基于多元时间序列分析的控制系统执行器故障诊断方法研究

为提高控制系统执行器故障实时诊断的准确率,该文提出一种基于多元时间序列分析的控制系统执行器在线故障诊断方法。首先分析了控制系统执行器故障机理,确定了表征执行器故障的关键信号;其次采用执行器历史数据,建立了时间卷积网络(TCN)在线预测模型,对执行器多通道信号进行在线预测;随后通过长短期记忆网络(LSTM)对多通道残差信号建立了故障分类模型;最后以燃气轮机控制系统执行器半物理试验平台中的电液执行器为例进行了多次重复试验验证。结果表明,基于TCN网络的在线预测模型相比传统循环神经网络(RNN)预测误差较小;基于LSTM网络的故障分类模型准确率较高;通过LSTM网络对多通道残差信号进行故障分类,比对原始故障数据分类故障准确率更高。

Research on Actuator Fault Diagnosis Method of Control System Based on Multivariate Time Series Analysis

In order to improve the accuracy of real-time fault diagnosis of control system actuator,an on-line fault diagnosis method of control system actuator based on multivariate time series analysis was proposed. Firstly,the actuator fault mechanism of the control system was analyzed,and the key signals representing the actuator fault were determined;Secondly,the temporal convolutional network(TCN) on-line prediction model was established by using the actuator historical data to predict the multi-channel signal of the actuator on-line;Then,the fault classification model of multi-channel residual signal was established by long short-term memory network(LSTM);Finally,the electro-hydraulic actuator in the semi physical test platform of gas turbine control system was taken as an example for repeated to verify the test. Results show that the online prediction model based on TCN has less prediction error than the traditional recurrent neural network(RNN);The fault classification model based on LSTM has high accuracy;The fault classification modeling of multi-channel residual signal through LSTM network has higher accuracy than the original fault data modeling.