微观力学机理与深度学习的融合方法——在ECC材料极限拉应变预测中的应用

Integrating Micromechanical Principles with Deep Learning—Application to Ultimate Tensile Strain Prediction of Engineered Cementitious Composites

摘要: 为克服传统水泥基材料依赖经验设计与重复试验、抗拉强度低且脆性易裂的问题，基于微观力学原理设计的高延性水泥基复合材料(ECC)显著提升了材料的延展性与控裂性能，其极限拉应变高出普通混凝土2~3个数量级，适用于工程结构抗震韧性提升。然而，ECC的配合比设计与性能优化仍依赖于试验测试与验证。因此，文章提出将深度学习(DNN)模型应用于ECC性能预测与验证，利用既有数据与微观力学原理筛选关键特征参数，构建DNN模型，实现对极限拉应变的准确预估，从而显著降低试验依赖。结果表明，该方法概念清晰、准确性高，对学生理论基础要求适中，适用于本科智能建造类课程教学，有助于培养学生跨学科解决实际工程问题的能力。

Abstract: To overcome the limitations of traditional cementitious materials, such as reliance on empirical design and iterative testing, low tensile strength, and brittle behavior, Engineered Cementitious Composites (ECC) have been developed based on micromechanical principles. ECC significantly enhances material ductility and crack control, with its ultimate tensile strain exceeding that of ordinary concrete by 2~3 orders of magnitude, making it highly suitable for improving seismic resilience in engineering structures. However, the mix design and performance optimization of ECC still heavily depend on experimental testing and validation. Therefore, this paper proposes the application of a Deep Neural Network (DNN) model for predicting and validating the mechanical properties of ECC. By leveraging existing experimental data and incorporating micromechanical principles to screen key feature parameters, a DNN model was constructed to accurately predict the ultimate tensile strain, thereby substantially reducing reliance on physical testing. The results demonstrate that the proposed method is conceptually clear and exhibits high predictive accuracy. With moderate requirements for students’ theoretical background, this approach is suitable for integration into undergraduate courses such as intelligent construction, fostering students’ ability to solve practical engineering problems through interdisciplinary means.