摘要
选用核电用钢A508为研究对象,对K近邻回归(KNN)、核回归(KR)、线性回归(LR)和随机森林回归(RF)四种算法拘束相关断裂韧性的预测能力进行了考察,发现四种算法对拘束相关断裂韧性的预测效果为RF>LR>KNN>KR.进一步,基于RF算法,在数据中加入了平面应变状态下的数据进行数据增强,对单边缺口弯曲(SENB)试样的拘束相关断裂韧性进行了预测与验证.并基于验证过的模型,向单边裂纹拉伸(SENT)、紧凑拉伸(CT)和中心裂纹拉伸(CCT)试样进行了移植.结果表明:通过添加数据增强策略,试样在边界点处的断裂韧性预测能力得到有效提高,预测结果更加准确,基于RF算法并添加数据增强策略的模型具有较好的泛化能力.最后,在基于RF算法和数据增强的基础上,加入试样类别这一特征,构建了准确度较高的统一预测模型.
The study of data-driven predictions for constraint-related fracture toughness is an interdisciplinary scientific problem relevant to mechanics,mechanical engineering,as well as computer science and technology,and is of great significance for accurate structural integrity assessment.This research focused on nuclear power steel A508.The predictive capabilities of four algorithms,namely the K-nearest neighbors(KNN)regression,kernel regression(KR),linear regression(LR),and random forest(RF)regression,for constraint-related fracture toughness predictions were investigated.The RF algorithm outperformed the others,while the KR algorithm had the least effective predictions.The prediction accuracy ranked as follows:RF>LR>KNN>KR.Furthermore,based on the RF algorithm,data under plane strain conditions were added for data enhancement,enabling the prediction and verification of constraintrelated fracture toughness for single-edge notch bending(SENB)specimens.The validated model was successfully transplanted to single-edge notch tension(SENT),compact tension(CT),and central crack tension(CCT)specimens.Results indicated that the RF algorithm with data augmentation improved prediction accuracy and capability,particularly at boundary points.The RF-based model,enhanced with additional data strategies,demonstrated strong generalization across different specimen types.For SENB and CT specimens,bending loads dominate at the crack tip;thus,altering a/W and B/W enhances restraint.For SENT and CCT specimens,where shear loads predominate at the crack tip,adjusting a and B proves more effective.Finally,a unified,high-accuracy prediction model was developed by incorporating sample category features using the RF algorithm and data enhancement strategies.
作者
单康中
王骁晓
刘芳
崔元元
杨杰
Kangzhong Shan;Xiaoxiao Wang;Fang Liu;Yuanyuan Cui;Jie Yang(Shanghai Key Laboratory of Multiphase Flow and Heat Transfer in Power Engineering,School of Energy and Power Engineering,University of Shanghai for Science and Technology,Shanghai,200093;Key Laboratory of Pressure Systems and Sa fety,Ministry of Education,East China University of Science and Technology,Shanghai,200237;School of Mechanical Engineering,University of Shanghai for Science and Technology,Shanghai,200093)
出处
《固体力学学报》
北大核心
2025年第1期105-116,共12页
Chinese Journal of Solid Mechanics
基金
国家自然科学基金项目(Nos.52375154和52311530067)资助.
关键词
拘束
断裂韧性
机器学习
随机森林算法
数据增强
constraint
fracture toughness
machine learning
random forest algorithm
data enhancement
