Title:
Deep Learning Based Approach for Evaluating Concrete Surface Integrity through Crack Identification and Analysis
Author(s):
Flah
Publication:
Web Session
Volume:
ws_S25_Flah.pdf
Issue:
Appears on pages(s):
Keywords:
DOI:
Date:
3/30/2025
Abstract:
Current prescriptive guidelines for concrete mixture design typically focus on limited performance criteria, primarily strength requirements, which can restrict the adoption of new, more sustainable materials. This limitation arises from the narrow scope of design considerations, which often fail to effectively accommodate alternative materials. Such constraints imposed by existing specifications and assessment methodologies pose significant obstacles to the integration of novel materials into concrete design processes. However, recent advancements in machine learning have demonstrated its capability to accurately predict various properties and performance aspects of concrete based on mixture design data. This study proposes a machine learning framework to expedite the goal-oriented and performance-driven design of low-carbon concrete incorporating a wide range of supplementary cementitious materials (SCMs). Leveraging the substantial progress in concrete technology over recent decades, a comprehensive database was compiled from peer-reviewed scientific literature, containing information on concrete mix constituents, curing techniques, and relevant performance metrics including plastic, mechanical, durability, and environmental characteristics. Utilizing this dataset, predictive machine learning models were developed to estimate the engineering properties of low-carbon concrete incorporating a wide array of alternative SCMs as high-volume replacement for ordinary Portland cement. An inverse design approach was then adopted to optimally design different concrete mixtures to satisfy predefined specific performance criteria, showcasing the potential of the machine learning framework to accelerate the performance-based design of sustainable concrete mixtures.