuning rules. This is then followed by the design of an adaptive diagnostic observer whose nonlinear functions can be approximated by their first order linearizations around the neighborhood of the observed state and the estimated faults. An adaptive tuning method has thus been formulated which ensures a good estimation of the actuator faults and guarantees that the observer vector is uniformly bounded