Modeling of Nonlinear Mechanical Behavior and Mix Design Optimization of Sustainable Recycled Aggregate Concrete Using Artificial Neural Network

Abstract

The increased demand for eco-friendly building materials has heightened the necessity for dependable forecasting tools that can model the intricate behavior of recycled aggregate concrete . This research formulates and assesses a feedforward Artificial Neural Network model intended to understand the nonlinear connections among essential mix design factors and the mechanical characteristics of RAC. The model integrates six input variables, which include water-to-cement ratio, the proportion of recycled aggregates, age of curing, alongside three categorical indicators related to the mix, predicting three properties: compressive strength, tensile split strength, and flexural strength. The architecture of the ANN consists of two hidden layers, each comprising 32 neurons, utilizing the ReLU activation function with the Adam optimization method. Findings indicate exceptional predictive effectiveness for compressive strength with R2 = 0.9781 and RMSE = 1.2257 MPa, while the tensile and flexural strengths exhibit moderate levels of predictive accuracy, highlighting the impact of additional microstructural elements not explicitly assimilated in the input parameters. A comparative analysis reveals that the ANN model considerably exceeds the performance of conventional empirical equations, which displayed extremely high error values and lacked explanatory capability. Residual diagnostics validate the lack of systematic bias and reveal a stable error distribution centered around zero. Additionally, optimization driven by the ANN identifies ideal combinations of water-to-cement ratio, recycled aggregate content, and curing age that improve both mechanical strength and durability. The results affirm the efficacy of machine learning techniques in modeling diverse concrete systems and underscore the promise of ANN as a trustworthy predictive and prescriptive resource for sustainable RAC mix design, minimizing the reliance on comprehensive laboratory testing.

Keywords

References