Deep Learning-Based Estimation of Thermal Micro-Crack Evolution in Ultra-High-Performance Concrete
Deep Learning-Based Estimation of Thermal Micro-Crack Evolution in Ultra-High-Performance Concrete
Amirmasoud Naderi1 Amirkhosro Najafi2 Hasan Mohammadi Dehcheshmeh3 Soran Karami4
1) M.Sc., Department of Civil Engineering (Geotechnical), Faculty of Engineering, Shahrekord University, Shahrekord, Iran.
2) M.Sc., Department of Civil Engineering (Geotechnical), Faculty of Engineering, Shahrekord University, Shahrekord, Iran.
3) M.Sc., Department of Civil Engineering (Structural), Farsan Branch, Islamic Azad University, Farsan, Iran.
4) M.Sc., Department of Civil Engineering (Structural), Faculty of Engineering, University of Kurdistan, Kurdistan, Iran.
IOI Code
:
ABEB-ACZKAB
Publication :
4th International Conference on Recent Advances in Engineering, Innovation and Technology - Belgium(eitconf.com)
Abstract :
In this study, the microstructural response of Ultra-High-Performance Concrete (UHPC) to elevated temperatures is investigated through deep learning and the automated analysis of FESEM imagery. To this end, a UHPC mixture was designed based on the modified Andreasen particle packing model. Following standard curing, the specimens were exposed to six temperature levels ranging from 100∘C to 600∘C, and their fracture microstructures were imaged across various magnifications. A semantic segmentation framework, built upon the U-Net architecture with a ResNet34 backbone, was then developed to automatically detect and quantify thermal micro-cracks in the captured images. Based on the network outputs, parameters such as crack density and width were extracted, and a “Microstructural Degradation Index” (IDB) was defined to characterize the progressive stages of thermal damage. The results indicate that the evolution of the micro-crack network in UHPC exhibits a three-phase, highly non-linear behavior. Upon exceeding the critical threshold of 300∘C–400∘C, both the IDB and crack density experience a rapid surge, driving the microstructure toward a state of near-complete collapse at 600∘C. The proposed framework successfully transitions the evaluation of FESEM imagery from a qualitative, operator-dependent process into a quantitative, reproducible, and automated system, offering a robust method for assessing the thermal durability of UHPC and other cementitious composites.
Keywords : ultra high performance concrete (uhpc) thermal micro cracking fesem imaging u net thermal behavior