Using crack width for shear, stiffness, and stirrup strain history predictions for reinforced concrete beams

Abstract

Shear failures in reinforced concrete structures occur with little or no warning. Reinforced concrete members with shear cracks should be evaluated to ensure safety. Existing evaluation methods have large variability, require time-consuming modeling or expert opinion. This study uses machine learning to investigate correlations of crack width with shear loading, stiffness, and stirrup strain histories. Experimental literature enables the assembly of a database of rectangular reinforced concrete slender beams with crack width measurements for beams with shear reinforcement amounts smaller and larger than the minimum required by ACI 318-19. Measured data include crack widths from 122 beams, load–displacement relationship from 100 beams, and stirrup strains from 46 beams. Gaussian Process Regression is used to correlate crack width and geometric, material and design properties to shear loading, stiffness, and stirrup strain histories. Ten-fold cross validation training shows mean absolute percent errors of 18, 33 and 77% for shear, stiffness, and stirrup strain history predictions. The proposed algorithms can be used to accelerate the evaluation of in-service structures and can be updated upon availability of additional data.