An intelligent fault diagnosis method of high voltage circuit breaker based on improved EMD energy entropy and multi-class support vector machine
Introduction
The high voltage CB is one of the most critical power apparatus in power systems. It is used to cut off the current of power system when necessary and to isolate faulty parts of the system as a part of the protective relaying operation. Therefore, diagnosis of potential faults is essential for ensuring stable electrical power supply to consumers. The second CIGER inquiry on high voltage CB reported that 44% of major failures and 39% of minor failures are of mechanical origin [1]. Therefore, it is of great significance to study the machinery failure of high voltage CB. Because vibration signals, generated while high voltage CB opening/closing operation, contain important information about health status of device, vibration signal analysis has become a hot research in recent years [2], [3]. Poor lubrication, slow relays and latches, malfunction of shock absorbers are typical problems that can be detected by vibration analysis [2], [3], [4], [5]. The selection of good features is an important phase in pattern recognition. The conventional vibration analysis, which is based on Fourier transformation to represents the frequency composition of vibration signatures required linear system and a stationary signal, is used in [3], [4]. However, vibration signatures collected during the operation of high voltage CB contain numerous non-linear, non-stationary and fractal characteristics [2], [3]. In recent years, some new methods have been reported in [6], [7], [8], [9], [10], [11], which are based on wavelet theory, envelope analysis and pattern recognition. All these studies have been applied successfully in a certain extent to identify faulty high voltage CB vibration natures, but they have not achieved a perfect success rate.
Wavelet analysis can provide the local features of the signal in both the time and frequency domains, so it has been widely used in the machinery fault diagnosis [12], [13]. In Ref. [6], [11], it has been proven that wavelet analysis really has efficiency in detecting the fault type for high voltage CB. However, the wavelet analysis is essentially an adjustable windowed Fourier transform. Due to the limitation of the wavelet bases’ length, energy leakage would occur in wavelet transformation [14]. Furthermore, once the wavelet bases and the decomposition scales are determined, the results of wavelet transform would be the signal under a certain scale, whose frequency components related only to the sample frequency rather than the signal itself. Therefore, wavelet analysis is not a self-adaptive signal processing method in nature [15].
Recently, a new signal analysis method, namely empirical mode decomposition developed by Huang et al., is based on the local characteristic time scale of the signal and can decompose the complicated signal into a number of IMFs. By analyzing each resulting IMF component which involves the local characteristic of the signal, the characteristic information of the original signal can be extracted more accurately and effectively. In addition, the frequency components involved in each IMF not only relate to the sampling frequency but also change with the signal itself. Therefore, EMD is a self-adaptive signal processing method that can be applied to non-linear and non-stationary process perfectly [15], [16], also EMD has overcome the limitation of the Fourier transform and has high signal noise ratios (SNR) as well.
Among the various methods for condition monitoring of machinery, artificial neural network (ANN) has become an outstanding method of exploiting their non-linear pattern classification properties in the recent decades. Although ANN offers advantages for automatic detection and identification of electrical equipment failure conditions, it does not require an in-depth knowledge of the behavior of the system [17]. This method is based on an empirical risk minimization principle and has some disadvantages such as local optimal solution, low convergence rate, obvious ‘over-fitting’ and especially poor generalization when the number of fault samples is limited [18].
SVM is a new machine-learning tool which is especially suitable for classification, forecasting and estimation in small sample cases. SVM has been used widely for fault identification [19], [20], [21], [22]. SVM aims at the optimal solution in the available energy rather than the optimal solution when the sample number tends towards infinitely large. It solves satisfactorily the ‘over-fitting’, the local optimal solution, the low convergence rate and some other problems, and it has a good generalization even when the samples are few [19].
SVM theory is proposed for binary classification. Many approaches have been proposed to extend the binary SVM to multi-class problems [20], [21]. The common scheme is that a multi-class SVM is designed to deal with the problem as a collection of two classifications that can be solved by binary SVM. In this study, we have developed a ‘one against others’ SVM algorithm for the multi-class fault diagnosis.
The rest of the paper is organized as follows. In Section 2, the EMD method will be introduced, while SVM algorithm will be described in Section 3. In Section 4, we are going to explain the method based on improved EMD energy entropy feature extraction and fault diagnosis steps. In Section 5, we are going to explain the experimental setup, ‘one against others’ SVM algorithm, the optimal parameters of the MSVM mode and also the detailed analysis and discussion of the major findings. Finally, we will summarize our results and make some conclusions.
Section snippets
EMD method
EMD method is developed from the simple assumption that any signal consists of different simple intrinsic modes of oscillations. Each linear or non-linear mode will have the same number of extrema and zero-crossings. There is only one extremum between successive zero-crossings. Each mode should be independent of the others. In this way, each signal could be decomposed into a number of IMFs, each of which must satisfy the following definition [15]:
- (1)
In the whole data set, the number of extrema and
Multi-class SVM algorithm
SVM comes from an optimal separating hyper-plane in case of linearly separable [23]. Its core idea is to map the original pattern space into the high dimensional feature space through some non-linear mapping functions, and then constructs the optimal separating hyper-plane in the feature space.
The input vector x is mapped into the high dimensional feature space through the non-linear mapping function φ(x), and the linear function setsThe discussion above deals with binary
Signal envelope extraction
To extract more information, each of these vibration signals is decomposed by the EMD method. The first some IMFs containing almost all valid information are selected. Energy entropies of different vibration signals show that the energy of a vibration signal will change in different frequency bands when a machinery fault occurs [22]. Signal envelope method is useful for signal analysis.
The Hilbert transformation of a signal is defined as follows:With this definition,
Data acquisition and improved EMD energy entropy extraction
For accurate vibration measurements on high voltage CB, accelerometer performance should be selected based on the maximum vibration burst that can be expected for the breaker drive mechanism. The accelerometer, M353B02 (PCB Piezotronic), is used to record the vibration signals of high voltage CB during the closing operation. Performance parameters of M353B02 includes: sensitivity of ;20 mV/g (g = 9.8 m/s2), measuring range of ±250 g, an upper frequency limit of 7 kHz, non-linearity of 1% and weight
Conclusion
This paper present a new method based on improved EMD energy entropy and MSVM to diagnose machinery fault for high voltage CB using vibration analysis under no-load. Three kinds of multi-class support vector machines with cross-validation technique and gradient method to sort the optimal parameters are used and compared to determine the fault types. The testing results show that the proposed MSVM with OAOT algorithm can correctly and effectively diagnose multi-class machinery faults for high
References (25)
- et al.
An expert system for fault diagnosis in internal combustion engines using wavelet packet transform and neural network
Expert Syst. Appl.
(2009) - et al.
On the energy leakage of discrete wavelet transform
Mech. Syst. Signal Process.
(2009) - et al.
An expert system for fault diagnosis in internal combustion engines using wavelet packet transform and neural network
Expert Syst. Appl.
(2009) - et al.
A neural network-based scheme for fault diagnosis of power transformers
Electr. Power Syst. Res.
(2005) - et al.
An intelligent fault diagnosis method based on wavelet packer analysis and hybrid support vector machines
Expert Syst. Appl.
(2009) - et al.
Fault diagnosis of power transformer based on multi-layer SVM classifier
Electr. Power Syst. Res.
(2005) - et al.
Support vector machines-based fault diagnosis for turbo-pump rotor
Mech. Syst. Signal Process.
(2006) - et al.
Use of wavelet transform to enhance piezoelectric signals for analytical purposes
Anal. Chim. Acta.
(2002) - et al.
High-performance implementation of wavelet algorithms on a stand PC
Microprocess. Microsyst.
(2002) - CIGRE Working Group 13.06, Final report of the second international enquiry on high voltage circuit breaker failures...
Event timing and shape analysis of vibration bursts from power circuit breakers
IEEE Trans. Power Deliv.
Vibration analysis for diagnostic testing of circuit breakers
IEEE Trans. Power Deliv.
Cited by (132)
Fault diagnosis of high voltage circuit breaker based on multi-sensor information fusion with training weights
2022, Measurement: Journal of the International Measurement ConfederationA novel adaptive single-phase reclosure scheme based on improved variational mode decomposition and energy entropy
2020, International Journal of Electrical Power and Energy SystemsDynamic feature information extraction using the special empirical mode decomposition entropy value and index energy
2020, EnergyCitation Excerpt :The establishment of AR model parameters is reflected in the IMF’s characteristic frequency. The Hilbert Transform is used in the EMD’s HT time–frequency analysis method, which is used during evaluation of instantaneous frequency [14–16]. Firstly, apply Hilbert Transform to IMF, then parse the form of analytical signals and finally obtain the instantaneous frequency after the derivative of the analytical signal phase position.