基于机器学习的在轨航天器泄漏源实时定位方法研究

Real time location method for leakage sources in orbiting spacecraft based on machine learning

该文章针对航天器在轨泄漏问题,提出了一种基于机器学习的泄漏源实时定位方法。该方法通过合理特征化泄漏在器壁中激发的弹性波信号,结合有限元仿真技术获得有效的训练数据,设计并实现了一种多层感知机网络模型,从而准确完成泄漏源与传感器间的距离信息估计,同时结合计算弹性波数据时空相关性得到的相对角度信息,可以快速稳定的获得泄漏源的空间位置。该方法仅采用一个布放在器壁上的压电阵列式传感器采集泄漏激发的弹性波数据,结构相对简单。试验结果表明,基于该文章设计的多层感知机模型,在1 m2试验板范围内该方法对泄漏源与阵列式传感器距离估计准确率为100%,最大定位误差为1.2 cm。

Here,a machine learning-based real-time location method for leakage sources of spacecraft in orbit was proposed.This method could obtain effective training data by reasonably characterizing elastic wave signals excited by leakages in vessel wall and combining with finite element simulation technique.A multi-layer perceptron(MLP)network model was designed and implemented to correctly complete estimation for distance information between per leakage source and sensor.Meanwhile,combing relative angle information obtained by calculating the time-space correlation of elastic wave data,spatial positions of leakage sources could be obtained quickly and stably.In this method,only one piezoelectric array sensor placed on vessel wall was used to collect elastic wave data excited by leakages,the structure was relatively simple.The experimental results showed that based on the MLP network model designed here,the accuracy of the distance estimation between per leakage source and array sensor is 100%within the range of 1 m 2 test board,and the maximum location error is 1.2 cm.