基于神经网络的表面波频散曲线反演温度相关杨氏模量  

Inversion of temperature dependence of Young’s modulus from surface wave dispersion based on neural network

作　　者：贾昊天 刘祥恩 沈中华[1] LOMONOSOV Alexey JIA Haotian;LIU Xiang’en;SHEN Zhonghua;LOMONOSOV Alexey(College of Science,Nanjing University of Science and Technology,Nanjing 210094,China;General Physics Institute,Russian Academy of Sciences,Moscow 119991,Russian Federation)

机构地区：[1]南京理工大学理学院,南京210094 [2]俄罗斯科学院通用物理研究所,莫斯科119991

出　　处：《无损检测》2022年第7期19-22,28,共5页Nondestructive Testing

基　　金：国家自然科学基金资助项目(61975080)。

摘　　要：提出了一种应用神经网络从色散曲线反演材料弹性常数的温度依赖性的方法。采用有限元方法计算了毫秒激光加热铝材料形成的瞬态温度场。在假设不同杨氏模量温度依赖性的条件下,计算了表面波在激光加热区传播时不同的频散曲线。利用正向计算的结果来训练神经网络。神经网络经过训练后,输入材料的表面波频散特性,可反演出材料杨氏模量与温度的关系。为验证该方法的反演能力,对比了不同噪声情况下的反演结果。对比结果表明该方法具有很好的鲁棒性。A method was proposed to inverse the temperature dependence of the elastic constant of a material from the dispersion curve using a neural network. The finite element method was employed to calculate the transient temperature field formed by millisecond laser heating of aluminum material. Under the assumption of various temperature dependence of Young’s modulus, various dispersion curves of surface waves propagating in the laser heating zone were calculated. The results of the forward calculation were used to train the neural network. After the neural network was trained, by inputting the surface wave dispersion characteristics of the material, the temperature dependence of Young’s modulus of materials was inversed. In order to verify the inversion capability of this method, the inversion results under different noises conditions are compared. The inversed results show that this method has good robustness.

关 键 词：温度场 弹性模量的温度依赖性 分布不均匀 神经网络 

分 类 号：TG115.28[金属学及工艺—物理冶金]