Nondestructive Damage Detection of Concrete With Alkali-Silica Reactions Using Coda Wave and Anomaly Detection

Taeyong Shin,Hongbin Sun,Jinying Zhu,Ying Zhang
DOI: https://doi.org/10.1109/jsen.2022.3149721
IF: 4.3
2022-03-15
IEEE Sensors Journal
Abstract:An anomaly detection model for early damage detection for concrete structures undergoing alkali-silica reaction (ASR) is presented. It is difficult to detect ASR initiation and early damage without a reference expansion measurement. Coda waves, or the multiply scattered portion of ultrasonic waves, have been found to be indicative of small changes in complex material such as concrete. The relationship between concrete damage and relative velocity change and decorrelation of coda waves has been studied, but a generalized model which detects when damage occurs in a concrete structure is still lacking. The presented method uses features extracted from coda waves to detect early damage in concrete structures. The model uses unsupervised learning and only requires data from undamaged structures for training. During the training process, the reconstruction error of the training data is minimized. When the data collected from damaged concrete structures is used as an input of the model, it returns high reconstruction errors that indicate the occurrence of damage in the structures. The performance of the model is validated using experimental studies and has been shown to generalize across two different ASR specimens.
engineering, electrical & electronic,instruments & instrumentation,physics, applied
What problem does this paper attempt to address?