In geotechnical engineering, the mechanical properties of interfaces between geological bodies and engineering structures represent a critical research focus, particularly in seismically active regions where damage at rock–concrete interfaces poses severe threats to structural safety. Accurately identifying damage modes at rock–concrete interfaces under complex loading conditions remains a significant challenge, primarily due to the limitations of conventional methods in distinguishing mixed tensile–shear fracture mechanisms. This study adopts an integrated experimental and machine learning framework that combines three-point bending (TPB) tests, acoustic emission (AE) technology monitoring, and a Gaussian mixture model–support vector machine (GMM–SVM) algorithm. Unlike the RA–AF analysis method reliant on empirical parameters, this approach enables more precise discrimination between tensile and shear damage modes. Experimental results demonstrate that the crack–depth ratio significantly influences AE hit counts under monotonic loading, while distinct Kaiser and Felicity effects emerge under cyclic loading, revealing characteristic damage accumulation behaviors. The quantitative damage identification method based on the GMM–SVM model effectively characterizes damage types at the interfaces of rock–concrete specimens, thereby addressing the shortcomings of traditional analytical methods that depend on empirical parameters. By correlating the roughness of natural rock fracture surfaces with the fracture performance of specimens, this research establishes a connection between the morphological characteristics of natural rock fracture surfaces and mechanical properties. This study not only provides a more reliable tool for damage identification in composite interfaces, but also offers insights into the failure mechanisms of rock–concrete composites, contributing to enhanced safety assessment and design of underground structures in high seismic-risk regions.Highlights<UnorderedList Mark="Bullet"> <ItemContent> Machine learning-based method accurately classified tensile–shear damage of interfaces. </ItemContent> <ItemContent> Granite substrates with natural fracture surfaces were prepared using artificial splitting for roughness studies of rock–concrete interfaces. </ItemContent> <ItemContent> Fracture energy was analyzed via roughness of naturally fractured rock–concrete interfaces. </ItemContent> </UnorderedList>

Machine Learning-Based Study on Fracture Behavior and Damage Identification of Naturally Fractured Rock–Concrete Interfaces

期刊论文
发表日期：

2025年11月6日

查看全文

Yuzhu Guo,
Zhixuan Gao,
Xudong Chen,
Xiangyi Zhu,
Yuntian Wang

摘要

In geotechnical engineering, the mechanical properties of interfaces between geological bodies and engineering structures represent a critical research focus, particularly in seismically active regions where damage at rock–concrete interfaces poses severe threats to structural safety. Accurately identifying damage modes at rock–concrete interfaces under complex loading conditions remains a significant challenge, primarily due to the limitations of conventional methods in distinguishing mixed tensile–shear fracture mechanisms. This study adopts an integrated experimental and machine learning framework that combines three-point bending (TPB) tests, acoustic emission (AE) technology monitoring, and a Gaussian mixture model–support vector machine (GMM–SVM) algorithm. Unlike the RA–AF analysis method reliant on empirical parameters, this approach enables more precise discrimination between tensile and shear damage modes. Experimental results demonstrate that the crack–depth ratio significantly influences AE hit counts under monotonic loading, while distinct Kaiser and Felicity effects emerge under cyclic loading, revealing characteristic damage accumulation behaviors. The quantitative damage identification method based on the GMM–SVM model effectively characterizes damage types at the interfaces of rock–concrete specimens, thereby addressing the shortcomings of traditional analytical methods that depend on empirical parameters. By correlating the roughness of natural rock fracture surfaces with the fracture performance of specimens, this research establishes a connection between the morphological characteristics of natural rock fracture surfaces and mechanical properties. This study not only provides a more reliable tool for damage identification in composite interfaces, but also offers insights into the failure mechanisms of rock–concrete composites, contributing to enhanced safety assessment and design of underground structures in high seismic-risk regions.

Highlights

Machine learning-based method accurately classified tensile–shear damage of interfaces.

Granite substrates with natural fracture surfaces were prepared using artificial splitting for roughness studies of rock–concrete interfaces.

Fracture energy was analyzed via roughness of naturally fractured rock–concrete interfaces.