Fault Diagnosis Of Power Electronic Circuits Based On Wavelet Transform And Support Vector Machine

With the advances of power electronics, power electronics systems have become more complicated, and the reliability and safety of systems have become an important factor to the protection of economic and social benefits, so the fault diagnosis has been paid attention by scholars from various countries and become one research focus. With the development of theory and practice in recent twenty years, fault diagnosis technology breaks through the traditional method, such as model dependence, poor robustness and difficult to pinpoint. As the signal processing, fault diagnosis technology integrate with pattern recognition, artificial intelligence and other disciplines more closely together, so as to direction to the development of intelligent and precise positioning.According to the characteristics of power electronic circuit fault, this paper propose a diagnosis method of wavelet transform and support vector machines for fault diagnosis, and made the following research and analysis.Firstly, we use the Matlab simulation toolbox to establish the model, and take the failure of the power electronic circuits as the main object of study to analysis of types of the fault. We choose several types of representative failure as the study objects for this paper and obtain the output voltage signal from the test point under various fault conditions.Then, we use the method of multi-resolution analysis theory to decompose rectifier output voltage fault signal and propose the method of extracting eigenvectors that the energy of high frequency part every scale. In the meanwhile, we prove the reasonableness that taking the energy of high frequency part every scale as eigenvectors. According to the characteristics of power electronic circuits, we compare a variety of wavelet with making a lot of experiment in this paper, and identify the wavelet function and decomposition scale as Db40 and eight scales. Then we use Matlab wavelet toolbox to decompose the signal, extract the scale wavelet transform coefficients to calculate the energy of the scale and establish the energy feature space of various failure types.Based on the theory of support vector machine, in order to establish an effective classifier, we choose the radial basis function as the inner product function and use the bilinear grid search algorithm to determine the corresponding optimal error penalty parameter and Gaussian kernel parameter. According to the characteristics of power electronic circuits fault, we propose an improved one-against-all classification algorithm and build some classifiers to classify fault features. Through calculating classifier with the algorithm, the next step will be calculated no longer when a classifier can be regarded as a value of 1and it can be determined as the type of the circuit that the fault type this time. This algorithm can reduce the computation effectively and improve the efficiency of the test with this method. Simulation and test show that the proposed method can effectively diagnose power electronic circuit.