Alkali-aggregate reaction (AAR) is a prominent degradation mechanism of concrete structures, which results from the dissolution of reactive silicate aggregates and the associated formation of damaging, expansive AAR gels. Due to the complex nature of AAR reactions, it has remained ambiguous which factors contribute the most to its occurrence on a global scale. Similarly, concrete monitoring in Finland has only recently begun to adapt to the reality that AAR often occurs concomitantly with other degradation mechanisms such as freeze-thaw damage, highlighting the need for critical evaluation of current methods to identify AAR occurrence and distinguish it from these other mechanisms. Machine learning (ML) provides a data-driven framework for both evaluating the relative “importance” of various data features, as well as predicting their consequent influence on AAR damage to concrete. Building on the success of previous such data-driven learning for the prediction of well-defined concrete properties such as compressive strength and setting time, the current work evaluates the feasibility of extending ML methods to AAR-relevant predictions. Results provide new insights into several of the most relevant concrete characteristics linked with AAR occurrence, and establish a basis for future work to extend and enhance such predictions to supplement monitoring and risk management of concrete structures.
|Journal||Revista portuguesa de engenharia de estruturas|
|Publication status||Published - Mar 2021|
|MoE publication type||A1 Journal article-refereed|