Mallat小波快速变换与IDRNN在卫星实时故障检测与识别中的应用

Application of Mallat wavelet fast transforms and IDRNN in real-time fault detection and identification for satellites

系统研究了面向复杂系统监测时变信号的实时故障检测与识别问题.采用滑窗Mallat小波快速变换克服传统小波变换的时域全局依耐性并提高计算效率,使之适应于实时故障检测;针对时变信号的故障模式识别难题,在故障检测基础上采用改进动态循环神经网络(improved dynamic recurrent neural network,IDRNN)进行智能故障识别.最后将滑动时窗小波检测模块及最优IDRNN网络模块嵌入某型完整卫星姿态控制系统仿真平台进行在线故障诊断.试验结果表明:实时条件下的滑动窗口小波变换与传统小波变换具有一致性,IDRNN对于时变信号识别具有较好的时域泛化能力,将滑窗移动时不变小波方法与IDRNN结合可以实现面向复杂系统监测实时信号的故障检测及复合模式分类.

A real-time fault detection and identification （FDI） scheme of time-variant signals for a complex system was studied. A sliding-window Mallat wavelet fast transform was first introduced to avoid depending on the signals in all periods for the classical wavelet transform, and the computing effect was improved, which makes sense that the real-time fault detection is effective. Secondly, aimed at the problem that it is difficult to identify the fault by using time-variant signals, an improved dynamic recurrent neural network （IDRNN） was utilized to identify the fault intelligently after detecting the fault. Finally, the scheme, including fault detection based on the sliding-window Mallat wavelet and fault isolation based on the optimized IDRNN, was applied into a satellite attitude control simulation platform to verify the online diagnosis result. Experimental results show that the sliding-window Mallat wavelet fast transform is consistent with the classical wavelet transform in real-time scenarios, IDRNN has a better generalization ability for identifying time-variant signals, and the scheme including the sliding-window Mallat wavelet and IDRNN can implement detecting the faults and classifying the multiple faults based on real-time monitoring signals for the complex system.