The Research On Fault Diagnosis Of Power Electronic Equipment

Power electronics technology is widely used in national defense and military and industrial production, once the power electronic equipment have fault, it may cause device or system damage, even threaten people'life and property. Therefore, power electronic equipment fault detection and diagnosis is very meaningful. Because power electronic devices are low overload, and damage fast, the fault information always exist for less than ten milliseconds after the fault occurs, so it's necessary for power electronic devices to have dynamic monitoring and on-line diagnostics.In this paper, we take three-phase inverter power for example to conduct the power electronic device fault diagnosis, primarily complete the following work:First, classify the common fault of power electronic devices, and using MATLAB to build PID control circuit model of three-phase inverter power, then we can get a variety of common faults waveforms as fault diagnosis sample database in the state of simulation.Then use the BP neural network which has powerful classification capability and one-versus-one support vector machine which has unique advantages on solve small sample data and non-linear problem for fault diagnosis of power electronic devices separately. The extraction of main fault feature and the design of the sample learning parameters are finished in this paper. MATLAB is used to train the neural network and support vector machines respectively, and training results are also analyzed. From the training results we can see that the neural network and support vector machine method are both have good recognition ability in fault diagnosis.Finally, use the TMS320F2812 DSP for the design of three-phase inverter circuit fault diagnosis system, mainly designed the power supply, fault detection circuit and the external circuit. Then introduced the layout,cable and interference of PCB board design. In the end, using the neural network design the software of fault diagnosis and do the experiment validate, and get the result.