A Study Of Machine Learning Applications For Solving Problems In Structural Engineering

Abstract

Structural engineering, a sub-discipline of civil engineering, involves the design and analysis of structures. The goal of structural engineering is to ensure that these structures are safe, stable, and capable of performing their intended functions throughout their lifespan. Meanwhile, machine learning (ML), a subset of artificial intelligence, utilizes statistical methodologies to learn and generate predictions or decisions directly from data. This dissertation explores the integration of these two fields, offering innovative approaches to solve structural engineering problems through the application of ML.The work comprises three projects that showcase the application of ML in the domain of structural engineering. The first project focuses on the predicting structural strength of steel circular hollow section (CHS) X-joints. Using support vector machines and deep neural networks, this project demonstrates how machine learning can effectively manage structural strength prediction tasks, pointing towards a promising future for ML in this field. The second project ventures into seismic risk analysis, a crucial part of structural safety evaluations. The use of advanced ML algorithms, including the discussion of hyperparameter tuning and model optimization, allowed for a more efficient and accurate prediction of seismic impact on structures such as railway bridges. The last project adopts machine learning for structural damage detection, using pre-trained convolutional neural networks tailored for image-oriented input. Key structural dynamic properties are transcribed into scalograms via wavelet transform, serving as training samples for the machine learning model. The promising outcomes from this project endorse the potential of machine learning in augmenting the efficiency and accuracy of processes for detecting and evaluating structural damage. Throughout the dissertation, a progressive learning journey unfolds, detailing how the understanding and application of ML evolved from basic techniques to more advanced methodologies. Each project enhances the subsequent one, demonstrating a continuous improvement in the application and understanding of ML. This research demonstrates that machine learning can provides new perspectives and methods for tackling topics in structural engineering, greatly enhancing the efficiency and effectiveness of problem solving in the field. The integration of ML can circumvent the complex experiments, simulations, and calculations that are typically required in structural design and analysis. The work encourages future discussion in the field, refining ML applications, exploring more innovative techniques, and ultimately continuing to push the boundaries of what can be achieved in structural engineering.