Confinement behavior and prediction models of ultra-high strength concrete using metaheuristic tuned neural network

Ultra-High Strength Concrete (UHSC) is known for its brittleness compared to traditional concrete, which can lead to sudden collapses. When it comes to columns, failures are particularly serious and require the use of confinement models to accurately predict the strength and strain of confined UHSC columns. While previous confinement models exist, many equations either underestimate or overestimate the confinement of concrete due to idealized assumptions and the exclusion of significant variables. This study employs a hybrid machine learning approach to capture the complex interactions in confinement behavior and accommodate a broader range of peak strength and axial strain parameters in UHSC. Statistical performance measures indicate the superiority of the proposed models over existing equations. Through causal inference, the study assesses the effects and relative importance of each parameter on peak strength and axial strain. The visualizations provided by the performance plots helped identify patterns and correlations that would have been difficult to discern through numerical analysis alone. The developed NN-PSO models are proven effective in reasonably predicting the peak strength and axial strain of UHSC columns.