基于MD-CUSUM和TD-SVR的滚动轴承健康状态预测

Health state prediction of rolling bearings based on MD-CUSUM and TD-SVR

故障特征提取是轴承健康状态描述的关键,然而当前常用方法提取的特征往往维数较高或信息缺失,无法单调性地反应轴承健康状态,且预测结果不能有效反应轴承退化趋势。应用累积马氏距离(MD-CUSUM)实现特征降维并得到健康指标(HI),能够在低维层面上单调性地反应轴承健康状态;构建时滞性支持向量回归(TD-SVR)模型,提高滚动轴承健康状态预测精度。通过试验数据分析对比了MD-CUSUM与等距特征映射(ISOMAP)的优劣,结果表明MDCUSUM和TD-SVR相结合在轴承健康状态预测方面具有更好地效果。

Fault feature extraction is the key of bearings’health state description,however,features extracted with currently common methods may have higher dimensions or they are lack of information and unable to reflect bearings’health state monotonously,and the predicted results cannot effectively reveal the degradation trend of bearings.Here,the cumulative sum and Mahalanobis distance(MD-CUSUM)was adopted to realize feature dimension reduction,obtain the health index(HI),and reflect monotonously health state of bearings with lower dimensions.Furthermore,a time delayed support vector regression(TD-SVR)model was constructed to improve the prediction accuracy for health state of rolling bearings.The advantages and disadvantages of MD-CUSUM and those of the isometric feature mapping were compared through test data analysis.The results showed that the combination of MD-CUSUM and TD-SVR has a better effect on predicting health state of bearings.