Application of machine learning models for the optimisation of compressive strength and water resistance of geopolymer stabilised compacted earth
Thanh-Phong Ngo, Ho-Nam Vu, Quoc-Bao Bui
Abstract
Earthen materials are attracting the interest of the public due to its sustainable properties. However, in several regions, especially in tropical zones, the water resistance is usually a barrier for the wide application of earthen materials. This paper presents an investigation on geopolymer stabilised compacted earth (GSCE). Influences of different parameters on the unconfined compressive strength (UCS) and the water resistance of GSCE have been analysed by machine learning models. First, the experiments were presented. The variations in Na 2 SiO 3 /NaOH ratios, curing temperatures, curing durations, specimens immersed in different water types, the number of wetting-drying cycles, and the corresponding UCS obtained were presented. Then, the machine learning models were applied, including different techniques: extremely gradient boosting (XGBoost), random forest, adaptive boosting, gradient boosting regression tree. The results showed that XGBoost demonstrated the highest performance, with the highest accuracy in UCS prediction. Partial Dependence Plots (PDPs) were also employed to visualize the impact of each input on UCS. The most impacting parameter observed was the curing temperature. The methodologies applied in this study (geopolymer stabilisation and optimization by machine learning) can also be applied for other types of earth, or for other earth construction techniques.