基于收缩自编码器和局部保持投影的机械故障特征提取  

Mechanical fault feature extraction based on contractive auto-encoder and locality preserving projections

作　　者：郝宇星 刘庆强[1] HAO Yuxing;LIU Qingqiang(School of Electrical and Information Engineering,Northeast Petroleum University,Daqing 163318,China)

机构地区：[1]东北石油大学电气信息工程学院,大庆163318

出　　处：《现代制造工程》2025年第2期84-93,137,共11页Modern Manufacturing Engineering

基　　金：海南省自然科学基金资助项目(623MS071)。

摘　　要：局部保持投影算法的性能主要依赖于构造的最近邻图,而构造最近邻图时容易受到原始数据冗余信息的干扰,以及没有良好的依据选择合适的热核参数带来的影响,导致不能充分挖掘高维数据的局部结构信息,在低维嵌入过程中也易对噪声和异常值较为敏感,影响其在故障诊断应用中的特征提取能力。针对以上问题,提出基于收缩自编码器和流形排序的局部保持投影算法(Locality Preserving Projections algorithm based on Contractive Auto-Encoder and Manifold Ranking,CAE-MRLPP),并用于机械设备故障诊断。首先,将样本标签信息和斯皮尔曼相关系数结合,预调整样本间距;其次,引入流形排序思想,根据样本点与邻域点在彼此邻域集中的排序位置信息以及二者的互邻个数信息来构造权重;最后,将收缩自编码器与基于流形排序的局部保持投影相融合,通过梯度下降法迭代优化求解出最优的投影矩阵,进而得到故障数据的低维表示。分别在滚动轴承数据集和抽油机数据集上进行了多项验证,故障识别准确度均在98%以上,表明该算法具有良好的特征提取能力,能够有效提高故障识别准确度,同时具有较好的鲁棒性和泛化能力。The performance of the locality preserving projections algorithm mainly depends on the construction of the nearest neighbor map.The construction of the nearest neighbor map was easily affected by the interference of redundant information in the original data and the impact of not having a good basis for selecting the appropriate heat kernel parameters.As a result,the local structural information of high-dimensional data cannot be fully explored,and it was also easy to be more sensitive to noise and outliers in the low-dimensional embedding process.And its feature extraction ability in fault diagnosis applications was affected.To address the above problems,a Locality Preserving Projections algorithm based on Contractive Auto-Encoder and Manifold Ranking(CAE-MRLPP)was proposed and used for mechanical equipment fault diagnosis.Firstly,it combined the label information and Spearman correlation coefficient to pre-adjust the sample spacing.Secondly,the idea of manifold ranking was introduced to construct the weights based on the information of the sorting position of the sample points and the neighboring points in each other′s neighborhood sets and the information of the number of mutual neighbors of the two.Lastly,the contractive auto-encoder was fused with the locality preserving projections based on the manifold ranking,and the optimal projection matrix was solved by iterative optimization of the gradient descent method.Then the low-dimensional representation of the fault data was obtained.A number of verifications were carried out on the rolling bearing dataset and the pumping unit dataset,and the fault identification accuracy was more than 98%.This indicated that the algorithm has good feature extraction ability,can effectively improve the fault identification accuracy,and has good robustness and generalization ability.

关 键 词：局部保持投影 特征提取 故障诊断 收缩自编码器 抽油机 滚动轴承 

分 类 号：TH165.3[机械工程—机械制造及自动化]