Fault diagnosis method for liquid rocket engines based on LSTM-LightGBM

Abstract

This paper proposes a fault diagnosis method for liquid rocket engines (LREs) that integrates physical simulation with a hybrid deep learning algorithm, aiming to address the scarcity of fault samples and the strong temporal coupling of multi-sensor data in practical engineering. The study first establishes a nonlinear dynamic mathematical model of the LRE, covering turbopumps, combustion chambers, and valve pipelines, based on a modular modeling approach. A simulation dataset containing normal and various fault modes is generated through fault factor injection. On this basis, a hybrid diagnostic model is constructed by serially connecting Long Short-Term Memory (LSTM) networks and Light Gradient Boosting Machine (LightGBM) to extract deep dynamic features from multivariate time series and achieve efficient nonlinear classification. Experimental results show that the proposed method achieves a fault diagnosis accuracy of 99.31%, significantly outperforming traditional SVM and LSTM networks.