Prediction of Ultimate Strain in Anchored Carbon Fibre-Reinforced Polymer (CFRP) Laminates using Machine Learning

Abstract

Anchoring carbon fibre-reinforced polymer (CFRP) laminates to concrete using CFRP spike anchors effectively mitigates the unfavorable debonding failure mode in strengthened concrete beams. However, the strain and strength enhancement resulting from anchoring CFRP laminates have not been thoroughly quantified in the literature and existing practice codes. This study investigates the prediction of ultimate strain in anchored CFRP laminates, which is critical for assessing the flexural strength of strengthened concrete beams. Statistical regression analysis and machine learning models are employed to develop a predictive equation for the ultimate strain in CFRP laminates due to anchorage, using data from prior flexural tests on concrete prisms. The study examines various parameters, including CFRP sheet width, anchor design details (such as diameter and embedment depth), number of CFRP layers and anchor-to-sheet material ratio. Linear regression models, Support Vector Regression and Decision Trees were tested and compared for their accuracy in predicting the ultimate strain in CFRP laminates. The linear regression model, with highest performance indicators, was selected. Additionally, the study provides derived predictive equations, offering a practical implication for design optimization. Finally, sets of design charts were proposed to achieve specific values of ultimate strain in CFRP-strengthened and anchored concrete beams.