基于功耗残差的航天器CMG退化特征提取方法

为实现航天器控制力矩陀螺（CMG）性能退化状态评估，提出了一种基于卷积神经网络（CNN）与功耗残差的CMG退化特征提取方法。由于CMG控制系统对高速转子运动状态的高精准控制，CMG退化特征难以从转子运动状态数据中直接提取。针对该问题，从转子系统的能量损耗角度出发，通过分析CMG工作机理确定了影响单位时间内转子电机功耗的变量，并通过CNN建立了CMG运行状态参数与电机功耗之间的映射。将退化状态下电机实际功耗与模型输出的残差作为退化特征对CMG退化状态进行评价。通过某型号CMG的加速寿命实验数据进行验证，结果表明:构建的退化特征能够表征CMG转子轴承的性能退化情况，从而为CMG状态监测和故障预警提供参考。 

Degradation indicator extraction for aerospace CMG based on power consumption analysis

Control moment gyro (CMG) is the actuator for the attitude control of large spacecraft. In order to evaluate the performance degradation state of CMG, a convolutional neural network (CNN) and residual power consumption-based degradation feature extraction method is proposed. The high-precision control of the CMG control system makes it difficult to extract degradation features from the operational state of the CMG rotor. To solve this problem, a CNN model is introduced to establish the mapping between CMG operating state parameters and motor power consumption, and the degradation feature is defined as the residual error between the model output and actual power consumption of the motor in the degraded state. For approach validation, an accelerated life test dataset of a real CMG was used. The results show that the constructed degradation feature can reflect the performance degradation of the CMG rotor bearing.