Intelligent Fault Diagnosis Method for Spacecraft Fluid Loop Pumps Based on Multi-Neural Network Fusion Model

Fluid loop pumps, critical to spacecraft thermal control subsystems, are more prone to failures than other spacecraft components. Timely fault diagnosis is therefore crucial to ensure operational reliability. This paper proposes a multi-neural network fusion model (MNN) to improve the fault diagnosis accuracy for spacecraft fluid loop pumps. The proposed model integrates four neural network algorithms—back propagation neural network (BPNN), particle swarm optimization-back propagation neural network (PSO-BPNN), genetic algorithm-back propagation neural network (GA-BPNN), and fuzzy neural network (FNN)—through a model scoring and weighting mechanism. Additionally, a dedicated software has also been developed and implemented for the intelligent fault diagnosis of fluid loop pumps in Chinese spacecraft. By analyzing a dataset derived from on-orbit telemetry and expert knowledge, the proposed model demonstrates superior performance over individual models, achieving significant improvements in key metrics such as mean squared error (MSE), prediction stability, correlation coefficient (R), Accuracy, Precision, Recall, and F1-score. Furthermore, validation using both on-orbit telemetry data and ground test data confirms that the model can accurately diagnose both normal operations and various types of faults, making it a reliable and practical tool for on-orbit fault detection. The study provides an efficient, stable, and practical solution for intelligent fault diagnosis of spacecraft fluid loop pumps, with significant engineering application value.