基于神经网络的双面散热功率模块本构热阻建模与表征  

Intrinsic Thermal Impedance of Double Sided Cooling Power Module Via Neural Network:Modeling and Characterization

作　　者：梁钰茜 孙鹏[1] 牛富丽[1] 梁森浩 曾正[1] LIANG Yuxi;SUN Peng;NIU Fuli;LIANG Senhao;ZENG Zheng(State Key Laboratory of Power Transmission Equipment&System Security and New Technology(Chongqing University),Shapingba District,Chongqing 400044,China)

机构地区：[1]输配电装备及系统安全与新技术国家重点实验室(重庆大学),重庆市沙坪坝区400044

出　　处：《中国电机工程学报》2024年第15期6095-6109,I0022,共16页Proceedings of the CSEE

基　　金：国家自然科学基金项目(52177169);重庆市基础研究与前沿探索项目(cstc2022ycjh-bgzxm0155)。

摘　　要：双面散热(double-sided cooling,DSC)功率模块的结-壳热阻,是其电热设计、状态监测与失效分析的关键性能指标。然而,现有的结-壳热阻模型及其测试标准,仅针对单通道传热的功率模块,难以匹配多通道传热的DSC功率模块,无法识别DSC功率模块的本构热阻信息。首先提出DSC功率模块的本构热阻模型,阐释DSC功率模块本构热阻的物理意义,分析重构单通道传热本构热阻的可行性和唯一性;其次,评估所提本构热阻与传统测试热阻的本质差异。基于商业化DSC功率模块,采用多物理场分析方法,构建DSC功率模块的热阻数据库;然后,基于神经网络学习方法和所构建的数据库,建立DSC功率模块的本构热阻模型,并通过大量训练数据和测试数据的交叉验证及多款DSC功率模块的实测结果,验证神经网络重构本构热阻的泛化能力和兼容能力,为多通道散热功率模块的热阻建模与表征,提供新的研究思路和技术途径。In terms of the electro-thermal design,condition monitoring,and failure analysis,the junction-case thermal resistance is a key specification of the double-sided cooling(DSC)power module.Nevertheless,due to the inherited single-channel thermal path of the single-sided cooling power module,the traditional thermal model and measurement standard are challenged by the emerging DSC power module with multiple-channel thermal paths.As a result,the intrinsic thermal impedance of the DSC power module is missed.On the basis of the machine learning method,the intrinsic thermal impedance model of the DSC power module is proposed by using the neural network(NN).The concept of the intrinsic thermal impedance is clarified for the DSC power module.Besides,the feasibility and uniqueness of the reconfigured intrinsic thermal impedance are confirmed.Moreover,the difference between the traditional and proposed thermal impedances is assessed.Concerning the commercial DSC power module,the thermal impedance database of the DSC power module is created by using the multiphysics tool.Furthermore,the NN-based intrinsic thermal impedance of the DSC power module is trained and cross-checked with aid of the created massive training and testing datasets.Additionally,the measured and characterized thermal impedances of several commercial DSC power modules are demonstrated.In this way,the generalization and compatibility features of the proposed NN model to reconfigure the intrinsic thermal impedance are ensured.The findings might support the modeling and characterization of the multiple-channel thermal impedance with insightful research routine and technology solution.

关 键 词：功率模块 双面散热 本构热阻 多物理场分析 神经网络 

分 类 号：TM131.2[电气工程—电工理论与新技术]