Development of fault diagnosis platform for liquid rocket engine based on experimental data

Aiming at the defects of difficult to establish fault diagnosis models caused by the small scale of liquid rocket engine test data, imbalanced categories and high fault coupling, the fault diagnosis platform with 10 fault diagnosis models are developed for fault diagnosis, covering K-nearest neighbor (KNN) model, optimized KNN model, logistic regression (LR) model, optimized LR model, support vector machine model, K-means model, decision tree model, neural network model, random forest model and light-GBM model. The prediction accuracies of these models are validated based on the experimental data. Among these models, the light-GBM model provide the best prediction accuracies, and 5 models have the prediction accuracy larger than 98%.