A new formulation for predicting the ultimate capacities of FRP-confined concrete using advanced machine learning framework: Developed structural reliability analysis

Abstract

Maintaining the safety levels of concrete structures is critical, where boosting the strength and deformability of such components utilizing Fibre Reinforced Polymer (FRP) composites is well adopted. However, the outside environment and applied loads can influence the FRP confined concrete. Therefore, accurate assessment of the ultimate capacities of FRP-confined concrete is highly essential. This paper proposes an advanced machine-learning framework using a novel approach called Group Method Data Handling Neural Network (GMDH-NN) for developing novel formulations that properly predict the ultimate capacities of FRP-confined concrete. The GMDH-NN ability in terms of tendency, agreement, and accuracy is compared to the existing empirical correlations and well-known machine-learning methodologies, notably Artificial Neural Network (ANN) and Response Surface Method (RSM). Furthermore, to address the failure modes of strength and ductility, the reliability analysis of FRP-confined concrete using combined Monte Carlo simulation (MCS) with the proposed GMDH-NN is examined for three types of composite sheets: Aramid, Carbon, and Glass. Results prove the GMDH-NN outperformance for modeling the ultimate strength (R2 =0871) and strain (R2 =0.79) capacities for FRP-confined concrete. The structural reliability analysis indicated that the choice of FRP sheets had a substantial influence on the increase of the safety levels of confined concrete cores, as well as the requirement for early prediction of these parameters to avoid failures.