二维非负矩阵分解在齿轮故障诊断中的应用

Application of Two-Dimensional Non-negative Factorization for Gear Fault Diagnosis

针对齿轮故障信号时频分布识别问题,提出采用二维非负矩阵分解技术提取时频分布矩阵特征参数的方法。采用S变换技术将齿轮故障信号变换至时频域,为克服传统的一维非负矩阵分解对矩阵向量化带来的维数过高和结构信息损失问题,提出采用二维非负矩阵分解技术直接对信号时频分布矩阵提取特征参数。对齿轮5种状态下信号时频分布矩阵的特征提取和分类结果表明,二维非负矩阵分解技术无论在计算效率还是分类精度上都明显优于一维非负矩阵分解技术。

A new feature extraction scheme utilizing two-dimensional non-negative matrix factorization (2DNMF) for classification of time-frequency distributions of gear defect signals is presented in this work. The S transform is employed to generate the time-frequency distributions of gear defect signals. The newly developed 2DNMF, which can overcome the high dimension and structural information loss problem of traditional non-negative matrix factorization (1DNMF), is used to extract feature subsets for classifying the time-frequency matrices. The application to the practical gear fault diagnosis has revealed that the 2DNMF demonstrates higher computation efficiency and classification rates compared with the traditional 1DNMF.