DEEP LEARNING-BASED FAULT DETECTION IN CIVIL ENGINEERING STRUCTURES: A COMPUTATIONAL MATHEMATICS APPROACH USING VIBRATION DATA

Abstract

Structural safety is a critical concern in civil engineering, as the failure of bridges, buildings, and other infrastructures can lead to catastrophic human and economic losses. Traditional fault detection methods, such as modal analysis and finite element simulations, have proven effective but are often limited by high computational demands, noise sensitivity, and difficulties in scaling to real-time monitoring. Recent advances in “deep learning (DL)” offer promising solutions for extracting fault signatures directly from vibration data, enabling automated, accurate, and scalable detection. This paper investigates a computational mathematics framework for vibration-based fault detection that integrates signal preprocessing, feature transformation, and deep neural architectures such as “convolutional neural networks (CNNs)”, “recurrent neural networks (RNNs)”, and hybrid models. Experimental results using benchmark vibration datasets show that deep learning models can achieve fault classification accuracies exceeding 95%, outperforming conventional machine learning approaches. Furthermore, the study highlights the role of mathematical techniques including wavelet transforms, Fourier analysis, and sparse representation in improving feature extraction and robustness against noise. The paper concludes that deep learning-based fault detection offers a paradigm shift in civil infrastructure monitoring, combining the rigor of computational mathematics with the predictive power of data-driven AI systems.