Prediction of Volumetric Strain in Concrete Due to ASR Reactions Using Acoustic Emission Technique and Artificial Neural Network

Abstract

An alkali-silica reaction (ASR) is a swelling reaction that occurs over time in concrete between the alkaline in cement paste and the reactive silica found in many common aggregates used to make concrete. The swelling reaction causes the initiation and propagation of cracks in concrete structures, leading to their deterioration and eventual failure. Development of a reliable structural health monitoring method for detecting crack initiation and propagation, and evaluating concrete condition is important for assessing the integrity of structures. In this paper, we investigate the correlation between acoustic emission (AE) data and volumetric strain for concrete during ASR and develop an artificial neural network (ANN)-based model to predict volumetric strain using AE signals. The AE data and volumetric strain are collected from three concrete specimens. Two of the specimens contain reactive aggregates that accelerate the ASR reactions, while one specimen serves as a controlled specimen without ASR reactions. AE events of all three specimens are monitored using an array of piezoelectric sensors, and the volumetric strain of each specimen is measured monthly. The AE data collected from the sensors are then preprocessed to obtain damage sensitive features, such as count, amplitude, and signal strength. Using the preprocessed data, two back-propagation feedforward networks are trained to predict volumetric strain using either cumulative number of counts only or both cumulative number of counts and the number of counts per hour as inputs, respectively. The results show that the trained neural networks provide good prediction accuracy for the volumetric strains of concrete with ASR reactions after the volumetric strain exceeds 0.036%.