Artificial Intelligence (AI) technology has emerged as a powerful tool for addressing various complex issues within engineering structures. Accurately assessing the fire resistance performance of Concrete-Filled Steel Tube (CFST) columns poses a significant challenge for researchers. The scarcity of CFST column fire resistance test data presents a hurdle in the development of Machine Learning (ML) models. To tackle this issue, this study proposes a novel ML approach that integrates Conditional Tabular Generative Adversarial Networks (CTGAN) and Random Oversampling (ROS) for evaluating CFST column fire resistance performance. CTGAN technology is used to generate synthetic datasets to alleviate data scarcity issues, while ROS technique is used to balance the dataset. To validate the effectiveness of this method, detailed experimental tests using eight ML algorithms were conducted on both real and synthetic datasets, with results indicating superior performance on the synthetic dataset compared to the real dataset. The optimal model in the synthetic dataset achieves an Accuracy, Precision, Recall, and F1-Score of 0.972, 0.973, 0.972, and 0.972, respectively. Furthermore, extensive comparative analyses and practical applications of the proposed method were conducted, revealing its exceptional performance in evaluating the fire resistance of CFST columns. To facilitate ease of use and operation in practical engineering, a visual interface was developed. The introduction of this technique offers a novel solution for researchers and engineers in the field of structural fire engineering.
Read full abstract