Research On Fault Identification And Health Condition Assessment Of Power Converters

Power electronic systems have been widely used in new energy generation,smart grids,electric vehicles,rail transportation,aerospace and other fields.However,with the introduction of concepts such as multi-electric/all-electric aircraft,power system power electronics,power electronic transformers,etc,the requirements for the reliability,safety,and guaranteed effectiveness of power electronic systems have also gradually increased.To reduce the damage and economic loss caused by system performance degradation or failure,the Prognostics and Health Management?PHM?has been greatly valued and promoted by many domestic and foreign research institutions and many industrial enterprises.This advanced technology can predict the occurrence of failures in advance effectively,forecast the remaining useful life of the system,minimize the number of unplanned maintenance,and reduce the life cycle cost.This article focuses on the key PHM technologies of power converter systems,including fault diagnosis,condition monitoring and health status assessment.The main research contents and innovations include:?1?A structural fault diagnosis method for power converter based on improved deep belief network?DBN?model is studied.Because there are many types of fault patterns in power converters,there are similarities in fault characteristics between different fault types,which make it difficult for traditional shallow neural network diagnosis methods to identify and locate in this situation.Therefore,based on deep belief network for superior pattern recognition capability,a structural fault diagnosis method for SEPIC converter based on CSA-DBN is proposed.By optimizing the number of hidden layer nodes to improve the classification performance of DBN network when there are many types of failure modes,as a result,these hard faults can be identified accurately.Also,it can improve fault diagnosis accuracy and fault location ability.?2?An open circuit fault diagnosis method of brushless DC motor full-bridge DC-AC driving circuit based on the automatic extraction feature of the stacked denoising autoencoder?SDAE?is studied.In view of the uncertainty brought by the traditional manual extraction features and the complexity brought by the preferred features,based on the deep learning network has adaptively extracted features from high-dimensional big data,the SDAE can establishe the mapping relationship between data samples and failure modes.The methoed based on SDAE can extract deep fault feature information in order to achieve accurate fault identification and fault location.?3?Considering the influence of parasitic components of the key components?diodes,inductors,power MOSFETs and electrolytic capacitors?on the performance degradation of the Boost converter,the hybrid system model for Boost converter is deduced which takes the drain source voltage of the power MOSFET and the output voltage of the circuit as state variables.A Chaos Whale Optimization Algorithm?CWOA?is proposed to obtain the fault feature parameters of each key component,and the parameter identification problem is converted into the objective function optimization problem,so as to realize the non-invasive power converter condition monitoring.This overcomes the traditional method disadvantages that based on the inductive current and the output voltage as the state variables,due to it is impossible to identify the internal resistance of the diode,the internal resistance of the inductor and the on-resistance of the power MOSFET.?4?A data-driven CSA-KELM method is proposed to realize the identification of the key components'characteristic parameters of the power converter.It is difficult to accurately describe the dynamic behavior of the system according to the mathematical model when the equivalent model of the system or component is complex.Thus,in order to solve this problem,only the voltage across the diode and the output voltage of the circuit need to be collected,which can identify the characteristic parameters of the inductor and the electrolytic capacitor.?5?The change of operating temperature will result in the inconsistency between the estimated value of the characteristic parameters of the component and the failure criterion?at 25°C?,resulting in inaccurate health assessment results.The mathematical model between the electrolytic capacitor characteristic parameters C,R_C and the drain on-resistance R_on of the power MOSFET and the operating temperature were established through experimental research and the normalized factor model of the characteristic parameters of the components have been obtained.Meanwhile,the quantitative evaluation of health index HI expressions for electrolytic capacitors and power MOSFETs is proposed based on the component failure criterion.A method of dynamically updating normalized factor model parameters based on Unscented Particle Filter?UPF?is proposed to realize the health status assessment of key components of Boost converter under variable operating temperature conditions.