A case study has been performed to get insight in different defect interpretation strategies. It was felt that, before comparing experimental results of unknown defects with simulated results of known defects, it is important to "clean" ultrasonic images from non-relevant indications. In the final evaluation, knowledge based systems or neural network classifyer can be used to classify the image or features derived from the image. A different way would be to use geometrical features for different planes which can be created by a 3D-reconstruction program.
