Title:
Deep-Learning-Informed Design Scheme for Prediction of Interfacial Concrete Shear Strength
Author(s):
Tarutal Ghosh Mondal, Nikkolas Edgmond, Lesley H. Sneed, and Genda Chen
Publication:
Structural Journal
Volume:
122
Issue:
1
Appears on pages(s):
51-62
Keywords:
deep learning; interfacial shear strength; learning-informed design; neural additive models; neural network; reinforced concrete; shear friction.
DOI:
10.14359/51743291
Date:
1/2/2025
Abstract:
Current design provisions pertaining to the shear transfer strength
of concrete-to-concrete interfaces, including those of the AASHTO
LRFD design specifications and ACI 318 Code, are based on
limited physical test data from studies conducted decades ago.
Since the development of these design provisions, many studies
have been conducted to investigate additional parameters. In addition, modern concrete technology has expanded the range of materials available and often includes the use of high-strength concrete and high-strength reinforcing steel. Recent studies examined the applicability of current shear-friction design approaches to interfaces that comprise high-strength concrete and/or high-strength steel and identified a need for revision to the existing provisions. To this end, this study leveraged a comprehensive database of test results collected from the literature to propose a deep-learningbased predictive model for normalweight concrete-to-concrete interfacial shear strength. Additionally, a new computation scheme is proposed to estimate the nominal shear strength with a higher prediction accuracy than the existing AASHTO LRFD and ACI 318 design provisions.