Application of machine learning in predicting workability for alkali-activated materials

Alkali-activated materials (AAMs) have been extensively studied for their superior performance and eco-friendliness. While previous researches have primarily focused on the hardened properties of AAMs, the assessment of their fresh properties has often been overlooked. The preparation process of AAMs involves key factors in mix design that significantly impact their workability. This study comprehensively evaluates the workability of different types of AAMs, analyzing a total of 402 mixtures extracted from 26 individual papers. The examination focuses on key factors in AAM mix design, specifically the precursors, alkali activator, and aggregate phases. Finally, a mathematical model for predicting the workability of AAMs was constructed based on the LightGBM (LGBM) algorithm. In this model, the reactivity of precursors, alkali activator, geopolymer paste volume, superplasticizer content, and aggregate were set as the inputs, and the flowability was set as the output. Additionally, the predictive efficiency of LGBM model was evaluated and compared to the multi-linear regression model. Meantime, a validation experiment for proving its accuracy was also conducted. This study largely advanced the understanding of the workability of AAMs by providing practical guidelines on AAM mix design with high workability and consistency.

Automated summary

Alkali-activated materials (AAMs) have been extensively studied for their superior performance and eco-friendliness. However, the assessment of their fresh properties, which play a crucial role in their workability, has been overlooked in previous research. This study comprehensively evaluates the workability of different types of AAMs and constructs a mathematical model for predicting their flowability. The study provides practical guidelines for AAM mix design with high workability and consistency.