基于神经网络的非线性观测器及在线故障检测

提出一种基于径向基函数神经网络的非线性观测器的设计方法，并将其应用于复杂非线性系统的故障检测与隔离。该方法将神经网络离线学习与在线学习相结合，获取系统输入输出的非线性动力学特性，进而实时计算出残差并进行逻辑判决，可显著提高故障检测的快速性、鲁棒性及准确率。最后，针对非线性同步交流电机的结构损伤故障进行了仿真，结果表明本文所提方法的有效性。

A Nonlinear Observer Based on Neural Networks and On Line Fault Detection

A new nonlinear observer based on radial basis function (RBF) networks is presented, and it is applied to fault detection and isolation (FDI) in complex nonlinear systems. The RBF network is used to model a multi input and multi output nonlinear dynamic system by off line and on line learning rule. The output prediction error, generated from the real output and the RBF estimated output, is used as a residual error to indicate the occurence of any faults. The approach improves the accuracy and rapidness of FDI, and also has robustness pro perty under the disturbance and noise. Simulation in nonlinear synchronous motors show that structural damage faults can be reliably detected and effectiveness of the fault detection approach is revealed.