Fully convolutional network-based ultrasonic inversion for multi-layered bonded composites

Ultrasonic methods are widely used for the detection and characterisation of defects in multi-layered bonded composites. However, quantitative reconstruction of defects, such as disbonds, which can affect adhesive bond integrity and severely reduce the strength of assemblies, remains challenging. In this work, a supervised full convolutional network (FCN)-based ultrasonic method is used to quantitatively reconstruct defects hidden in multi-layered bonded composites. This proposed method consists of a training process and a predicting process. In the training process, the FCN builds a non-linear mapping from the ultrasound data to the corresponding longitudinal (L-wave) velocity model. In the predicting process, the network obtained from the training process is used to directly reconstruct the L-wave velocity models from the new measured ultrasonic data of adhesively bonded composites. The simulation results show that the FCN-based ultrasonic inversion method has the ability to achieve the accurate quantitative reconstruction of ultrasonic L-wave velocity models of the high contrast defects, which has potential in online detection of multi-layered bonded composites.