Evaluation of machine learning method on predicting shear strength of reinforced concrete beams with and without shear reinforcements

• Main Author: Ang, Shao Shen
• Other Authors: Li Bing
• Format: Final Year Project
• Language: English
• Published: Nanyang Technological University 2024
• Subjects: Engineering
• Online Access: https://hdl.handle.net/10356/181647
• Institution: Nanyang Technological University

This study investigates the application of machine learning (ML) techniques to enhance the accuracy of shear strength prediction in reinforced concrete (RC) beams. Traditional design codes, such as ACI 318-14 and Eurocode 2 (EC2), often provide conservative estimates due to their reliance on empirical models that fail to capture the intricate, nonlinear relationships between critical parameters like shear span-to-depth ratio, effective depth, and reinforcement configuration. To address this limitation, we trained multiple ML algorithms, including Ridge Regression, K-Nearest Neighbours (KNN), Gaussian Process Regression (GPR), Support Vector Regression (SVR), and Random Forest, on a comprehensive dataset of 775 experimental observations encompassing both stirrup-reinforced and unreinforced beams. The results demonstrate that machine learning (ML) models, specifically Extra Trees (ET) for beams with stirrups and Random Forest (RF) for beams without stirrups, significantly outperformed traditional methods. These ML models achieved prediction errors below 20% in most cases, while the ACI and EC2 codes exhibited average errors exceeding 30%. Although ACI generally showed better accuracy for beams with stirrups, and EC2 performed better for beams without stirrups, both codes often underestimated the actual behaviour, leading to conservative designs that may be inefficient. The study concludes that ML models offer a promising avenue for more accurate and efficient reinforced concrete (RC) beam design. By capturing complex relationships between variables that conventional models may overlook, ML can enhance the precision and safety of RC structures. Integrating ML into structural design practices has the potential to reduce material usage and costs. However, further validation with larger datasets and careful consideration of safety margins are essential before widespread adoption..