A Health Monitoring Model for the Intake and Exhaust System of Locomotive Diesel Engine

The health monitoring for the intake and exhaust system of locomotive diesel engine is very important for railway safety and maintenance cost. The traditional model is difficult to deal with the multi-parameter dynamic correlation under complex working conditions, resulting in insufficient prediction accuracy. In this paper, a monitoring model based on graph self-learning and Huber-MTGNN is proposed. The sparse adjacency matrix between sensor parameters is dynamically constructed by the adaptive graph learning layer, and the spatial dependence is mined by the graph convolution module. The temporal convolution module extracts time features, and introduces the Huber loss function to suppress outliers. Experiments based on real data sets show that compared with Graph Attention Networks and Graph Convolutional Networks, the average absolute error of the model is reduced by 16 % and 26 %, the root mean square error is reduced by 16 % and 31 %, and the average absolute percentage error is reduced by 4 % and 6 %, respectively. The high precision and robustness under dynamic conditions are verified, which provides an effective solution for locomotive diesel engine health monitoring under unknown graph structure.