The Research On The Fault Diagnosis Methodsfor The Switching Power Supply

The switching power supply is the core of every electrical system. If the position of the fault cannot be located and made the corresponding processing in time when a fault occurs in the switching power supply, it will cause the damage of the components, or even cause the breakdown of the electrical system, leading to great economic losses and the threats to the personal safety. As a result, it becomes an increasingly important task which the faults of the switching power supply are needed to be found quickly and accurately. At present, the research on the fault diagnosis methods of the switching power supply considers two aspects: how to extract the fault features effectively and how to establish an efficient diagnostic model.For the difficulties to extract the fault features of the switching power supply caused by the nonlinear and the tolerance of the components and other problems, the wavelet packet transform and the principal component analysis method are combined to extract the fault features effectively. Firstly, the signals are denoised by the wavelet packet transform. Secondly, the fault features of the switching power supply are extracted based on the technique of wavelet packet frequency band energies, solving the difficult problems of fault features extraction. Thirdly, to extract the fault features effectively, the principal component analysis is adopted to eliminate the redundancies, solving the multicollinearity problems caused by the relevant inputs.Since the switching power supply has no unified fault diagnosis model, general fault diagnosis methods such as fault dictionaries are difficult to meet the diagnostic requirements of the switching power supply. Although the neural networks provide a way to the fault diagnosis of the switching power supply, every neural network has its own limitations. To solve the above problems, a new fault diagnosis model based on the combination of the improved gravitational search algorithm and neural network(IGSA-NN) is put forward for the switching power supply. Firstly, to solve the problem of easily falling into local optimal and the problem of low optimizing accuracy when solving the function optimization by the standard GSA, an improved gravitational search algorithm based on the time-varying weights and the boundary mutation strategy is proposed. The simulation results show that the IGSA has better optimization performance. Secondly, taking the IGSA optimizing the BPNN and the RBFNN for examples, the detailed operation steps are given respectively. Thirdly, the IGSA-BPNN and IGSA-RBFNN are used for the fault diagnosis of switching power supply respectively. The experiments indicate that the IGSA-NN model has higher speed and accuracy of diagnosis than the general neural networks.In the research on the theory and application of the fault diagnosis for the switching power supply, an open and high performance experimental platform is essential in the fault diagnosis of the switching power supply. The laboratory virtual instrument(Lab VIEW) technology is applied to the software system design of the fault diagnosis for the switching power supply, improving the efficiency and reliability of experimental datas effectively.