SaimVAE: An unlabeled intelligent ultrasonic NDT method for composite materials

In recent years, a lot of research has been carried out to achieve intelligent ultrasonic NDT of composite materials. This paper proposes an intelligent ultrasonic NDT method to assist manual inspection, which does not require additional labelled data. The technique utilizes an improved variational auto-encoder (VAE) to construct an ultrasonic signal classification module. It automatically identifies normal and abnormal signals in composites by reconstructing them and calculating the errors, with an average Area Under Curve (AUC) value of 0.958. The method also designs a probe tracking module to replace the conventional encoder, with a tracking success rate of 92.1% and a tracking speed of 35 fps. Finally, the composite inspection results are visually presented using the ultrasonic signal classification and probe tracking results. The results show that this method makes it easier and faster to show the internal state of the composite material.