A fault detection approach based on radial basis function neural network(RBFNN) was presented herein, this approach consisted of system identification, filtered residual generation and fault alarm concentration(FAC). The system identification was based on immune strategy RBFNN, and the residuals were generated by on-line comparing the system model outputs with the actual system ones. The generalization capability of immune strategy RBFNN was enhanced by introducing generalization capability interfering factor in the affinity function of the clone selection algorithm. The proposed immune model-based fault detection approach is only sensitive to the residuals caused by faults because of the introduction of residual filtering and FAC. Simulations on a parallel manipulator were conducted to evaluate and validate the effectiveness and robustness of above fault detection approach, simulation results show that it can detect and locate actuator faults and sensor faults efficiently, and it is sensitive to faults while at the same time insensitive to unspecified uncertainties.

提出了一种基于径向基函数（RBF）免疫神经网络的故障检测方法,该故障检测方法由系统辨识、残差过滤和故障报警浓度等功能模块构成。系统辨识基于免疫RBF神经网络,用于故障检测的残差是通过对系统的模型输出与系统的实际输出进行在线比较得到的。在克隆选择算法的亲和力函数中引入泛化能力干涉因子,增强了RBF网络的泛化能力。在该故障检测方法中,通过过滤残差和引入故障报警浓度,使得故障检测仅对因故障引起的残差敏感。并联机器人的故障检测实例表明,该方法能够有效地检测和定位出驱动器故障和传感器故障,具有良好的容噪性能。

故障检测, 人工免疫, 神经网络, 系统辨识