摘要
为解决在少量标记样本的条件下故障诊断困难的问题,提出一种基于改进半监督局部保持投影(ISS-LPP)的故障诊断方法。ISS-LPP算法利用部分标记样本的标签信息调整原始特征空间中样本间的权值矩阵,并根据所有样本在特征空间的分布情况自适应的调整邻域参数,寻找数据的低维本质流形,得到原始特征空间样本数据的低维特征向量和投影转换矩阵。以得到的低维特征向量为输入,建立分类器,识别和判断故障类型。将ISS-LPP算法应用于滚动轴承的故障诊断。实验结果表明:该方法能够在标记样本较少时,提高轴承的故障诊断精度。
In order to diagnose the fault effectively with a small number of labeled samples, a method of fault diagnosis based on improved semi-supervised locality preserving projections was proposed. The method of ISS-LPP used the information of some labeled samples to adjust the weight matrix among all samples in the original characteristic space. The neighborhood parameter could he adjusted automatically according to the distribution of the all samples. Therefore, the low-dimensional manifold could be found. So the low-dimensional eigenvectors and the projection matrix were achieved from the original characteristic space by ISS-LPP. With the low-dimensional eigenvectors as inputs, classifiers were established for identifying fault types. The method of ISS-LPP was applied for the fault diagnosis of roller bearing. The results indicate that the proposed method can diagnose bearing fault in high accuracy with a small number of labeled samples.
出处
《中南大学学报（自然科学版）》
EI
CAS
CSCD
北大核心
2015年第6期2059-2064,共6页
Journal of Central South University:Science and Technology
基金
国家自然科学基金资助项目(E51205405)~~
关键词
故障诊断
改进半监督局部保持投影
权值矩阵
邻域参数
滚动轴承
fault diagnosis
improved semi-supervised locality preserving projections
weight matrix
neighborhood parameter
roller bearing
