Calibrating equations to predict the compressive strength of FRP-Confined columns using optimized neural network model

Abstract

A procedure based on machine learning algorithms is used to develop design equations for the compressive strength of FRP confined columns. The procedure involved training and optimizing an Artificial Neural Network (ANN), and using the parametric variation of each input of the ANN to investigate nonlinear relationships between input parameters and model output. Results of the parametric study were illustrated using Partial Dependence Plots (PDP), and a polynomial regression model (PRM) was calibrated to offer a simple equation to calculate compressive strength of columns confined with FRP sheets. The ANN and PRM developed using this method provided more accurate and precise estimates of compressive strength of retrofitted columns than other models found in the literature. Moreover, investigating the relationships between each input variable and the output of the ANN model and coupling effects between pairs of input variables on the output of the ANN model using Partial Dependent and contour plots revealed that confined compressive strength was insensitive to the elastic modulus of the composite layers with the thicknesses of 2 mm or less. A new equation was recalibrated for specimens with composite layer thickness greater than 2 mm, which provided very accurate estimates of confined compressive strength for this empirical subset.