模糊支持张量训练机及其在滚动轴承故障诊断中的应用

Fuzzy support tensor train machine and its application in rolling bearing fault diagnosis

在进行样本不平衡数据建模时,支持向量机(SVM)无法保护故障信号的特征信息,针对这一问题,提出了一种基于模糊支持张量训练机(FSTTM)的状态评估方法。首先,利用多源故障信号在FSTTM中,构建了张量样本,并在模型中引入了张量训练(TT)分解方法,以提取高阶张量样本中包含的特征信息;然后,利用基于TT核函数,建立了线性不可分下的预测模型,解决了非线性数据的分类问题;最后,在目标函数中设计了模糊因子,使模型对数目较少一类样本及数目较多一类样本的倾向均衡,实现了对样本不平衡数据的有效分类。研究结果表明:采用FSTTM对两种不同的滚动轴承数据进行故障诊断实验,其故障识别准确率均在97%以上,且F-score指标达到0.9800以上;相对于传统支持张量机,FSTTM利用高阶张量和模糊因子构造预测模型,可实现对原始信号状态信息的充分利用和样本不平衡数据的准确分类。

Aiming at the problem that support vector machine(SVM)cannot protect the characteristic information of fault signal when modeling of sample imbalanced data,a fuzzy support tensor train machine(FSTTM)method was proposed in this paper.Firstly,in FSTTM,tensor samples were constructed by using multi-source fault signals,and tensor train(TT)decomposition method was introduced into the model to extract the feature information contained in high-order tensor samples.Then,the prediction model based on TT kernel function was established,which can solve the classification problem of nonlinear data.Finally,a fuzzy factor was designed in the objective function,which can make the tendency equilibrium of the type of sample with less number and the type with more number of samples,and realize the effective classification of sample unbalanced data.Two different roller bearing data were used for experimental analysis of fault diagnosis.The results show that the fault identification accuracy of FSTTM is more than 97%and the Fscore index is more than 0.9800.Compared with the traditional support tensor machine,high-order tensor and fuzzy factor are used FSTTM to construct the prediction model,which can make full use of the original signal state information and accurately classify the sample unbalanced data.