多退化变量下基于实时健康度的相似性寿命预测方法

Multi-index modeling for similarity-based residual life estimation based on real-time health degree

鉴于目前对多退化变量下的相似性寿命预测方法的研究较少,且所用的建模方法仅限于线性回归、较为单一,提出多退化变量下基于实时健康度的相似性寿命预测方法。该方法先采用主成分分析法、支持向量数据描述法、马氏距离和负向转换函数等将多退化变量融合为能反映系统退化状态的定量指标——实时健康度;依据设备的实时健康度,采用面向单退化变量的相似性寿命预测方法预测设备的剩余寿命;通过陀螺仪剩余寿命预测的实例对该方法进行验证分析。研究结果表明,该方法可行并具有一定的优越性,能提供统计意义上更精确的剩余寿命预测结果(即更小的预测误差)。

Since the research on multi-index modeling for the similarity-based residual life estimation were rare and modeling methods were limited to linear regression and thus relatively single,the multi-index modeling method for similarity-based residual life estimation based on real-time health degree was proposed.The principal component analysis,support vector data description,Markov distance and negative conversion function were all hired to fuse multiple degradation variables into a quantitative index namely real-time health degree which could reflect the system degradation state;based on the equipment real-time health degree,the equipment residual life could be predicted by using the similarity-based residual life prediction method with single degradation variable;the recommended method was validated through the case analysis of gyroscope residual life prediction.Results showed that the suggested method was feasible and had certain superiority in terms of statistically more accurate prediction（i.e.smaller prediction error）.