Data-driven prediction of UHPC compressive strength using a hybrid 1D CNN–GRU network

Ultra-high-performance concrete (UHPC) is one of the cutting-edge materials in the concrete industry. UHPC possesses superior mechanical properties compared to conventional concrete, bringing many breakthroughs in construction. Compressive strength is one of the most essential properties of UHPC and is determined through destructive laboratory tests. These tests are often costly and time-consuming. In this study, a data-driven approach is proposed to predict the compressive strength of UHPC using a hybrid deep learning model combining a Convolutional Neural Network (CNN) with a Gated Recurrent Unit (GRU). The 1D CNN component effectively extracts local feature patterns among material properties, while the GRU module captures sequential and interdependent relationships. A comprehensive dataset, including various mix designs, was used for model training and validation. The performance of the hybrid CNN–GRU network was compared with the standalone CNN and LSTM models. The results demonstrate that the proposed hybrid model achieves superior accuracy, exhibiting lower mean absolute error (MAE) and mean square error (MSE) on the test dataset. This study highlights the potential of data-driven hybrid neural networks in improving the prediction of UHPC compressive strength, providing practical insights for optimizing mix design in UHPC applications.