Bearing fault diagnosis based on shared neighbors weighted local linear embedding
LIU Qingqiang, SUN Yanru, LIU Yuanhong, WU Li
1. School of Electrical and Information Engineering, Northeast Petroleum University, Daqing, Heilongjiang 163318, China; 2. Training Center of Natural Gas Branch, Daqing Oilfield Company Limited, Daqing, Heilongjiang 163457, China
摘要 针对传统局部线性嵌入算法在挖掘局部流形结构时未充分考虑样本邻居分布信息，且在降维过程中默认样本具有相同的重要性导致提取鉴别特征不明显的问题，提出基于共享近邻的加权局部线性嵌入（weighted local linear embedding based on shared neighbors, SN-WLLE）算法，并用于滚动轴承故障诊断.该算法首先使用余弦距离划分样本邻域；其次计算样本邻域对相似度用以评估样本共享近邻信息，并结合样本的6种邻居分布修正局部结构挖掘，提高多共享近邻的k近邻重构准确性；接着从多流形的角度评估样本点与近邻点间的稀疏分布一致性，以获得样本的重要性指标，并在低维空间保持该信息，进而提取准确的鉴别特征；最后结合KNN分类器构建出完备的轴承故障诊断模型.采用凯斯西储大学轴承数据集和实验室测试平台轴承数据集，从可视化评估、定量聚类评估、故障识别精度评估及鲁棒性评估等方面进行分析.结果表明：SN-WLLE算法的F值保持在108以上水准，平均故障识别精度最低可达0.973 4，不仅具有较好的类内紧致性与类间可分性，还对近邻参数k具有低敏感性.
Abstract： To solve the problems that the neighbor distribution information of the samples was ignored for local linear embedding during mining the local manifold structure, and the default samples had the same importance in the dimensionality reduction process, leading to the inconspicuous feature extraction, the weighted local linear embedding based on shared neighbors (SN-WLLE) was proposed for bearing fault diagnosis. The cosine distance was adopted to divide the sample neighborhood in SN-WLLE. The sample neighborhood pair similarity was calculated by Jaccard coefficient to evaluate the sample shared neighbor information, and the local structure mining was modified by combining six neighbor distributions of the sample to improve the accuracy of the k-nearest neighbor reconstruction of multiple shared neighbors. The consistency of the sparse distribution between the sample and neighbors was evaluated to obtain the importance index of sample from the perspective of multi-manifolds, and the information was maintained in the low-dimensional space to extract accurate identification features. The complete bearing fault diagnosis model was constructed by combining KNN classifier. The bearing dataset of Case Western Reserve University and the bearing dataset of the laboratory test platform were used to analyze visual evaluation, quantitative clustering evaluation, fault identification accuracy evaluation and robustness evaluation. The results show that the F-value of SN-WLLE is maintained above 108, and the average fault identification accuracy can reach the minimum value of 0.973 4, which not only has good intra-class compactness and inter-class separability, but also has low sensitivity to the nearest neighbor parameter k.
刘庆强， 孙艳茹， 刘远红， 吴丽. 基于共享近邻加权局部线性嵌入的轴承故障诊断[J]. 江苏大学学报（自然科学版）, 2024, 45(1): 85-91.
LIU Qingqiang, SUN Yanru, LIU Yuanhong, WU Li. Bearing fault diagnosis based on shared neighbors weighted local linear embedding[J]. Journal of Jiangsu University(Natural Science Eidtion)
, 2024, 45(1): 85-91.
LIN Y, LONG W, LI Y Y, et al. Bearing condition monitoring and fault diagnosis based on mathematical morphology and fractal theory and PNN[J]. Journal of Sichuan University (Natural Science Edition), 2021,58(5):71-78. (in Chinese)
YU Z, FU J M, LIU L J. Probabilistic neural network fault diagnosis model based on FFT[J]. Journal of Sichuan University (Natural Science Edition),2020,57(5):909-914. (in Chinese)
［3］
WANG Q. Knowledge-based approach for dimensionality reduction solving repetitive combinatorial optimization problems[J]. Expert Systems with Applications, DOI: 10.1016/j.eswa.2021.115502.
［4］
DZEMYDA G, SABALIAUSKAS M. Geometric multi-dimensional scaling: a new approach for data dimensio-nality reduction[J]. Applied Mathematics and Computation, DOI: 10.1016/j.amc.2020.125561.
［5］
LUO J Q, XU T F, PAN T, et al. An efficient method of hyperspectral image dimension reduction based on low rank representation and locally linear embedding[J]. Integrated Ferroelectrics, 2020, 208(1): 206-214.
［6］
YAO B B, SU J, WU L F, et al. Modified local linear embedding algorithm for rolling element bearing fault diag-nosis[J]. Applied Sciences, DOI:10.3390/app7111178.
［7］
WU Q, JING R R, WANG E. Improved weighted local linear embedding algorithm based on Laplacian eigenmaps[J]. International Journal of Knowledge-based and Intelligent Engineering Systems, 2020, 24(4):323-330.
［8］
MEHTA K, TYAGI G, RAO A, et al. Modified locally linear embedding with affine transformation[J]. Natio-nal Academy Science Letters, 2017, 40(3): 189-196.
［9］
ZIEGELMEIER L, KIRBY M, PETERSON C. Sparse locally linear embedding[J]. Procedia Computer Science, 2017, 108: 635-644.
WANG X Y, DING S F. Pairwise constrained spectral clustering algorithm based on shared nearest neighborhood[J]. Computer Engineering and Applications, 2019,55(2):142-147. (in Chinese)
［11］
ZHU X S, ZHANG J, XU Y P, et al. Single-cell clustering based on shared nearest neighbor and graph partitioning[J]. Interdisciplinary Sciences: Computational Life Sciences, 2020, 12(2): 117-130.
［12］
XIE J Y, WANG M Z, XU S Q, et al. The unsupervised feature selection algorithms based on standard deviation and cosine similarity for genomic data analysis[J]. Frontiers in Genetics, DOI:10.3389/fgene.2021.684100.
［13］
BAHARAV T Z, KAMATH G M, TSE D N, et al. Spectral jaccard similarity: a new approach to estimating pairwise sequence alignments[J]. Patterns, DOI: 10.1016/j.patter.2020.100081.
