Effects of binder proportion and curing condition on the mechanical characteristics of volcanic ash- and slag-based geopolymer mortars; machine learning integrated experimental study

Contradicting natures of volcanic ash (VA)-and ground granulated blast furnace slag (GGBFS)-based geo-polymers with regard to their need for being exposed to a heating source for conducting alkali-reactions make meaningful differences in the strength and flowability of mortars. To address the impacts of types of the geo-ploymerizing agent, their proportions, and different curing regimes on their strength and shrinkage behavior, three different curing scenarios including curing at room temperature, RT curing; resting under room temper-ature for various curing periods, and then hydrothermally curing at an elevated temperature for one day, RO curing; and curing in the oven for one day and then resting at the room temperature, for various curing periods, OR curing, are considered. Based on the flowability, setting time, unconfined compressive strength (UCS), and drying shrinkage tests on eight different mixtures, the effects of binder proportion and curing conditions are evaluated. It was observed that with the substitution of 30% VA with GGBFS, specimens cured at RT condition outperformed OR specimens. This threshold was 45% while comparing RT and RO specimens and with higher slag content, RT specimens gain higher UCS. Also, for GGBFS proportion less than 45%, specimens cured at RO conditions showed higher long-term UCS compared to the OR specimens, and this behavior was reversed for higher GGBFS content. Scanning electron microscopy test results indicated that N-A-S-(H) type gel is the dominant geopolymeric gel in VA-based specimens, and as the amount of slag increases, the gels formed change to N-(C)-A-S-H and with further addition of GGBFS, C-A-S-H type gel is the dominant gel which results in strength development in GGBFS-based specimens. Based on the results of 80 UCS and 140 drying shrinkage tests, different machine learning (ML) algorithms and an evolutionary model were employed to present accurate and cost-effective predictive models as useful and capable methods for practitioners and researchers for the pre-liminary design stages of projects. A comparative study revealed the superior prediction efficiency of the gradient boosting (GB) model with the R2 score of 95% among 8 different ML algorithms and advanced ensemble models. In addition, evolutionary models with the coefficients of determination of 92% and 95% are, respectively, proposed for the drying shrinkage and the UCS. This study advances the field by assessing VA-GGBFS blend and curing effects on geopolymer mortars and developing ML-based models for compressive strength and drying shrinkage, facilitating efficient optimization in practical applications.

Automated summary

This study investigates the strength and shrinkage behavior of volcanic ash (VA) and ground granulated blast furnace slag (GGBFS) based geo-polymers under different curing conditions. The effects of binder proportion and curing conditions on the properties of mortar are evaluated. Machine learning algorithms and an evolutionary model are employed to develop accurate predictive models for compressive strength and drying shrinkage.