融合失效样本与截尾样本的滚动轴承寿命预测

Life prediction for rolling bearings utilizing both failure and truncated samples

针对常规寿命预测方法依赖于失效样本、无法有效利用截尾样本的局限性,提出一种融合失效样本和截尾样本的滚动轴承寿命预测方法。基于函数型主成分分析方法对反映轴承退化的特征量建立趋势模型,将各特征量分解为均值、特征向量和主成分得分向量;通过最小化截尾样本与失效样本主成分得分向量间的相似性指标估计各截尾样本最优寿命值;基于特征量趋势模型估计各样本全寿命阶段内特征值,生成训练数据;采用最小二乘支持向量机建立预测模型用于轴承寿命估计。滚动轴承寿命预测试验表明该方法能利用截尾样本提高寿命预测精度,且对一定程度的数据缺失具有鲁棒性。

To overcome the limitations that the traditional bearing life prediction method relies on a database of failure samples and it cannot effectively utilize truncated samples,an intelligent method utilizing both failure and truncated samples was proposed for bearing life prediction. Firstly,the trend model for features characterizing bearing degradation was constructed based on the function principal component analysis（ FPCA）,and each feature was decomposed into a mean value,an eigenvector and a score vector of function principal components（ FPC-scores）. Secondly,the optimal life value of each truncated sample was estimated by minimizing the similarity index between its score vector and those of failure ones. Thirdly,all features in the whole life duration of each sample were estimated and reconstructed based on the feature trend model to generate training data. Finally,the prediction model was constructed based on a least square support vector machine for bearing life prediction. The test results of rolling bearings＇ life prediction showed that the proposed method can improve the bearing life prediction accuracy with truncated samples,and it is robust to a certain level data missing.