基于注意力残差网络的航天器测控系统故障诊断

随着航天器数量的不断增加,快速而准确地对航天器测控系统进行故障诊断尤为重要。 针对航天器所处空间环境变化 较大、遥测数据成分复杂和故障诊断准确率不高的问题,提出了一种基于注意力残差网络(AM-ResNet)的航天器测控系统故障 诊断方法。 首先,将原始遥测数据转换成灰度图像;其次,将图像依次通过残差网络和注意力模块,获取具有全局依赖关系的特 征图;最后经过卷积、池化操作后利用 Softmax 分类器进行分类,实现航天器测控系统的故障诊断。 实验结果表明,所提出的基 于注意力残差网络的航天器测控系统故障诊断方法可将诊断准确率提升至 95. 68% ,与 ResNet-18、AlexNet 和 LeNet-5 故障诊断 模型相比,诊断准确率分别提高了 3. 53% 、5. 62% 和 16. 43% ,验证了该方法可以有效提高航天器测控系统故障诊断性能。 关键词: 深度学习;故障诊断;残差网络;航天器;注意力机制

Fault diagnosis method of spacecraft tracking telemetry and control system based on the attention residual network

As the number of spacecrafts increasing, it is particularly important to diagnose the fault of spacecraft tracking telemetry and control (TT&C) system quickly and accurately. To address the problems of large changes in the space environment, complex telemetry data components and low accuracy of fault diagnosis, a fault diagnosis method of spacecraft TT&C system based on the attention mechanism residual network (AM-ResNet) is proposed. Firstly, the telemetry data are converted into grayscale image. Secondly, the image is passed through the residual network (ResNet) and attention module to obtain feature map with global dependence. Finally, the softmax classifier is used to achieve image classification after convolution and pooling operations to realize the fault diagnosis of spacecraft TT&C system. Experimental results show that the fault diagnosis method of spacecraft TT&C system based on the proposed AM-ResNet can improve the accuracy of fault diagnosis to be 95. 68% . Compared with ResNet-18, AlexNet and LeNet-5 fault diagnosis models, the diagnostic accuracy is increased by 3. 53% , 5. 62% and 16. 43% , respectively, which prove that the method can effectively improve the fault diagnosis performance of the spacecraft TT & C system.