Implementation of Machine Learning in Predicting Pin-Bearing Strength of Aged and Nonaged Pultruded GFRP Composites

Abstract

Glass fiber–reinforced polymer (GFRP) composites are widely used materials in construction. Their use in structural applications, especially in civil infrastructure, requires structural analysis and design of their connections with proper safety and reliability levels. This work implements complex machine learning tools and schemes on a large amount of pin-bearing strength data on pultruded GFRP materials from the literature to extract correlations between various input parameters that affect the resulting bearing strength. This is performed to bridge the gaps in the findings that are associated with these data and to verify the recommended formulas in the corresponding design manuals and codes that are used for these materials in construction. The results of the derived gradient boosting model (GBM) could be used to predict the bearing strength of any pultruded composite without implementing experimental testing, therefore reducing operational costs and time. The derived model was exploited to investigate the effect of the hole diameter, the angle between the applied load and the pultrusion direction, and the effect of service temperature on the bearing strength of pultruded composites. The results imply the need to modify the used formulas in design standards; therefore, the proposed modifications are highlighted to properly predict the bearing strength for composite materials to be used in construction.