Evaluation of Heat-Induced Damage in Concrete Using Machine Learning of Ultrasonic Pulse Waves.

This study investigated the applicability of using ultrasonic wave signals in detecting early fire damage in concrete. This study analyzed the reliability of using the linear (wave velocity) and nonlinear (coherence) parameters from ultrasonic pulse measurements and the applicability of machine learning in assessing the thermal damage of concrete cylinders. While machine learning has been used in some damage detections for concrete, its feasibility has not been fully investigated in classifying thermal damage. Data was collected from laboratory experiments using concrete specimens with three different water-to-binder ratios (0.54, 0.46, and 0.35). The specimens were subjected to different target temperatures (100 °C, 200 °C, 300 °C, 400 °C, and 600 °C) and another set of cylinders was subjected to room temperature (20 °C) to represent the normal temperature condition. It was observed that P-wave velocities increased by 0.1% to 10.44% when the concretes were heated to 100 °C, and then decreased continuously until 600 °C by 48.46% to 65.80%. Conversely, coherence showed a significant decrease after exposure to 100 °C but had fluctuating values in the range of 0.110 to 0.223 thereafter. In terms of classifying the thermal damage of concrete, machine learning yielded an accuracy of 76.0% while the use of P-wave velocity and coherence yielded accuracies of 30.26% and 32.31%, respectively.