Customized Deep Convolutional Neural Network Modelling Toward Porous Concrete in the Green Infrastructure for Water Management

The growing demand for sustainable urban infrastructure emphasizes the need for advanced materials and predictive modeling. This paper proposes a new method using a customized deep Convolutional Neural Network (CNN) to predict the performance of porous concrete mixtures, specifically focusing on mixtures incorporating steel slag as a partial replacement for traditional stone aggregates. The porous concrete plays a vital role in green infrastructure by mitigating flooding, reducing noise, and lowering heat absorption in urban areas. However, predicting the performance of different mixtures remains challenging. This proposed work introduces a Hyperbolic Tangent Activated Customized CNN (HTAC-CNN) to improve predictive accuracy for key performance indicators, including Marshall Stability, Marshall Flow, Air Voids, Voids in Mineral Aggregate (VMA), Voids Filled with Bitumen (VFB), Unit Weight, Permeability, and Abrasion Loss. The HTAC-CNN leverages dual convolutional paths with distinct normalization and activation functions, capturing complex, nonlinear relationships in the data. Experimental results show that steel slag enhances the mechanical properties of porous concrete, with an optimal performance at 20% and 40% replacement levels. The model outperforms conventional methods, achieving an MSE of 0.0228, MAE of 0.1487, MSLE of [Formula: see text]05, and RMSE of 0.1510.