摘要
液体火箭发动机健康检测技术是提高火箭安全性和可靠性的重要技术之一,对其进行研究具有重要的学术和工程应用价值。目前的健康检测方法大多基于特征提取和专家经验,智能检测技术水平急需提高。提出了一种基于卷积自编码器的液体火箭发动机健康状态智能检测方法,对发动机多传感器监测数据进行无监督的特征提取和重构,完成对训练集的学习,并基于隐含层特征和重构误差确定健康状态的中心和监测阈值,然后对测试集各机组状态进行识别,通过对各部件的损失计算可以进一步确定各部件的健康状态。提出方法通过热试车实验数据进行了验证,该方法在测试集上取得了88.9%的准确率,并能够监测整机及各部件的健康状态和退化趋势,结果表明提出方法的有效性和应用潜力。
The health state detection technology of liquid rocket engine is very important to improve the safety and reliability of rockets,which has important academic research and application value.The current research methods are mostly based on feature extraction and expert experience,while the research on intelligent detection technology needs to be improved.In this paper,Multi-component Two-stage Fusion Convolutional Denoising Auto-Encoder(MTAE)for intelligent detection of liquid rocket engine health state was proposed.First,unsupervised feature extraction and reconstruction of multi-sensor monitoring data were performed.After completing the learning of training set with the MTAE,the center of health and the monitoring threshold were determined based on the hidden layer features and reconstruction errors.Then,the state of each group of the testing set would be identified.In addition,the health state of each component could be further determined by calculating the loss of each component.Finally,the proposed method was verified by the data collected from the static firing test process.The method achieves an accuracy of 88.9%on the test set and could monitor the health and degradation trends of the whole engine and each component.The results show the effectiveness and application potential of the proposed method.
作者
王珺
吕海鑫
陈景龙
刘子俊
袁军社
WANG Jun;LV Haixin;CHEN Jinglong;LIU Zijun;YUAN Junshe(Xi'an Aerospace Propulsion Institute,Xi'an 710100,China;State Key Laboratory of Manufacturing and Systems Engineering,Xi'an Jiaotong University,Xi'an 710049,China)
出处
《火箭推进》
CAS
2021年第4期52-58,共7页
Journal of Rocket Propulsion
关键词
健康检测
液体火箭发动机
多传感器数据
卷积自编码器
无监督学习
health state detection
liquid rocket engine
multi-sensor data
convolutional autoencoder
unsupervised learning
