A convolution-based deep learning approach for estimating compressive strength of fiber reinforced concrete at elevated temperatures

Fiber-reinforced concrete (FRC) exhibits high fire-resistance capacity and maintains structural integrity at elevated temperatures. However, conventional approaches for optimizing its mixture design and predicting its corresponding mechanical responses following fire exposure present particular difficulties in efficiency, accuracy, and safety issues. To address these issues, a convolution-based deep learning model was developed in the present paper. A dataset with 19 features, including concrete mix proportioning, heating profile, and fiber properties, was collected from previous experimental recordings to evaluate the model performance. The feasibility and generality of the proposed model were validated through the collected dataset and another widely used concrete dataset, where our model performs the best compared with multiple machine learning baseline models. In addition, the correlation between temperature and the relative compressive strength obtained by the proposed model echoes with Eurocode 2, which further demonstrates that our proposed model can accurately estimate the mechanical performances of FRC exposed to high temperatures. It is envisioned that the proposed deep-learning approach serves as an accurate and flexible property assessment tool that aids researchers and engineers in mixture design optimization and compressive strength estimation of FRC for different engineering needs.

Automated summary

The paper introduces a convolution-based deep learning model to accurately assess the mechanical performance of Fiber-reinforced concrete (FRC) exposed to high temperatures. The feasibility and accuracy of the model were validated using experimental data, where it outperformed multiple machine learning baseline models. The proposed model can assist researchers and engineers in optimizing FRC design and estimating compressive strength for various engineering needs.