Machine learning prediction of structural response for FRP retrofitted RC slabs subjected to blast loading

Abstract

Explosive and impact events, which in recent years have inflicted colossal human and economic losses, are dire warnings that civil infrastructure is not immune to blast and explosion scenarios. Thus, designing resilient new civil infrastructure and retrofitting the existing one to enhance its blast resistance is paramount. Retrofitting reinforced concrete (RC) slabs using external fiber-reinforced polymer (FRP) is known to enhance blast resistance, mitigate displacements and cracking, and contain debris. However, existing approaches to simulate the structural response to blast loading require competent knowledge, substantial modeling efforts, and high computational cost. Therefore, this study investigates the practicality of deploying machine learning to predict the maximum displacement of FRP strengthened RC slabs under blast loading as a novel approach to achieve simplified and accurate predictions. A Gaussian process regression algorithm was implemented for model development considering several influential features of the application. Due to the limited pertinent data in the open literature, a novel approach based on Tabular Generative Adversarial Network (TGAN) was used to generate 200 additional synthetic data used for model training. A design parameter prediction model was also proposed to predict the cross-section of FRP retrofitting considering blast parameters and the displacement specified by the design code. Statistical performance metrics including MAE, MAPE, and R2 indicate that the developed model achieved predictions with superior accuracy. Feature importance analyses were also conducted and corroborated by existing experimental and numerical studies. Based on the proposed model and its validation and feature importance analyses, the implementation of ML was proven to be a viable approach for structural response predictions under blast loading. Thus, the proposed model can potentially provide designers with accurate results for FRP retrofitting of RC slabs against blast loading through a highly simplified approach at low computational cost.