Permutation entropy-based 2D feature extraction for bearing fault diagnosis
- PDF / 4,599,332 Bytes
- 15 Pages / 547.087 x 737.008 pts Page_size
- 70 Downloads / 231 Views
ORIGINAL PAPER
Permutation entropy-based 2D feature extraction for bearing fault diagnosis Mantas Landauskas · Maosen Cao · Minvydas Ragulskis
Received: 6 April 2020 / Accepted: 9 October 2020 © Springer Nature B.V. 2020
Abstract Bearing fault diagnosis based on the classification of patterns of permutation entropy is presented in this paper. Patterns of permutation entropy are constructed by using non-uniform embedding of the vibration signal into a delay coordinate space with variable time lags. These patterns are interpreted, processed and classified by employing deep learning techniques based on convolutional neural networks. Computational experiments are used to compare the accuracy of classification with other methods and to demonstrate the efficacy of the presented early defect detection and classification method. Keywords Permutation entropy · Convolutional neural network · Feature extraction · Fault detection 1 Introduction Rolling and other types of bearings play an important role in different kind of machinery. Excess vibraM. Landauskas · M. Ragulskis (B) Center for Nonlinear Systems, Kaunas University of Technology, Studentu 50-146, Kaunas, LT 51368, Lithuania e-mail: [email protected] M. Landauskas e-mail: [email protected] M. Cao Department of Engineering Mechanics, Hohai University, Hohai 210098, China e-mail: [email protected]
tions in bearings might induce other mechanical faults, increase the wear of devices, or even be a serious safety threat. However, a direct inspection of bearings is usually an unfeasible approach due to the complexity of the machinery, work safety problems or costs related to time constraints. Non-intrusive early fault diagnosis of bearings is usually based on intelligent computational analysis of experimental vibration data. Methods developed in this paper use vibration data in order to make informed decisions about the identification and classification of early defects in rotational ball bearings. Machine learning (ML) algorithms have been successfully used for early fault detection in rotational bearings. Support vector machines (SVM) and artificial neural networks (ANN) are applied on features extracted from vibration data in [16] (these features are mainly basic statistical measures). The fault classification performance of this approach is evaluated by the confusion matrix [16]. Confusion matrices are a common tool for evaluating the fault classification quality. More extensive approach of fault detection in rotating machinery is discussed in [2]. Support vector classification (SVC) analysis with a number of other ML techniques are applied on features of centrifugal pump vibration data; the McNemar’s test on confusion matrix is used to compare different ML methods and rank them in terms of their performance in [2]. A review of broader class of methods (Artificial intelligence (AI) in particular) for fault diagnosis can be found in [9,25]. A short
123
M. Landauskas et al.
overview of the main ANN architectures used for fault diagnosis is discussed
Data Loading...
