Title:
Modeling of Peak and Ultimate Stress-Strain States in Confined Ultra-High-Performance Concrete (UHPC) Using Novel Hybrid Machine Learning Approach with Conditional Tabular Generative Adversarial Network
Author(s):
Tadesse Wakjira
Publication:
Web Session
Volume:
Issue:
Appears on pages(s):
Keywords:
DOI:
Date:
10/29/2023
Abstract:
Owing to its outstanding physical and mechanical properties, as well as sustainability, ultra-high-performance concrete (UHPC) has emerged as a highly desirable material for use in concrete structures. To realize its potential in large-scale structural applications, a comprehensive understanding of the compressive behavior of confined UHPC is essential. A significant number of analytical studies have focused on predicting the peak and ultimate conditions of unconfined UHPC. In comparison, a comprehensive framework for stress-strain characterization of confined UHPC has yet to be established, making efficient design and wider use of UHPC, particularly in seismic areas, challenging. Developing an accurate, design-oriented model for UHPC confined with either normal or high-strength steel is crucial for ensuring the safe design and modelling of UHPC structures. This study offers a solution by introducing a novel framework that integrates a novel hybrid machine learning (ML) model and a state-of-art-the art conditional tabular generative adversarial network (CTGAN) for accurate prediction of the stress-strain response of UHPC confined with either normal strength or high strength steel. An existing database of confined UHPC stress-strain response was compiled from the literature. To overcome the problem of limited database, CTGAN was developed by conditioning on the compiled experimental dataset to sample hypothetical synthetic confined UHPC samples. The generated synthetic data was then utilized to develop a novel hybrid ML model for predicting the stress-strain response of confined UHPC at peak and ultimate conditions. The predictive accuracy of the proposed hybrid ML model is benchmarked against various ML models with varying complexity and prediction accuracy. The results demonstrated that the proposed ML model is the most accurate and robust model for predicting the stress-strain response of confined UHPC with either normal strength or high strength steel.