Open AccessUnsupervised fault diagnosis method based on iterative multi‐manifold spectral clustering
Author(s) -
Song Wenbin,
Wen Long,
Gao Liang,
Li Xinyu
Publication year - 2019
Publication title -
iet collaborative intelligent manufacturing
Language(s) - English
Resource type - Journals
ISSN - 2516-8398
DOI - 10.1049/iet-cim.2019.0003
Subject(s) - cluster analysis , nonlinear dimensionality reduction , computer science , spectral clustering , fault (geology) , pattern recognition (psychology) , tangent space , outlier , artificial intelligence , data mining , dimensionality reduction , mathematics , seismology , geology , mathematical analysis
Fault diagnosis is very important to the modern manufacturing system. As a powerful data‐driven method, machine learning (ML) has been widely used for fault diagnosis. However, a certain amount of labelled data is essential to most of the ML methods, which is not always available. In this study, an unsupervised fault diagnosis method based on iterative multi‐manifold spectral clustering (IMMSC) is proposed. The IMMSC, that the affinity matrix is constructed based on local tangent space, can improve the performance of spectral clustering on multi‐manifold distributed data. The IMMSC is conducted after feature extraction to divide data into different clusters first. Then local outlier factor is utilised to identify the normal condition. A fault index is found to recognise the fault conditions. The proposed IMMSC is tested on five simulation datasets and the proposed method is validated on the motor bearing dataset provided by Case Western Reserve University. The mean accuracy of IMMSC on simulation data is 97.85%, which outperforms the other spectral clustering and traditional methods. Furthermore, the result of the IMMSC is stable over 30 individual tests. The mean accuracy of the proposed method on motor bearing dataset is 98.53%. The result demonstrates its potential in fault diagnosis field.