基于神经网络的高温反应堆蠕变本构模型研究  

Research on Creep Constitutive Model for High Temperature Reactor Based on Neural Network

作　　者：张帆 戴守通[1] ZHANG Fan;DAI Shoutong(Department of Nuclear Engineering Design,China Institute of Atomic Energy,Beijing 102413,China)

机构地区：[1]中国原子能科学研究院核工程设计研究所,北京102413

出　　处：《原子能科学技术》2024年第12期2581-2591,共11页Atomic Energy Science and Technology

摘　　要：新型反应堆的设计温度越来越高,寿命越来越长,其金属构件的蠕变现象越发显著。蠕变分析需要基于实验数据建立精确的本构模型,但获取长时实验数据代价高昂,因此需要研究由短时实验数据预测长时蠕变性能的可靠模型。本文使用不锈钢3000 h的蠕变实验数据,先进行降噪处理,结合蠕变的物理特性找到蠕变第一、二阶段分界点,然后基于人工神经网络和Norton-Bailey幂律对实验数据进行分段训练和反向参数标定,最终得到优化的高温蠕变本构模型。经过与实验数据对比可知,该模型既能够准确描述材料蠕变变形行为,又能预测长达10000 h的蠕变变形,且预测结果与实验数据的相对误差不超过3%,其精确度远高于单一使用蠕变理论或神经网络训练得到的结果。本文结果为反应堆结构高温蠕变本构模型研究提供了有效的新方法,对高温反应堆蠕变分析和寿命评价具有理论指导意义和工程参考价值。This research aims to develop a novel creep constitutive model,leveraging neural networks to precisely predict long-term creep behavior from short-term data.The new types of nuclear reactors require higher design temperatures and longer service life than before.Consequently,the investigation of the long-term creep in metallic components utilized in high temperature environments gains greater importance.The analysis of creep necessitates precise constitutive models which need a substantial quantity of long-term experimental data.However,the acquisition of such data through experiments incurs considerable costs,thus the need is highlighted to develop a dependable model that can forecast the long-term creep behavior of materials based on short-term creep test data.What’s more,traditional approaches often struggle with the precision required to capture the nonlinear deformation mechanisms of the primary creep,as well as difficulties in parameter calibration.At present,the amalgamation of artificial neural network(ANN)techniques in the creep of reactor design lacks any established precedent.In this paper,an ANN model that characterizes the creep deformation properties of materials and predicts the long-term creep deformation behavior was presented.This approach eliminates the need to prematurely define a cut-off point between the primary and secondary creep stages,which can lead to overestimating the slope of the second stage and introduce challenges in engineering design.Other key highlight of this research is the introduction of a machine learning-based method to identify the cut-off point without compromising the accuracy of the primary creep stage prediction.This approach ensures that the model captures the true creep behavior throughout all stages,resulting in more reliable predictions of material performance and lifetime.After identifying the cut-off point between the primary and secondary creep stages,neural networks were used for fitting in the primary creep.This allowed the model to capture the intricate de

关 键 词：人工神经网络 Norton-Bailey幂律 蠕变本构模型 蠕变分界点 蠕变性能预测 

分 类 号：TL341[核科学技术—核技术及应用]