基于一维卷积残差网络的微热管几何结构及制造工艺参数预测  

Parameter prediction of micro heat pipe structure and manufacturing process based on a one-dimensional convolutional residual network

作　　者：李勇[1] 张靖昊 刘苑喆 高昂 陈昕宇 张泽华 LI Yong;ZHANG JingHao;LIU YuanZhe;GAO Ang;CHEN XinYu;ZHANG ZeHua(School of Mechanical and Automotive Engineering,South China University of Technology,Guangzhou 510640,China)

机构地区：[1]华南理工大学机械与汽车工程学院,广州510640

出　　处：《中国科学:技术科学》2025年第2期281-294,共14页Scientia Sinica(Technologica)

基　　金：广东省自然科学基金(编号:2023A1515011635)资助项目。

摘　　要：压扁型微热管在散热领域应用广泛,微热管传统设计方法依赖人工经验及多次重复实验以确定其几何结构及制造工艺参数,耗时费力.本文提出一种基于copula熵(copula entropy,CE)和一维卷积残差网络的微热管几何结构及制造工艺参数预测方法.通过功率测试设备收集微热管样品的传热性能数据,构建性能数据集;利用copula熵表征微热管结构设计需求参数及最大传热功率间关联程度,通过特征筛选获得微热管设计需求关键参数;通过训练一维卷积残差网络建立微热管设计需求关键参数与其几何结构及制造工艺参数映射关系.将该模型与传统人工神经网络进行对比,以微热管设计需求关键参数为输入且具有残差结构的一维卷积网络模型精度更高,在测试集上误差仅为0.66%.最后,根据实际微热管设计需求,对所提出方法进行验证,结果表明该方法能较好实现微热管几何结构及制造工艺参数预测.Flattened micro heat pipes have a wide range of applications in the field of heat dissipation.However,their traditional design relies on manual experience and several repeated experiments to determine their geometric structure and manufacturing process parameters,which is time-consuming and laborious.Thus,a method for predicting the geometric structure and manufacturing process parameters of micro heat pipes based on copula entropy(CE)and one-dimensional convolutional residual network is proposed.A micro heat pipe data set is established using micro heat pipe power testing equipment to perform performance tests on micro heat pipe samples.The degree of correlation between each micro heat pipe structural design requirement parameter and power is characterized by CE,and the key micro heat pipe design requirement parameter is obtained through feature screening.The mapping relationship between the key micro heat pipe design requirement parameter and geometric structure and manufacturing process parameters is established by training a one-dimensional convolutional residual network.The model is compared with a traditional artificial neural network.The one-dimensional convolutional network model with the key parameters of micro heat pipe design requirements as inputs and the residual structure is more accurate,with an error of only 0.66%on the test set.Finally,the proposed method is validated according to actual micro heat pipe design requirements,and the results show that the method can better predict micro heat pipe geometric structures and manufacturing process parameters.

关 键 词：微热管 参数预测 copula熵 一维卷积残差 

分 类 号：TK1[动力工程及工程热物理—热能工程]