Framework for Risk Assessment of Mixture Designs Incorporating Alkali-Silica Reactive Aggregates

Abstract

Alkali aggregate reaction (AAR) affected structures show reduced serviceability and premature distress in over 50 countries worldwide. Once triggered, remediation is challenging and uncertain; therefore, prevention remains the most effective strategy against AAR. Several laboratory test protocols have been developed over the years to evaluate the potential reactivity of aggregates under varying conditions. Among them, the Accelerated Mortar Bar Test (AMBT) and Concrete Prism Test (CPT) are the most widely used. However, exposure site data has revealed notable discrepancies between laboratory outcomes and field performance. Therefore, this PhD project proposes a probabilistic framework integrating laboratory and field data with logistic regression to predict ASR risk. First, a systematic review and comparative analysis of laboratory methodologies and field behaviour is conducted, supported by a comprehensive bibliometric analysis. In addition, a robust database integrating laboratory and field data has been established to support the research’s future steps. Next, the study focuses on the reliability of laboratory tests, indicating moderate accuracy in predicting field performance for the AMBT and the CPT. Then, by incorporating a multifactorial analysis integrating laboratory and field data with statistical and probabilistic modelling to account for variability in the test outcomes, this research assesses the risk of AAR occurrence in the field. More specifically traditional statistical methods, Bayesian analysis, and Beta distribution model are employed resulting in the probability distribution of AAR occurrence given laboratory test outcomes, environment and alkali loading. The findings highlight an opportunity for recalibration of these methods through advanced analytical models that account for environmental conditions, alkali content, and the presence of SCMs to improve predictive accuracy. Therefore, machine learning (ML) techniques, including decision tree classifiers and logistic regression are employed to appraise and adjust the thresholds of AMBT and CPT. Finally, the study proposes that thresholds should be observed dynamically, through a flexible coupled threshold-time (FCTT) approach, taking into account expansion levels, test duration, environmental factors, and alkali content to more accurately predict AAR occurrence. This dynamic risk assessment framework enables more informed decisions regarding mitigation strategies based on a structure's specific information, improving alignment between laboratory outcomes and real-world durability.