Artificial Neural Network-Based Predictive Modeling of Mechanical Properties in Nano-Modified Concrete Incorporating Nano-Silica and Nano-Alumina

Abstract

This study presents the development of Artificial Neural Network (ANN) models for predicting the mechanical properties of M30-grade nano-concrete incorporating nano-silica and nano-alumina. The models were built on a comprehensive experimental dataset comprising 17 mix designs and over 250 test specimens, capturing variations in strength due to nano-material dosages ranging from 0% to 4%. Using MATLAB’s Neural Network Toolbox, five separate ANN models were developed to predict 7-day, 14-day, and 28-day compressive strengths, as well as tensile and flexural strengths. Each model utilized seven input variables—cement, fine and coarse aggregates, water, superplasticizer, nano-silica, and nano-alumina—and a single mechanical output. The networks followed a feed-forward backpropagation architecture and were trained using the Levenberg–Marquardt algorithm. All models demonstrated high accuracy, with correlation coefficients exceeding 0.93 and low mean squared error (MSE) values. The tensile and flexural models showed R values of 0.9678 and 0.9883, respectively, while compressive strength models achieved R values of 0.979 (7-day), 0.969 (14-day), and 0.930 (28-day). Prediction accuracy exceeded 93% in all cases. The models also performed well on 48 new datasets featuring intermediate nano-material combinations, confirming their generalization capability. Optimal strength performance was consistently observed at 2% nano-silica and 1% nano-alumina. This research highlights the effectiveness of ANN in modeling complex concrete behavior and offers a data-driven approach for optimizing nano-concrete mixes. The framework reduces reliance on trial-based experimentation and supports the development of high-performance, sustainable concrete materials.