Title:
Physics-Guided Bayesian Framework for Bond Strength Modeling with Explicit Aleatoric–Epistemic Uncertainty Quantification (Prepublished)
Author(s):
Hung La and Tan Nguyen
Publication:
Structural Journal
Volume:
Issue:
Appears on pages(s):
Keywords:
aleatoric and epistemic uncertainty; bayesian calibration; bond strength; physics-informed modeling; reinforced concrete corrosion; reliability-based assessment
DOI:
10.14359/51751827
Date:
7/1/2026
Abstract:
Accurate prediction of bond strength between steel reinforcement and concrete is essential for assessing structural safety and durability, particularly under reinforcement corrosion, where existing design codes such as ACI 408R-03 lack explicit modifiers. This study presents a physics-guided Bayesian framework that integrates dimensional analysis, probabilistic calibration, and uncertainty quantification to model bond strength while explicitly accounting for experimentally observed factors, including corrosion. Starting from a dimensionless power-law derived via Buckingham’s π-theorem, bond strength is expressed with globally interpretable exponents for embedment-to-diameter and cover-to-diameter ratios, while the scaling coefficient is adaptively modeled as a nonlinear function of experimental variables—including corrosion level, stirrup ratio, rebar type, and test method—through Random Fourier Features. Bayesian inference with Markov Chain Monte Carlo enables calibrated predictions with explicit decomposition of aleatoric and epistemic uncertainty, providing transparent insights into variability sources. Model performance and uncertainty behavior are examined through cross-validation and external validation. Beyond predictive performance, posterior analysis yields a concise, physics-consistent bond strength equation that explicitly incorporates corrosion effects and quantifies uncertainty, providing a practical and interpretable tool for reliability-based assessment of reinforced concrete structures.