ScholarWorks@성균관대학교

초록

Monitoring residual tensile force (RTF) in prestressed tendons is critical for ensuring structural integrity throughout all construction stages. Existing EMI sensor-based methods face significant limitations, including the need for extensive initial calibration through lab-scale or mock-up tests, and restricted applicability to specific target materials due to the nonlinear relationship between tension and electromagnetic response. These limitations lead to inefficiencies in both time and cost. This study aims to overcome these challenges by developing a novel transformation framework for EMI sensors that eliminates the need for initial calibration, enhancing versatility across various arrangement conditions. A finite element (FE) framework was used to determine magnetic responses in the nonloaded state, with its accuracy verified through experiments. Furthermore, machine learning (ML) and deep learning (DL) models were implemented to address the complexity of the nonlinear relationship between electromagnetic response and tension. Seven models were used, including XGBoost, Feedforward Neural Network (FFNN), TabNet, and other commonly adopted ML models for solving engineering problems. Among them, XGBoost, FFNN, and TabNet demonstrated superior prediction accuracy, with XGBoost achieving a mean absolute error (MAE) of 0.845, FFNN reaching 0.813, and TabNet reaching 1.124. The findings indicate that the developed framework effectively predicts RTF without relying on costly and time-consuming calibration procedures, providing a cost-effective and reliable solution for monitoring prestressed tendons in various arrangement conditions. © 2025 Elsevier Ltd

키워드