Advanced concrete pavement internal crack monitoring using wave response variation and deep learning

This paper presents an in-depth investigation into internal crack monitoring in concrete pavement through the application of wave response variation (WRV) and deep learning techniques. The study examines wave scattering in both homogeneous (HM) and inhomogeneous (IHM) media, validating WRV by analyzing forward scattering due to interval vertical cracks through laboratory tests and analytical solutions. It compares various laboratory tests with experimental and finite element (FE) results. The analysis reveals that shallower cracks result in peaks at lower impulse frequencies on the WRV curve, while deeper cracks correspond to peaks at higher frequencies, as observed from both forward and incident waves. The study also finds that the complexity of IHM, influenced by random aggregate size and distribution, significantly affects WRV patterns, with larger aggregates causing greater energy attenuation in forward waves compared to smaller aggregates. Machine learning (ML) techniques are employed to predict cracks and clarify the impact of aggregate information on WRV. This research establishes a framework for understanding internal damage in IHM using ML technology. This study enables effective monitoring of internal vertical cracks, such as reflective cracks in bridge decks, pavements, and airport runways. •In IHM, larger aggregates more significantly affected WRV, increasing energy attenuation in the forward wave than smaller ones.•The combination of random aggregate size and distribution results in a more intricate WRV pattern.•GANs can generate synthetic samples based on images, creating a robust database for training purposes.•ALE clarifies the ML model's black box, showing how aggregate information affects the WRV.•The CNN achieved impressive training accuracy when using WRV images with varying internal crack information.