Research On Fault Prediction And Remaining Useful Life Estimation Of Train Equalizing Reservoir Control System

The train braking system is used to realize the braking and releasing of the train,which is very import to ensure the safe operation of the train.The equalizing reservoir control system is the key functional module of the braking system.Due to the increase of frequency of service and running time,the components of the equalizing reservoir control system will degenerate and cause system failure,which seriously threatens the running safety of the train.In order to improve the reliability of the braking system,it is necessary to predict the fault of the equalizing reservoir control system and estimate its remaining useful life.However,the characteristics of the equalizing reservoir control system with many components and complex degradation process bring great challenges to the realization of the above goals.To this end,this paper proposes a fault prediction method based on long short-term memory neural network and a remaining useful life estimation method based on transfer learning.The main research work is as follows:Firstly,in order to address the issue that the features in system fault prediction are difficult to extract and the prediction results are difficult to explain,a fault prediction method based on long short-term memory neural network is proposed.The working principle of the equalizing reservoir control system is fully studied,and the key features that characterize the health of the system are extracted from the current curve and the pressure curve respectively.The long short-term memory neural network is designed,the mapping relationship between features and potential faults is mined from time series data,and the layer-wise relevance propagation method is introduced to explain the correlation between features and prediction results.Then,due to the insufficient data labels for the actual operation of the train,a transfer learning-based remaining useful life estimation method is proposed.Taking train operation data as the target domain dataset,multiple source domain datasets of different operating conditions are generated through the test bench,and a source domain selection strategy based on information divergence is proposed to select the source domain closest to the target domain.Using the instance-based adaptive boosting transfer learning algorithm,by continuously adjusting the sample weights of the source domain and the target domain,multiple support vector regression models are trained and an integrated estimation model is generated,so as to use the life degradation knowledge in the source domain data to assist the task of estimating the remaining useful life of the target domain.Finally,the proposed fault prediction method is proved not only achieves high accuracy but also has good interpretability based on the actual train operation data.At the same time,the validity of the proposed remaining useful life estimation method is verified based on the data of the built test bench of the equalizing reservoir control system.