Optimal Curing Temperature for Determination of Compressive Strength of Geopolymer Concrete via Artificial Neural Network (ANN)

Geopolymer concrete offers a promising alternative to traditional Portland cement concrete, exhibiting comparable mechanical and durability performance while reducing environmental impacts. However, achieving desirable properties in geopolymer concretes through heat curing remains challenging. This study proposes the use of an unsupervised Artificial Neural Network (ANN): Self-Organizing Map (SOM) to determine the optimal curing temperature of geopolymer concrete based on experimental datasets. The novelty of this study lies in utilizing SOM for clustering and pattern recognition to establish the relationship between curing temperatures and compressive strength, providing a novel data-driven methodology for enhancing material performance. Data on compressive strength at different curing temperatures were collected and used to train and validate SOM models. Fly ash based geopolymer concretes of size 100mm3 cubes were prepared in two sets of activators; sodium hydroxide (NaOH) and a combination of sodium silicate (Na2SiO¬3) with NaOH. These samples underwent curing under three conditions: ambient, 60° and 80° for 28 days. Clustering analysis generated by the SOM model provides valuable insights into the relationship between curing conditions, activator dosages, and compressive strength. Consequently, a cluster of mix proportion was developed, enabling the selection of specific curing conditions that result in targeted compressive strength. The results show that curing temperatures of 80°C offers optimal compressive strength ranging from 27MPa to 34MPa. This method introduces a novel "cluster mix proportion" for selecting curing parameters and demonstrates the potential of machine learning in advancing sustainable construction materials. The approach provides a distinct advantage by reducing reliance on trial-and-error methods, saving time and resources, and establishing a foundation for further exploration of data-driven techniques in cement and concrete research.