Robust unsupervised feature selection via sparse and minimum-redundant subspace learning with dual regularization

Zeng, Congying; Chen, Hongmei; Li, Tianrui; Wan, Jihong

doi:10.1016/j.neucom.2022.09.074

Cited by 8 publications

(1 citation statement)

References 45 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Guo [19] used Unsupervised Feature Selection With Adaptive Structure (FSASL) method to eliminate the impact of noise, redundancy, and missing values on the original signal, improving the effectiveness of feature selection. Zeng [20] eliminated the influence of noise or outlier in real fault data through Robust Unsupervised Feature Selection (RUFS) algorithm. Zhu [21] combined the construction of similarity matrices with the feature selection process through the Graph Learning Unsupervised Feature Selection (GLUFS) algorithm, improving the effectiveness and superiority of the feature selection process.…”

Section: Introductionmentioning

confidence: 99%

A Novel Fault Diagnosis Method Based on NEEEMD-RUSLP Feature Selection and BTLSTSVM

Lu,

Xu,

Zhou

et al. 2023

IEEE Access

View full text Add to dashboard Cite

The vibration signal of rolling bearings is a nonlinear and non-stationary signal, which is affected by the working condition change and background noise, and the reliability of traditional feature extraction methods and fault identification methods is low. In order to effectively extract feature vectors and improve the accuracy and reliability of fault identification, we propose a new fault diagnosis method based on noise eliminated ensemble empirical mode decomposition and robust unsupervised feature selection with local preservation (NEEEMD-RUSLP) and binary tree least squares twin support vector machine (BTLSTSVM). Firstly, NEEEMD is introduced to suppress background noise and decompose the vibration signal into a series of intrinsic mode functions (IMF), and the wavelet packet energy entropy, packet energy coefficient, and Gini coefficient of each IMF are extracted to construct time-frequency domain features. Then, 16 time-domain features and 13 frequency-domain features of the original signal are extracted and combined with the time-frequency domain features of each IMF to construct a high-dimensional feature space. In order to reduce the feature dimension and improve the diagnostic accuracy of the model, the RUSLP feature selection method is introduced to select effective low-dimensional features from the high-dimensional features. In addition, the binary tree (BT) strategy is introduced into the LSTSVM binary classifier to construct the BTLSTSVM multi-classifier, which aims to improve the recognition accuracy of lowdimensional features. In the bearing fault diagnosis of Case Western Reserve University, the fault diagnosis accuracy obtained by the proposed method is improved by 10.67%. In the bearing fault diagnosis of the University of Ottawa, the fault diagnosis accuracy obtained by the proposed method is improved by 10%. In the fault diagnosis of check valve in the actual industrial production environment, the fault diagnosis accuracy obtained by the proposed method is improved by 22%. The results show that the proposed method can not only effectively extract and select the low-dimensional fault characteristics of the bearing, but also achieve competitive fault diagnosis accuracy. Therefore, this method can provide a new method reference for the field of fault diagnosis, and has great theoretical significance and application value.INDEX TERMS empirical mode decomposition, unsupervised feature selection, mixed domain, fault diagnosis, least square twin support vector machine

show abstract

Section: Introductionmentioning

confidence: 99%