Research On Diagnosis Of Open Circuit Fault For Three-level Inverter Based On Neural Network Of MLD Model

In this paper,the grid-connected inverter,the core of the wind power system,is studied.Based on the topology structure of the three-level inverter and its working principle,the conduction logic and coupling relationship between output voltage and current of each bridge IGBT is deeply analyzed with the diode-clamped(NPC)three-level inverter as the research object.Proposed a method to accurately locate and determine the cause of the failure of IGBT,in the light of the same output voltage and current waveforms of the two kinds of failure modes when there are double IGBT fault and single IGBT fault in the middle part of the same phase upper bridge(lower bridge)arm;at the same time for the same output characteristics situation of the open circuit fault caused by the drive failure or the device break.The MLD(Mixed Logical Dynamic)model of three level inverter is constructed,and its validity is verified on the Simulink platform.In order to accurately diagnose the fault and locate the failure driven parts,the MLD model expression of the inverter is equivalently transformed into DMLD(Diagnosis Mixed Logical inverter Dynamic)model,making the output voltage and current are the same as the simulation during normal operation,not same under fault conditions,to effectively locate failure pulse.Secondly,the line current residuals outputted from both the DMLD model and the actual simulation model are the fault signal sources,and the residual maximum value Max,the minimum Min,Add = Max + Min and App = Max-Min are used as the Characteristic expression of IGBT trigger pulse failure.The harmonic content of the clamped voltage across the clamp diode is analyzed by FFT(Fast Fourier Transform),and the effective characteristics of the device are diagnosed by 1,2,4 and 6 harmonics.And reassembles the two features to generate a new feature set that includes both the trigger failure state and the device state.The three-layer BP neural network is used to identify the fault feature set.The results show that the proposed the fault diagnosis method is effective.