Automated ultrasonic-based diagnosis of concrete compressive damage amidst temperature variations utilizing deep learning

Abstract

Ultrasonic-based non-destructive testing technologies have been extensively applied for detection of internal damage in concrete. However, it is vulnerable to environmental temperature variations. An automated ultrasonic-based diagnosis approach integrating the continuous wavelet transform, and the transfer learning enhanced deep convolutional neural networks is proposed to evaluate compressive damage amidst temperature variations. The ultrasonic tests were conducted on pre-damaged concrete specimens, considering both temperature variations and damage levels as variables. The results indicate that the temperature fluctuations significantly influence the ultrasonic parameters of concrete compression damage. The proposed method effectively identifies the concrete damage state amidst temperature variations. Furthermore, it is recommended that the temperature range within the training set should uniformly cover the expected temperature range throughout the lifespan of concrete structures. This study offers novel perspectives for ultrasonic testing of concrete subjected to environmental variations.