Abstract: In the context of industrial big data,it is important to utilize multiple sensors deployed in complex equipment to obtain condition monitoring data for remaining useful life prediction to ensure the reliability and safety of the system.A relative density high-dimensional kernel estimation remaining useful life prediction method based on multi-sensor data fusion was proposed to fuse multi-sensor monitoring data more effectively.On the basis of one-dimensional kernel density estimation,a nonparametric high-dimensional kernel estimation features fusion model based on multi-sensor data fusion was established.The idea of k-nearest-neighbor relative density was introduced into the selection of the window width of high-dimensional kernel estimation,which could adaptively select a more reasonable window width according to the sparsity of data.A relative density high-dimensional kernel diffeomorphism transformation method was established using a spatial mapping approach to solve the boundary shift problem of high-dimensional kernel estimation in prediction,which could improve prediction accuracy.The accuracy and effectiveness of the proposed method were verified through experimental analysis of the aero-engine dataset and the gearbox dataset.

Key words: remaining useful life prediction, high-dimensional kernel estimation, feature fusion, relative density, kernel diffeomorphism transformation

摘要： 在工业大数据的时代背景下,利用复杂系统中部署的多种传感器获取状态监测数据进行剩余寿命预测,对于保证系统的可靠性与安全性有着重要意义。为了更有效地融合多传感器监测数据,提出一种基于传感器数据融合的相对密度高维核估计剩余寿命预测方法。首先,建立基于多传感器数据融合的非参数高维核估计特征融合模型,并将K近邻相对密度的思想引入高维核估计窗宽的选择中,可以根据数据的稀疏性自适应地选择更合理的窗宽;其次,在剩余寿命预测模型上利用空间映射的方法建立自适应相对密度高维核微分同胚变换方法以解决高维核估计在预测中的边界偏移问题,从而提高剩余寿命预测结果的准确性。最后,分别对航空发动机数据集和齿轮箱数据集进行了实验分析,验证了所提方法的准确性和有效性。

关键词: 剩余寿命预测, 高维核估计, 特征融合, 相对密度, 核微分同胚变换