开展了航天器流体回路泵智能故障诊断方法研究,将 BP 神经网络、粒子群优化(PSO)-BP神经网络、遗传算法(GA)-BP 神经网络和模糊神经网络 4 种模型应用于故障诊断,以温度、流量、出口压力、转速 4 类状态参数作为神经网络输入,状态标志作为输出进行训练,通过均方误差、相关系数、模型训练评分对模型训练效果评价,从而实现模型优选,完成故障诊断.利用在轨遥测数据进行应用验证,结果表明可以准确识别流体回路泵的正常和叶轮卡死泵功能丧失两种在轨实际状态类型,且模糊神经网络相对其他 3 种神经网络具有更好的诊断效果.
Intelligent Fault Diagnosis Method for Fluid Loop Pumps in Spacecraft
A study is conduted on the research of intelligent fault diagnosis methods for spacecraft fluid loop pump in this paper.Four models,namely BP neural network,Particle Swarm Optimi-zation(PSO)-BP neural network,Genetic Algorithm(GA)-BP neural network,and fuzzy neural network,are applied to fault diagnosis.These models utilize temperature,flow rate,outlet pres-sure,and rotational speed as inputs to the neural network,,with state labels as outputs for train-ing.The training effectiveness of the models is evaluated through mean square error,correlation coefficient and model training score,thereby achieving model optimization and completing fault diagnosis.The application validation of the proposed method is performed using on-orbit teleme-try data,demonstrating its ability to accurately identify normal operation and impeller jam pump function loss of the fluid loop pump,indicating that the fuzzy neural network has better diagnos-tic performance compared to the other three neural networks in terms of diagnostic accuracy.
国家科技期刊平台
NSTL
万方数据
