基于特征处理的MVU算法在齿轮故障诊断中的应用

Application of MVU algorithm based on feature processing in gear fault diagnosis

针对不同故障类别齿轮的故障信息难以有效获取、齿面多类故障难以准确聚类的问题,提出一种基于特征处理的最大方差展开(Maximum Variance Unfolding,MVU)维数简约的齿轮故障诊断模型。首先对获取的振动信号进行最小熵反卷积(Minimum Entropy Deconvolution,MED)预处理,将高低频段进行分离并筛除不确定信号,并在多域上提取信息熵作为特征指标;而后,利用样本点分布矩阵筛选高效表征特征指标并构建高维特征空间,并利用改进的MVU算法对其进行维数简约,获取低维的真实子空间;最后,将其输入到超球多类支持向量机中进行超球构造与分类识别。通过实验数据的分析对比验证模型的有效性。

Aiming at problems of gear fault information being difficult to effectively extract and gear surface faults being difficult to correctly cluster,a gear fault diagnosis model using the maximum variance unfolding(MVU)algorithm based on feature processing was proposed here.Firstly,the collected faulty gear vibration signals were pre-processed with the minimum entropy deconvolution(MED),higher and lower frequency bands were separated and uncertain signals were excluded.The information entropy was extracted in multi-domain as feature indexes.Then efficient feature indexes were screened out with the sample point distribution matrix and they were used to construct high dimensional feature space,and the improved MVU algorithm was used to do dimension reduction,and obtain an actual subspace with lower dimension.Finally,the actual subspace was input into a hyper-sphere multi-class SVM to do hyper-sphere construction and classification identification.The effectiveness of the proposed model was verified through contrastively analyzing test data.