基于机器学习的蠕变断裂寿命预测方法  

Creep rupture life prediction method based on machine learning

作　　者：张效成[1] 宫建国[1] 轩福贞[1] ZHANG Xiaocheng;GONG Jianguo;XUAN Fuzhen(Key Laboratory of Pressure Systems and Safey,School of Mechanical and Power Engineering,East China University of Science and Technology,Shanghai 200237,China)

机构地区：[1]华东理工大学机械与动力工程学院,承压系统与安全教育部重点实验室,上海200237

出　　处：《压力容器》2021年第7期48-57,共10页Pressure Vessel Technology

基　　金：国家自然科学基金项目“基于超声非线性理论的高温结构损伤早期表征与评价基础研究”(51835503);“蠕变-疲劳载荷下高温构件不连续部位的损伤机理及设计准则研究”(51605165)。

摘　　要：工程结构的寿命预测问题是结构完整性领域的重要课题。通常,工程问题的解析解难以获得、经验解无法描述高维变量之间的非线性交互,而机器学习可以克服两者的局限性,是实现工程部件寿命预测的重要解决方案。基于此,提出了面向工程结构寿命预测问题的机器学习方案及优化的一般性方法。以316奥氏体不锈钢蠕变断裂寿命预测问题为例,基于高维数据集,使用BP神经网络、支持向量机、随机森林、高斯过程回归等机器学习模型,对316奥氏体不锈钢材料的蠕变断裂寿命进行预测。结果表明,基于机器学习的蠕变断裂寿命预测方法较之传统Larson-Miller方法具有更高精度。Life prediction of engineering structure is an important topic in structural integrity field.Usually,it is difficult to obtain analytical solution and empirical solution cannot describe the nonlinear interaction between high dimensional variables for a complex problem,whereas machine learning is an important solution for life prediction of engineering structure as it can overcome the limitation of both methods.Based on this,the machine learning solution oriented to life prediction problem of engineering structure and general method for optimization of it were proposed.Taking the creep rupture life prediction of 316 austenitic stainless steel as an example,based on high dimensional dataset,and by using machine learning models such as BP neural network,support vector machine,random forest and Gaussian process regression,the creep rupture life of the steel was predicted.Results show that creep rupture life prediction method based on machine learning could obtain higher accuracy than the conventional Larson-Miller method.

关 键 词：机器学习 蠕变断裂寿命 神经网络 支持向量机 

分 类 号：TH142[一般工业技术—材料科学与工程] TG111.8[机械工程—机械制造及自动化]