Abstract

Alkali-activated system is an environment-friendly, sustainable construction material utilized to replace ordinary Portland cement (OPC) that contributes to 9% of the global carbon footprint. Moreover, the alkali-activated system has exhibited superior strength at early ages and better corrosion resistance compared to OPC. The current state of analytical and machine learning models cannot produce highly reliable predictions of the compressive strength of alkali-activated systems made from different types of aluminosilicate-rich precursors owing to substantive variation in the chemical compositions and reactivity of these precursors. In this study, a random forest model with two constraints (i.e., topological network and thermodynamic constraints) is employed to predict the compressive strength of alkali-activated systems made from 26 aluminosilicate-rich precursors and distinct processing parameters. Results show that once the model is rigorously trained and optimized, the RF model can yield a priori, high-fidelity predictions of the compressive strength in relation to the physicochemical properties of aluminosilicate-rich precursors; processing parameters; and constraints. The topological network constraint provides the chemo structural properties and reactivity of the aluminosilicate-rich precursors. Whereas the thermodynamic constraint estimates the phase assemblages at different degrees of reaction of the aluminosilicate-rich precursors. Finally, the correlations between topological network constraint; phase assemblage; and compressive strength are demonstrated. When the topological network constraint equals 3.4, the alkali-activated systems can achieve their optimal compressive strength.

Department(s)

Electrical and Computer Engineering

Second Department

Materials Science and Engineering

Comments

National Science Foundation, Grant 1661609

Keywords and Phrases

Alkali-activated system; Compressive strength; Machine learning; Thermodynamic simulation; Topological constraint theory

International Standard Serial Number (ISSN)

0950-0618

Document Type

Article - Journal

Document Version

Final Version

File Type

text

Language(s)

English

Rights

© 2023 Elsevier, All rights reserved.

Publication Date

20 Jun 2022

Share

 
COinS