Synergistic effect of fly ash and stone dust on foam concrete under saline environment: Mechanical, non-destructive and machine learning approaches

A sophisticated machine learning (ML) strategy and a salty setting where fly ash and stone dust replace cement and sand may reveal eco-friendly lightweight foam concrete's implications that are not widely available or adequately stated in the literature. The study aims to conduct experimental programs to analyze the implications of utilizing various quantities of stone dust and fly ash as a substitute for sand and cement on foam concrete's fresh, hardened, and non-destructive testing (NDT) properties. The study explores foam concrete's compressive, splitting, permeability, pulse velocity, and microstructural properties and its potential as an alternative to conventional concrete in saline environments while integrating machine learning techniques like SVM and ANN for predicting compressive strength. The experimental study utilized three foam concrete batches with varying filler and binder proportions: Batch I with 100% sand and different fly ash percentages, Batch II with stone dust replacing sand and cement replaced by fly ash, and Batch III with 100% stone dust and different fly ash and cement ratios, all consistently using a foaming agent at 0.5% by binder weight. The study observed that replacing approximately 30% of cement with fly ash yielded the highest compressive strength, while substituting over 40% of sand with stone dust also showed promising results, achieving peak compressive strengths of 13-14 MPa. The study findings further revealed that the water absorption and permeability rate were minimal, and the rebound hammer, microstructure, and pulse velocity were ultimate when cement was replaced with roughly 30% fly ash. The experiments' outcomes are utilized to develop advanced machine learning methods for forecasting its strength. The ML technique demonstrates that an inferior regression coefficient (R²) support vector machine (SVM) contrasts dramatically with a larger R² value for the artificial neural networks (ANN), showing an excellent projection with experimental data.