齿轮故障声信号特征数据集具有高维和非线性特性,等距映射(ISOMAP)降维算法通过构造距离矩阵来测量样本点之间的测地距离,具有处理复杂非线性数据的能力,但其本身是一种无监督算法,不能有效利用样本点间的标签信息。文中设计了一种结合有监督等距映射(S-ISOMAP)算法和有向无环图支持向量机(DAG-SVM)的故障诊断方法,主要包括特征提取、降维和模式识别三个部分。利用梅尔频率倒谱系数(MFCC)提取齿轮故障声信号的特征信息,建立高维特征数据集,在计算欧式距离时引入调节因子,构建有监督的S-ISOMAP 降维算法对高维MFCC 特征数据集进行降维。引入有向无环图,构建DAG-SVM 分类器实现多分类。实验结果表明,该方法能有效准确的识别出旋转机齿轮的故障状态,识别准确率达到94. 67??, S-ISOMAP 相较ISOAMP、局部线性嵌入的降维效果更好,分类识别准确率更高。

The gear fault acoustic signal characteristic data has high-dimensional and nonlinear characteristics. ISOMAP dimensionality reduction algorithm can measure the geodesic distance between sample points by constructing distance matrix, which has the ability to deal with complex nonlinear data structure. However, the label information between sample points cannot be effectively utilized due to the nonsupervision nature of ISOMAP. In this paper, a supervised ISOMAP algorithm and a directed acyclic graph support vector machine (DAG-SVM) fault diagnosis method are proposed, which includes feature extraction, reduction and pattern recognition. Mel frequency Cepheid coefficient (MFCC) is used to extract the characteristic information of gear fault signal, and a high dimensional feature data set is established. A supervised ISOMAP dimensionality reduction algorithm is introduced in the calculation of Euclid distance to reduce the dimensionality of high-dimensional MFCC feature data set. A directed-acyclic-graph is introduced to construct a DAC-SVM classifier to realize multi-classification. The experimental results show that this method can effectively and accurately identify the fault state of the gear of the rotary machine, and the recognition accuracy reaches 94. 67%. S-ISOMAP has better dimension reduction effect and higher classification recognition accuracy than ISOAMP and local linear embedding.

TP391

国家自然科学基金资助项目(61571376)