Neural-Network-Based Ultrasonic Inspection of Offshore Coated Concrete Specimens

Abstract

A thin layer of protective coating material is applied on the surface of offshore concrete structures to prevent its degradation, thereby extending the useful life of the structures. The main reasons for the reduction in the protective capability of coating layers are loss of adhesion to concrete and flattening of the coating layer wall. Usually, the state of the coating layer is monitored in the setting of water immersion using ultrasonic inspection methods, and the method of inspection still needs improvement in terms of speed and accuracy. In this study, the ultrasonic pulse echo method was used in a water immersion test of the coated specimens, and continuous wavelet transform (CWT) with complex Morlet wavelets was implemented to define the received waveforms' time of flight and instantaneous center frequency. These allow one to evaluate the thickness of the coating layer during water immersion. Furthermore, phases of reflected echoes at CWT local peaks were computed using a combination of Hilbert transforms (HT) and wave parameters derived from CWT. In addition, three relative wave parameters of echoes were also used to train deep neural networks (DNN), including instantaneous center frequency ratio, CWT magnitude ratio, and phase difference. With the use of three relative waveform parameters of the DNN, the debonded layer detection accuracy of our method was 100%.