变工况滚动轴承异常状态局部切空间分类检测

Classification and Detection of Local Tangent Space in Abnormal State of Rolling Bearing under Variable Working Conditions

变工况滚动轴承异常状态数据在特征空间上呈现高维模糊分类特征,异常状态数据的子特征分区极为困难,增加了轴承异常检测的难度。为此,提出变工况滚动轴承异常状态局部切空间分类检测方法。采用局部切空间排列法,降维处理变工况滚动轴承数据,使其在局部切空间满足分类空间映射条件,再利用深度置信网络,通过异常数据训练提取数据的异常特征。将提取的特征输入到SVM分类器中,利用非线性映射函数将二维特征矩阵映射到三维分类空间中再将超平面结构加入其中。在多项式核函数的引导下,找到对应的子特征分类区域,根据分类结果检测变工况滚动轴承的异常状态。实验结果表明:在调整轴承承载负荷前后,该方法针对异常状态的检测率较高,早期异常点检出所花时间较少。

As the abnormal state data of rolling bearing under variable working conditions featured with high-dimensional fuzzy classification in the feature space encounters the extreme difficulty in partitioning the sub features of the abnormal state data,which increases the difficulty of bearing abnormal detection,a classification method of local tangent space for abnormal state of rolling bearing under variable working conditions is proposed.The local tangent space arrangement method is used to reduce the dimension of the rolling bearing data under variable working conditions,thus meeting the mapping conditions of the classification space in the local tangent space.The depth confidence network is applied to extract the abnormal features of the data through the training of the abnormal data.By using the nonlinear mapping function,the extracted features are input into the SVM classifier,and the two-dimensional feature matrix is mapped into the three-dimensional classification space,to which the hyperplane structure is added.Under the guidance of polynomial kernel function,the corresponding sub-feature classification region is found,and the abnormal state of rolling bearing under variable working conditions is detected according to the classification results.The experimental results show that the method has a high detection rate for abnormal conditions and takes less detection time for early abnormal points before and after adjusting the bearing load.