Prediction of compressive strength of FRP-confined concrete using machine learning: A novel synthetic data driven framework

Abstract

FRP-confined concrete is an effective approach to enhance mechanical properties of concrete columns. Hence, accurate prediction of the compressive strength of FRP-confined concrete has the significant importance. Machine learning is a technology for achieving high-precision prediction of material performance. However, constrained by the lack of abundant experimental data in engineering, the reliability and robustness of models are often questionable. To address the issue, a novel synthetic data driven framework was developed in this study. The latest and most extensive experimental dataset was collected firstly. Subsequently, the state-of-the-art conditional tabular generative adversarial network (CTGAN) algorithm was employed to construct the model for generating FRP-confined concrete data for the first time. Finally, by adopting the “Train on Real, Test on Synthesized” and " Train on Real, Test on Real” approaches, the models for predicting FRP-confined concrete strength were developed. The results indicate that the 3508 data generated by the developed CTGAN model exhibit high quality. The models exhibit the high prediction accuracy, strong generalization capability and robustness, achieving a coefficient of determination of 0.972, which significantly surpasses the prediction accuracy of existing models. Therefore, the model for predicting the compressive strength of FRP-confined concrete developed in this study can be utilized to guide the development and structural design of FRP-confined concrete, and it can also serve as an alternative to empirical models. Furthermore, the novel synthetic data driven framework provides a powerful solution for addressing the issue of the poor reliability and robustness of models resulting from the lack of abundant data in engineering problems.