Predicting The Compressive Strength of Geopolymer Concrete Using Machine Learning Techniques

Abstract

Concrete activated by sodium hydroxide and sodium silicate is the subject of this thesis. Geopolymers have recently become a very heated topic of research due to the environmental issues with ordinary Portland cement (OPC)-based concrete. Geopolymer not only contributes to reducing carbon dioxide emissions but also declines industrial waste deposition in the waste disposal area by using it as raw materials. The solution of raw material silicate and aluminium components and an aluminium tetrahedral reaction cause amorphous to the semi-crystalline geopolymer matrix to form. Many efforts were made to produce various types of geopolymers by changing the solid or liquid portions and curing conditions and time of geopolymers. A new type of geopolymer is currently being developed, the so-called FAGPC, fly-ash geopolymers. In this thesis, fly ash as industrial byproducts forms geopolymer concrete by using mix design.

In the present thesis, the fly ash geopolymer concrete mix design data is collected from different research papers to predict the compressive strength. Random forest regression (RFR), decision tree (DT), and support vector regression (SVR) compressive strength modelling are carried out. The thesis also includes geopolymer concrete simulation and modelling of RFR, DT, and SVR. Finally, the compressive strength of FAGPC samples collected from research papers based on various parameters is developed in the decision tree and supported by support vector regression and random forest regression. FAGPC is not only an environmentally friendly, silicate-reduced product but also has potential for infrastructural and construction applications because of its comparable physical and mechanical properties. Finally, the decision tree and random forest modelling show a correlation between the predicted results and the objectives introduced. The very reliable performance factors of the model that demonstrate the compressive strength prediction capacity of the used algorithm confirm this.