Date of Award

6-8-2026

Thesis Type

PhD

Document Type

Thesis

Degree Name

Doctor of Philosophy (PhD)

Divisions

Faculty of Engineering

Department

Department of Civil Engineering

Institution

Universiti Malaya

Abstract

Concrete-Filled Steel Tube (CFST) have been extensively utilized in modern engineering structures due to their superior mechanical performance. However, existing code-based design methodologies often fail to accurately capture the nonlinear coupling effects among multiple parameters under complex loading conditions including axial compression, flexure, shear, and eccentric compression. Furthermore, there is a notable lack of systematic design optimization frameworks that simultaneously balance structural safety, economic cost, and environmental impact. To address these challenges, this study proposes a comprehensive framework integrating Artificial Intelligence (AI)-aided prediction with surrogate model-driven design optimization to achieve high-precision performance evaluation and intelligent design optimization for CFST structures. The research comprises five main phases starting with the Axial Ultimate Load capacity of circular CFST columns, where an improved XGBoost prediction model using an Asymmetric Squared Error Loss function was constructed to suppress overestimation risks, followed by a multi-objective optimization framework using the Non-dominated Sorting Genetic Algorithm. Subsequently, regarding Flexural capacity, an XGBoost model optimized by the Jellyfish Search Optimizer was proposed alongside interpretability analyses, revealing the trade-off relationship between capacity and cost through bi-objective optimization. Addressing Shear Strength, an algorithm-optimized XGBoost model incorporating quantile regression was developed to quantify uncertainty, while a simplified calculation formula was derived based on the Genetic Algorithm to improve engineering applicability. For the Ultimate capacity under eccentric load, the concept of Materials Genome Engineering was introduced to establish an inverse design framework based on a CatBoost surrogate model, utilizing a "residual bearing capacity vcoefficient" to directly deduce optimal design parameters. Finally, the research scope was expanded to circular Concrete-Filled Double-Skin Steel Tubes by constructing a Bat Algorithm optimized CatBoost model and conducting cost-oriented design optimization. The results demonstrate that the proposed AI-aided prediction models, with R² values generally exceeding 0.99, significantly outperform existing design codes or peer machine learning models, and the established optimization frameworks can directly guide the structural design of CFST, while the developed Graphical User Interface (GUI) provides robust support for efficient and safe practical applications.

Initial

khm

Additional Information

Thesis (PhD) - Faculty of Engineering, Universiti Malaya, 2026.

Download

Available for download on Thursday, June 08, 2028

Share

COinS