Title:
Predicting Condition of Reinforced Concrete Beams with Shear Cracks using Machine Learning
Author(s):
Castillo
Publication:
Web Session
Volume:
ws_S23_Castillo.pdf
Issue:
Appears on pages(s):
Keywords:
DOI:
Date:
4/2/2023
Abstract:
Shear failures in reinforced concrete structures occur with little or no warning; therefore, members with shear cracks need to be evaluated to ensure safety. Existing evaluation methods for members with shear cracks have large variability in results, require time-consuming models or rely on expert opinion. This study uses machine learning to investigate correlations between crack width and load, stiffness, and stirrup strain histories. A shear test database is assembled for rectangular slender beams with more than six hundred crack width measurements, along with other measured data, including load-displacement relationship and stirrup strains. A Gaussian Process Regression model is used to predict crack width from shear history, stiffness, and stirrup strains. Beam design details were considered in the model design. The three outcomes (shear history, stiffness, and stirrup strain) offer information on the shear condition of a reinforced concrete beam for rapid evaluation of in-service structures. Ten-fold cross validation shows that the mean absolute percentage error is 18%, 33% and 77% for predicting shear history, stiffness and stirrup strain, respectively.