Content area
Full text
Received Dec 1, 2017; Accepted Feb 20, 2018
This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.
1. Introduction
With the increasing complexity of modern industry, fault diagnosis as accurately and timely plays an important role in industrial applications. Many fault diagnosis analysis methods have been developed to accurately and automatically identify faults in the past two decades. They usually use some basic measurements, like vibration, acoustic, temperature, and wear debris analysis [1, 2]. Although these methods have been beneficial, tests are still quite expensive and time-consuming. Fault diagnosis using the big data is a goal that has not yet been fully implemented. When the machine creates faults, the dynamic signals of the machine structure can be monitored in real time. The most effective fault diagnosis is feature learning from the monitoring information. Numerous researchers focus on fault diagnosis by faults patterns recognition, but most studies are concerned with the single fault patterns recognition [3–5]. However, the fault of the machinery may be the compound, which is influenced by two or three causes in practice.
Compared with single fault, compound faults lead to serious performance degradation and are more difficult to recognize. This leaves the challenging task to identify multiple faults effectively. There are a few studies on multiple faults patterns recognition [6–8]. Chen et al. integrate extreme-point symmetric mode decomposition with extreme learning machine to identify typical multiple patterns recognition [6]. Ranaee and Ebrahimzadeh use a back-propagation (BP) neural network to recognize multiple-fault patterns [7]. Lu et al. propose a hybrid system that uses independent component analysis (ICA) and support vector machine (SVM) for recognizing mixture patterns. That method initially applies the ICA to get the independent components (ICs), and then the ICs are used as the inputs for the SVM classifier [8].
Feature extraction has become a major technique for multiple-fault patterns recognition. Numerous previous studies have reported about signal processing [9–13], like Fourier Transform [9] and wavelet transform [10]. But these methods should select appropriate base functions in advance. It is difficult to get effective analysis results, because the data from real world machines are nonstationary and nonlinear. Empirical mode decomposition...
| |