计算机科学
环氧树脂
沥青
人工神经网络
机器学习
随机森林
混合(物理)
编织
近似误差
人工智能
多层感知器
均方预测误差
支持向量机
复合材料
极限学习机
Lasso(编程语言)
模数
剪切模量
决定系数
可靠性(半导体)
数据挖掘
算法
剪切(物理)
优化设计
动态模量
材料科学
实验数据
实验设计
沥青混凝土
性能预测
Boosting(机器学习)
万能试验机
启发式
弹性模量
作者
Ke Xu,Jian Xiong,Tengfei Zhao,Xinxian Li,Zhaodong Wang,Y F Zhang,Yutong Tan
标识
DOI:10.1016/j.procs.2026.05.195
摘要
To explore a novel approach for the performance-driven design of epoxy asphalt formulations, this study constructs a "performance-mix proportion" database for epoxy asphalt and proposes a matching design method based on performance requirements. Initially, using penetration, softening point, ductility, elastic recovery, 60°C dynamic viscosity, and complex shear modulus as inputs and mix proportions as outputs, three machine learning prediction models—Back Propagation Neural Network (BPNN), Extreme Gradient Boosting (XGBoost), and Random Forest (RF)—were developed based on small-sample experimental data. The optimal model was identified by comparing the coefficient of determination (R²). Subsequently, the small-sample "performance-mix proportion" database was expanded to 196,020 samples using the optimal model (XGBoost), and a MySQL database was established. Structured SQL queries were then employed to match mix proportions with desired performance criteria, proposing a data-driven design method for epoxy asphalt formulations. Finally, case studies validated the reliability of the method. The results indicate that XGBoost outperforms RF and BPNN with the highest R² (0.97). The relative error between case study predictions and laboratory tests was less than 3%, confirming the method’s accuracy. This research provides a significant reference for performance-driven epoxy asphalt mixed proportions design.
科研通智能强力驱动
Strongly Powered by AbleSci AI