基于迁移学习的综合传动装置健康预测方法

针对某型步战车综合传动装置变工况的工作特点,提出一种基于迁移学习的健康预测方法.采用灰色关联分析(grey relational analysis,GRA)提取源域和目标域内时序退化特征作为各部件健康指标(health indicator,HI),构建1维时序健康指标,通过动态时间规整(dynamic time warping,DTW)运算得到目标域特征与健康指标的关联度,提取源域与目标域的公共退化信息,构建面向健康预测的支持向量回归(support vector regression,SVR)模型进行健康预测,并以传动装置的变速结构为例进行验证.结果表明:基于迁移学习的健康预测结果更贴合实际健康变化趋势,有助于维修人员更为准确地判断传动装置的健康状态.

Health Prediction Method of Comprehensive Transmission Based on Transfer Learning

Aiming at the working characteristics of variable conditions of an integrated transmission device of a type of IFV, a health prediction method based on transfer learning is proposed. Use the grey relational analysis to extract the time series degradation features in the source and target domains as health indicators of each component, construct a one-dimensional series of health indicators, through the dynamic time warping operation, the correlation between the characteristics of the target domain and the health indicators is obtained, extract the public degradation information of the source and target domains, construct a support vector regression model for health prediction for health prediction, and take the transmission structure of the transmission device as an example to verify. The results show that the health prediction results based on transfer learning are more in line with actual health trends, and help the maintenance staff to more accurately judge the health status of the transmission device.