基于神经网络同步学习的功率模块散热器拓扑优化快速迭代方法  

Topology Optimization and Fast Iterative Method for Power Module Heat Sink Based on Neural Network Synchronous Learning

作　　者：朱高嘉 何函宇 李龙女 朱建国 梅云辉 ZHU Gaojia;HE Hanyu;LI Longnu;ZHU Jianguo;MEI Yunhui(School of Electrical Engineering,Tiangong University,Tianjin 300387,China;School of Electrical and Information Engineering,University of Sydney,Sydney 2006,Australia;CPSS)

机构地区：[1]天津工业大学电气工程学院,天津300387 [2]悉尼大学电气与信息工程学院,悉尼2006 [3]中国电源学会

出　　处：《电源学报》2024年第3期111-117,共7页Journal of Power Supply

基　　金：国家自然科学基金资助项目(52177189);天津市杰出青年基金资助项目(21JCJQJC00150);教育部“春晖计划”国际合作科研项目(HZKY20220604)。

摘　　要：随着功率模块集成化程度的提高,其散热结构优化已成为研发中的关键。拓扑优化可通过变换散热器形貌、结构来最大化地提升散热效果,因此受到了广泛关注。但在拓扑优化过程中,每步迭代均需要计算模块与散热器温度分布,占用较庞大的计算资源和计算时间。为加速传统散热器拓扑优化进程,在基于传统固体各向同性材料惩罚SIMP(solid isotropic material with penalization)散热器拓扑优化方法的基础上,提出一种嵌套神经网络NN(neural network)同步学习的快速迭代方法。首先,构建散热器基于编码器-解码器结构的NN预测模型,即基于散热器形貌迭代进化过程实现优化结构的快速预测;其次,将NN模型与散热器SIMP拓扑优化流程相嵌套,利用迭代过程中的中间形貌同步训练NN;最后,针对单芯片、两芯片模块结构,对比所提方法与传统迭代方法的拓扑优化结果,验证了所提NN同步学习方法的准确性和快速性。With the improvement of the integration degree of power modules,the optimization of their heat transfer structures has become a focus in the development.The topology optimization(TO)can maximize the cooling performance by transforming the morphology and structure of heat sinks,thus receiving extensive attention.However,in the TO process,the temperature distribution of modules and heat sinks needs to be calculated in each iteration step,consuming a large amount of computing resource and calculation time.To accelerate the TO process of traditional heat sinks,a fast iterative method combining neural network(NN)synchronous learning and the traditional solid isotropic material with penalization(SIMP)-based TO methods is put forward.First,an NN prediction model based on the encoder-decoder structure is constructed,which can iteratively evolve the shape of heat sinks to achieve a fast prediction of optimized structures.Second,the NN model is integrated into the TO process of the heat sink based on the SIMP method,and the NN is trained synchronously using the intermediate morphology obtained in the iteration process.Finally,aimed at the single-chip and dual-chip modules,the results obtained by the new method and traditional iterative methods are compared to validate the accuracy and rapidity of the proposed NN synchronous leaning method.

关 键 词：散热器结构优化设计 拓扑优化 变密度法 神经网络同步深度学习 

分 类 号：TN386.1[电子电信—物理电子学]