Application of Manifold Learning to Machinery Fault Diagnosis
The essence of machinery fault diagnosis is pattern recognition. Extracting the fault pattern contained in the vibration signal is the frequently used method to diagnose mechanical fault. Manifold Learning is widely used to extract the non-linear structur
- PDF / 632,811 Bytes
- 9 Pages / 439.37 x 666.142 pts Page_size
- 26 Downloads / 245 Views
stract. The essence of machinery fault diagnosis is pattern recognition. Extracting the fault pattern contained in the vibration signal is the frequently used method to diagnose mechanical fault. Manifold Learning is widely used to extract the non-linear structure within the data and could do the dimensionality reduction of high-dimensional signal. Therefore manifold learning is employed to diagnose the machinery fault. The feature space is constructed by characters in time-frequency domain of vibration signal firstly, and then the manifold learning method named as sparse manifold clustering and embedding is used to extract the essential nonlinear structure of feature space. Afterwards, the fault diagnosis is implemented with spectral clustering and support vector machine. The experiment demonstrates that the approach can effectively diagnose the fault of Machinery. Keywords: Manifold learning
1
· Fault diagnosis · Bearing
Introduction
With the increment of precision and operation complexity in industry, the techniques of machinery fault diagnosis, which is used for the safe guarantee in operation of mechanical equipments, has gained more and more attention. At present, data-driven methods play a important role in the mechanical fault diagnosis. The vibration signal contains a wealth of information which indicate the condition of machinery, therefore it is widely used in fault diagnosis that the method based on vibration signal processing [1]. The characters of vibration signal in time-frequency domain that is used to establish the initial feature space can be extracted by methods such as Fourier transform or wavelet transform etc. And then, the classifier is constructed with the further processing of initial feature space [1]. The dimensionality reduction method is taken to get a lowdimensional feature representation since the initial feature space is generally high-dimensional. The frequently used dimensional reduction methods, such as principal component analysis [2] or linear discriminant analysis [3], is the linear dimensional reduction method. For the nonlinear signal, the linear method is not the best choice. c IFIP International Federation for Information Processing 2016 Published by Springer International Publishing AG 2016. All Rights Reserved Z. Shi et al. (Eds.): IIP 2016, IFIP AICT 486, pp. 41–49, 2016. DOI: 10.1007/978-3-319-48390-0 5
42
J. Wang et al.
Manifold learning has been applied in many fields since three articles published in Science [4–6]. Manifold learning is a nonlinear data dimensional reduction method, which is used to extract low-dimensional manifold structure embed in high-dimensional space. Therefore there is a new way to diagnose the machinery fault. Generally speaking, the operation data of machinery with the same condition lie on the same manifold and the different condition lie on different manifold [7]. Based on this setting, manifold learning can be applied to machinery fault diagnosis. A typical manifold learning method includes Local Linear Embedding (LLE) [5], Isometric Map
Data Loading...
