基于神经网络的假设检验故障诊断算法研究

传统的χ2检验法通过比较先验信息与量测信息进行故障检测,当先验信息所依赖的模型出现错误时将无法进行有效的故障检测和故障隔离。提出利用神经网络的方法结合χ2检验法进行故障诊断,训练神经网络来跟踪系统模型,则正确的先验信息保存在训练好的神经网络中,系统模型输出与网络模型输出之差作为故障诊断的依据。针对INS/GPS组合导航系统进行算法仿真,该算法能够快速、准确地判断系统故障源,通过故障隔离和系统重构,使系统在故障情况下依然保持正常工作。

Hypothesis Testing FDI Algorithm Based on Neural Networks

The traditional cai-square hypothesis testing processes fault detection by comparing the apriori information with the observation information. If the INS goes wrong, a fault model can't provide accurate apriori information, thereby the fault detection and isolation can't be performed effectively. A new fault detection algorithm based on neural network was proposed. A neural network was trained according to INS model, so the accurate apriori information was held by the trained NN. The difference of the output between networks and model can be the parameter of fault diagnoses. The application on INS/GPS integrated navigation system demonstrates that the algorithm exhibits excellent fault detecting and identifying ability. Therefore the effective fault isolating can be performed to realize the fault tolerance navigation.