Aiming at the problem of imbalanced fault samples observed in practical engineering,a classification method called multitask game probability classification vector machine(MGPCVM)was proposed based on sparse Bayesian theory and fuzzy membership degree theory.Firstly,in the objective function of MGPCVM,a game factor was designed to assign each sample a specific sensitivity value based on the game information between the centroids of different classes.This was done to address the poor classification performance of traditional classifiers on imbalanced datasets.Secondly,in the Bayesian framework theory,a truncated Gaussian prior distribution was employed to achieve consistency between the signs of sample parameters and their corresponding label information,and to generate sparse estimation of centroid sensitivity values.Finally,the MGPCVM method was applied to validate the effectiveness of fault diagnosis using rolling bearing experimental data collected from two different experimental platforms.The research results indicate that,under different imbalance ratios(IR),the accuracy of the MGPCVM method remaines above 95%,which showes a 4%to 8%improvement compared to support vector machines(SVM),probabilistic classification vector machines(PCVM),and other methods.These results demonstrate that,in comparison with typical vector-based classification methods,the MGPCVM method exhibites superior classification performance under imbalanced data conditions,making it suitable for classification problems with imbalanced data in practical operating conditions.
关键词
滚动轴承/故障诊断/多任务博弈概率分类向量机/支持向量机/概率分类向量机/不均衡比/故障分类模型
Key words
rolling bearing/fault diagnosis/multitask game probabilistic classification vector machine(MGPCVM)/support vector machine(SVM)/probabilistic classification vector machine(PCVM)/imbalance ratios(IR)/fault classification model
维普
万方数据