Deep Learning-based Identification and Localization of Grout Failures in Post-tensioned Concrete Girders

Résumé fourni par la source

The structural integrity of post-tensioned concrete girders is vital for the safety and durability of civil infrastructure, particularly in bridge construction. Grout failures in post tensioned duct of these girders can significantly compromise their performance, making early detection crucial for effective maintenance and prevention of potential failures. This study presents a novel method for the identification and localization of grout failures in post-tensioned duct-grouted concrete girders, utilizing vibration-based analysis in conjunction with Deep Learning (DL) and Machine Learning (ML) techniques. A Finite Element (FE) modelling approach is employed to generate synthetic data that simulates the dynamic responses of girders subjected to various damage conditions, with an emphasis on grout-related failures and other non-grout related scenarios. The generated dataset serves as the basis for training a deep learning model to accurately detect and localize grout failures based on the vibrational patterns observed. The generated dataset serves as the basis for training a deep learning model to accurately detect and localize grout failures based on vibrational patterns. The model demonstrated exceptional performance, achieving 98.70% accuracy on training, testing, and validation datasets with a mean squared error (MSE) of 0.0001. To evaluate its real-world applicability, the framework was subjected to various levels of stochastic Gaussian noise and ambient fluctuations, under which it maintained a robust prediction accuracy of approximately 88.74%. These results demonstrate the model's stability and its ability to transition from numerical simulations to noisy environmental conditions, ensuring structural integrity prior to girder erection. Additionally, 5-fold cross-validation with a standard deviation of 0.0023 demonstrates that the model performance is stable and not dependent on the training dataset, indicating its robustness and generalizability. These results highlight the potential of deep learning-driven approaches, combined with vibration-based analysis and FE modelling, for improving damage detection in post-tensioned duct-grouted concrete girders before their launch into position, ensuring their structural integrity prior to erection. The suggested framework was validated using experimentally tested beams with different grouting conditions, demonstrating its potential for effective damage detection, and supporting maintenance strategies to ensure the safety and integrity of critical infrastructure.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.