Research On Fault Diagnosis And Early Warning Based On Data Driven

Recent years,with the improve of modern industrial production ability,people give the increasing concern about the safety of production process and the reliability of product quality.Data driven fault diagnosis and early warning technology has become the core technology to improve safety,reliability and work efficiency in industrial process.Point at the data-driven fault diagnosis and early warning methods,this paper takes the control system actuator as the object.The main work results are summarized as follows:(1)Based on the principle of Time Convolutional Network(TCN),combined with the timing feature of actuator signals,a TCN actuator state prediction method was proposed.Taking the electro-hydraulic actuator of gas turbine control system actuator failure semi-physical test platform as an example,the data of normal state was collected,and the state prediction model was established using RNN and TCN respectively,and the accuracy of two prediction methods was compared.The results show that the method has higher accuracy than the traditional RNN network;(2)Based on the principle of Long Short-Term Memory Network(LSTM),considering the correlation feature between control system actuator signals,an LSTM fault classification model was established,and a fault diagnosis method of multi-channel residual signals based on LSTM was proposed.The fault test of electro-hydraulic actuator in the actuator fault semi-physical test platform shows that the proposed actuator fault classification model and method have high accuracy;(3)The mechanism analysis were carried out for the gradual failures that are common and difficult to detect in control system actuators.According to the assumption of failure mechanism consistency under extreme working conditions,two typical gradual failures of electro-hydraulic actuator were simulated.Combined with the TCN state prediction model and LSTM regression model,an online early warning test was carried out,it was verified that the early warning method has high accuracy rate,shortens the warning time,and reduces the false alarm rate.