Automatic sample tracking in ultrasonic testing system utilizing deep learning

Abstract

Ultrasonic testing (UT) system is a family of non-destructive testing techniques has been developed to conduct examinations and make measurements of material, part or component without causing destruction. Recently, in almost UT system, in order to determine the coordinate dimension of a sample, it is manually operated. However, it took a long time and difficult to finish doing an operation. Furthermore, the system operator had to be trained, learning skill or getting qualification before use the machine or equipment. As the result of this inconvenience, a based-lab system for automatically detect sample is in needed. In this thesis, by combine image processing and YOLOv3 deep learning with the UT system, we introduce the vision solution can help identify, determine and track coordinate, size or shape of the sample. In addition, the combination of vision and motion control are used to optimize the scanning process, procedure and operation.