| IGBT power module is the core device of energy conversion and transmission.It is widely used in smart grid,rail transit,new energy generation and other fields,it is also used to improve the efficiency and quality of power consumption.And IGBT power module is a key device that helps the industry to realize the development of electrification,intelligence and network connection.It is of great significance to solve energy shortage and reduce carbon emissions.However,with the increase of the working time of IGBT power modules,the problem of aging failure becomes more serious and the replacement cost is higher.Therefore,online evaluation of the aging status of IGBT power modules is of great significance for predicting their remaining life and improving their reliability.This paper analyzes the failure mechanism of the module,obtains the degradation parameters of the IGBT power module by using the accelerated aging test platform,establishes an aging condition evaluation system,and conducts in-depth research on the aging condition evaluation method based on machine learning.This article focuses on the following aspects:Firstly,the basic structure and basic operating characteristics of the IGBT power module are introduced.The heat transfer mechanism of the module is analyzed starting from the package structure of the module,the package failure and failure modes of the module are analyzed in detail,and the state parameters related to aging failure are summarized.The correlation between the state parameters is analyzed,and the saturation voltage drop,junction temperature,and collector current are selected as the state characterization of the aging status assessment model.Secondly,this paper introduces the experimental scheme,experimental steps and data acquisition process of this paper.Through accelerated aging test and single pulse test,the saturation voltage drop,junction temperature,and collector current data under different aging conditions are obtained,and the interactions are analyzed.The relationship between the module’s thermal resistance increase and the number of aging times was analyzed,the relationship between the number of accelerated aging times and the aging state was found,and an aging status evaluation system was established.Thirdly,the aging state evaluation model of IGBT power module based on machine learning is presented,and the modeling process and data processing process of the model are explained.The aging status evaluation model based on BP neural network and the extreme learning machine(ELM)is compared.The comparison found that the ELM-based aging condition evaluation model is more suitable for module aging status assessment.Based on this,some improvement studies on ELM were conducted,and it was found that the ELM optimized based on swarm intelligence algorithm is more suitable for the aging status evaluation of IGBT modules than the cored ELM,and the evaluation effect is better.Finally,an improved aging state evaluation model of the improved gray wolf algorithm optimized extreme learning machine(IGWO-ELM)is proposed.The principle of the gray wolf algorithm is introduced,and an improved gray wolf optimization algorithm is proposed for its shortcomings.The performance test of the improved gray wolf algorithm(IGWO)is shown.The results prove that it is superior to other algorithms in convergence speed and accuracy.The IGWO is used to optimize the weights and thresholds of ELM,introduce the modeling steps,list the results of the model’s evaluation of the aging status of IGBT power modules,and prove through comparison with other aging status evaluation models.Based on IGWO-ELM’s IGBT power module aging status assessment model has higher accuracy and stronger applicability. |
