Modeling of Shear Strength of Ultra-High-Performance Concrete Beams Using Statistical Learning Tools

Introduction. Ultra-high-performance concrete (UHPC) combines high strength and crack resistance with low permeability, making it ideal for structures operating under aggressive environmental conditions and high loads. The growing use of UHPC in construction makes necessary the development of scientifically grounded methods for designing structures made with this material. The aim of this study is to develop engineering methods for calculating the shear strength of UHPC I-beams using statistical learning techniques. The models were based on extensive datasets, including both the authors' own experimental results and data from other researchers. Materials and Methods. Artificial neural networks (ANNs) and regression analysis methods were used to develop the models. The tasks were implemented using the STATISTICA software package. Results. Nonlinear expressions were developed for engineering calculations, allowing for the determination of the shear strength of UHPC-beams accounting for the shear span and structural parameters, including section geometry, UHPC strength characteristics, and fiber and shear reinforcement ratios. The proposed formulas reduce the discrepancy between theoretical and experimental data by up to 2.4 times compared to calculation by methods adopted in codes. The formulas are applicable to beams unreinforced in shear as well as those with fiber and shear reinforcement. Discussion and Conclusion. The results confirm the applicability of regression models and ANNs for calculating the shear strength of UHPC beams, particularly in cases where traditional analytical solutions are difficult to formalize. The reliability of the developed models is supported by statistical analysis, including verification of regression equation adequacy and comparison with existing code-based methods.