Model Development for Strength Properties of Laterized Concrete Using Artificial Neural Network Principles

This study develops predictive models for determination of strength parameters of laterized concrete made with ceramic aggregates, based on the principle of Artificial Neural Networks (ANN). The model development follows the results of the experimental phase (covering compressive and split-tensile strengths), where numerous materials were used in varying proportions: ceramics (fine and coarse fractions), river sand, and granite were substituted between 0 and 100%, laterite between 0 and 30%, and curing ages between 3 and 91 days. The cement proportion was maintained at 100%, and the water–cement ratio was 0.6. The model development was performed in MATLAB based on the Levenberg–Marquardt (LM) principles, where input data were separated in ratio 70%:15%:15% for learning, testing, and validation phases, respectively. After several trials, the selected model architecture, based on satisfactory performance in terms of means square error, contains eight-input layer, ten-hidden layer, and two-output layer neurons.