Amidst the evolving landscape of structural engineering, this study addresses the pressing need for a robust, data-driven model in the context of alternative reinforcement techniques such as Fiber-Reinforced Polymer (FRP) bars and the application of Machine Learning (ML) methodologies for predicting the punching shear strength (PSS). Structural engineering heavily relies on accurate predictions of PSS for ensuring the safety and durability of various constructions. However, existing models often lack the comprehensiveness and precision required to effectively address the diverse and complex factors influencing PSS in FRP-reinforced concrete structures. This research explored the shear strength of FRP-reinforced concrete slabs, with a particular focus on the different types of FRP, namely Carbon Fiber-Reinforced Polymer (CFRP), Basalt Fiber-Reinforced Polymer (BFRP), Glass Fiber-Reinforced Polymer (GFRP), and Hybrid Fiber-Reinforced Polymer (HFRP). Leveraging an innovative deep learning approach, the study utilizes a dataset comprising 285 samples gathered from 56 distinct studies to train models that incorporate 11 dependent variables. This comprehensive approach aims to capture the intricate interplay between geometric, material, and other pertinent factors influencing PSS. The study employs a connection weight method to evaluate the significance of various features and introduces a novel Graphical User Interface (GUI) to enhance user interaction with the Deep Neural Network (DNN) model designed for PSS prediction. Through meticulous experimentation and analysis, the research identifies an optimal DNN architecture that achieves a minimum Mean Absolute Error (MAE) of 0.81, validated through k-fold cross-validation. Key conclusions highlight the critical influence of specific features, such as slab depth, column-to-slab area ratio, and modulus of elasticity, on the predictive accuracy of the model. Conversely, certain factors like slab area, reinforcement ratio, and concrete compressive strength negatively impact predictions. SHAP analysis further elucidates the significance of FRP type and column length-to-width ratio as pivotal factors affecting PSS predictions. These findings underscore the potential of the developed model in enhancing the dependability and practicality of structural engineering applications. Moreover, they point towards future research directions aimed at refining predictive accuracy and exploring broader applications in the field of building engineering.
Read full abstract