基于流形学习和支持向量机的焊缝偏差识别方法

首先对基于旋转电弧传感的焊接电流信号进行小波滤波，预处理后构建样本数据集。然后建立基于支持向量回归机的Laplace特征映射外延算法，并对样本数据集和新样本进行维数约简，利用维数约简后的样本数据集训练支持向量回归机，并对新样本进行偏差识别。与不进行维数约简而是直接利用支持向量回归机进行偏差识别的方法进行对比实验，表明利用Laplace特征映射进行维数约简能提高焊缝偏差识别的精度。

Seam offset identification method based on manifold learning

The welding current based on the rotational arc sensor is filtered at first by wavelet. The sample data set is reconstructed by pretreatment. The extending algorithm of Laplacian Eigenmaps based on the support vector regression machine (SVR) is proposed in this paper. The dimensionality reduction is done for the sample data set and new sample. Then the SVR is trained with the sample data set. The offset identification is done for new sample. By comparing with traditional method without the dimensionality reduction, the experiments confirm that the offset identification precision can be improved by the dimensionality reduction of Laplacian Eigenmaps.