基于LSTM的风洞设备健康状态评估方法研究

The study of health assessment for wind tunnel based on LSTM

摘要:

针对风洞设备健康状态评估中特征提取困难、量化算法复杂等问题,提出了一种基于深度学习的健康度评估方法,利用正常状态样本数据训练LSTM编解码器网络并构建特征空间,利用测量数据特征向量与特征空间的欧氏距离衡量健康状态的退化程度,从而高效地实现了系统或设备的健康状态量化评估；经风洞试验室轴流风机转子不平衡故障、长轴轴承裂缝故障等两个数据集进行验证,取得了与设备工作状态一致的健康度评估值,具有很强的工程应用价值。

Abstract:

This paper presents a health assessment method for wind tunnel based on deep learning networks, which utilizes the LSTM encoder-decoder to build up the normal status features space via computing the hidden status of examples data. Meanwhile, the quantitative value of the diversity between the measurement data and example data is calculated using Euclidean distance from feature vector to the normal status feature space, and then hundred-mark health index can be specified. As result, health assessment method achieves high accuracy in evaluation of two fault data sets of wind tunnel, which show a significant effect and strong application value.