基于多SVM误差加权的轴承剩余寿命预测

Rolling bearing residual life prediction based on error weighted multiple improved support vector machine model

针对传统的以概率统计为基础的轴承寿命预测方法需要大量的轴承寿命试验数据,且运行周期较长,难以满足特定工况下轴承寿命状态预测的问题。通过多SVM的预测误差进行加权获得轴承的剩余寿命。首先采集多个轴承的全过程寿命振动数据,针对获取的多个轴承的振动数据进行特性提取并利用主成分分析进行特性约简,实现对轴承运行寿命评估特性指标的建立,并利用粒子群算法改进的多个支持向量机模型进行剩余寿命预测建模,利用所建立的多个模型对要预测的轴承特征数据进行预测,并通过各个模型的误差权重来实现轴承剩余寿命的准确预测。通过实验分析验证了模型的有效性。

Aiming at the problem that the traditional probability statistics based bearing life prediction method need a large number of bearing whole life test data, which will cost a long periods of running time, which is difficult to satisfy, especially for the life prediction of bearing working under specific conditions. The re- search proposed using the error weighting multi-SVM prediction method to obtain the bearing residual life. Firstly, the whole process life vibration data of multiple bearings are collected. Then, the character- istic feature data of the vibration signal about the multiple running bearings are extracted and dimension reduced by using the principal component analysis method to realize the establishment of the bearing life evaluation index. Finally, a multi- support vector machine model improved by particle swarm optimization is established and used to achieve the residual life prediction. The bearing life characteristic data to be predicted is forecasted by using the established models, the error of each SVM model is got, and the residual weight of the bearing is predicted by using the error weight. The proposed model is veri- fied effectively by some experimental analysis.