Soft Computing Techniques for the Prediction and Analysis of Compressive Strength of Alkali-Activated Alumino-Silicate Based Strain-Hardening Geopolymer Composites

Abstract

A futuristic class of concrete that has ductile nature with zeroed cement and eco-friendly materials is popularly known as Engineered Geopolymer Composites (EGC). Research on strain-hardening geopolymer composites have started a decade back and there is an extensive opportunity for utilizing this kind of eco-material in the construction sector for sustainable development. The focus of this article is to develop predictive models for the compressive strength of EGC with the objective of assisting the experimental researches and to analyze the type of locally available eco-materials that could be adopted in developing the novel ductile geopolymer composites for structural applications. A database has been created with ten mix-design factors, including material contents and curing conditions as inputs. Three soft-computing tools viz., Artificial Neural Networks (ANN), Response Surface Methodology (RSM) and Gene-Expression Programming (GEP) have been exercised to create, train and validate the predictive models. Also, a critical comparative analysis has been performed. The accuracy of predictive models is tested with regression tools. Among these artificial intelligence tools, the RSM model has shown an accuracy level of 96% with the least RMSE of 2.8 and the ANN [10:8:1] model has shown 93% accuracy with RMSE 3.4. Only 80% accuracy has been shown for the GEP model with RMSE 6.2. The sensitive parameter in concrete composites is compressive strength where the prediction error should be minimum. This article concludes that the developed ANN and RSM models worked effectively in prediction whereas GEP is comparatively less accurate, which can be improved when influencing mix-design parameters are lesser.