基于深度学习及GPU计算的航天器故障检测技术

由于航天器在高温、高压等恶劣环境中工作，采用传统故障检测方法自主性相对较差，缺少对故障特征的分析，导致检测精准度较低。提出了基于深度学习及GPU计算的航天器故障检测技术，依据航天器故障信号特征分析与检测原理，在GPU计算技术支持下，获取GPU图像，并在深度置信网络模型中引入该计算方法。根据构建的深度置信网络模型，预测轴承故障位置，经过GPU计算技术下提取的故障特征用于深度置信网络故障预测基本数据，将原始进行归一化处理，分析航天器轴承故障特征，并在不同参数支持下，利用深度学习算法自动确定网络关键参数，由此识别轴承故障，并学习故障特征，实现航天器故障检测。由实验结果可知，该技术检测精准度最高可达到98%，具有较强鲁棒性。

Spacecraft Fault Detection Technology Based on Deep Learning and GPU Computing

Because the spacecraft works in the harsh environment such as high temperature and high pressure, the autonomy of traditional fault detection methods is relatively poor, and the lack of analysis of fault characteristics leads to low detection accuracy. A spacecraft fault detection technology based on deep learning and GPU calculation is proposed. According to the principle of analysis and detection of spacecraft fault signal characteristics, the GPU image is obtained with the support of GPU computing technology, and the calculation method is introduced into the depth confidence network model. According to the built deep confidence network model, the fault location of bearings is predicted. The fault features extracted by GPU computing technology are used for the basic data of deep confidence network fault prediction. The original data are normalized and the fault characteristics of spacecraft bearings are analyzed. With the support of different parameters, the deep learning algorithm is used to automatically confirm the fault location of bearings. By defining the key parameters of the network, bearing faults can be identified, and fault features can be learned to realize spacecraft fault detection. The experimental results show that the detection accuracy of this technology can reach 98%, and it has strong robustness.