数据驱动的多元性能退化特征产品可靠性建模

Reliability Modeling of Products with Multiple Performance Degradation Characteristics Based on Data-driven

为解决多元性能退化特征产品可靠性建模困难,提出了一种通用的可靠性建模方法。首先,基于产品多元退化数据,运用ReliefF算法提取产品关键退化特征;然后,通过满窗宽矩阵的多元核密度方法,估计产品在各检测时刻的可靠度;最后,利用各检测时刻可靠度估计结果,拟合产品寿命分布。案例结果表明,该方法能够克服传统参数方法需对产品退化特征量分布作出具体假设的局限,并且能够获得比文献方法精度更高的产品可靠度估计值。另外,该方法可以为后续产品的可靠性试验减少检测工作量,节约试验成本。

The reliability of many products is usually assessed by multiple performance degradation characteristics. To address this concern, a general reliability modeling approach was proposed. Firstly, the ReliefF algorithm was applied to extract the key degradation features of the product based on the multivariate degradation data. Then,the reliability of the product at each testing moment was estimated by the multivariate kernel density method with a full window width matrix.Finally, the reliability estimation results at each testing moment were used to fit the product lifetime distribution. The case results show that this method can overcome the limitations of traditional parametric methods which require specific assumptions about the distribution of product degradation characteristics and can obtain product reliability estimates with higher accuracy than literature methods. In addition, the method can also reduce the testing workload and save the testing cost for the subsequent product reliability test.