Automating defect visibility assessment in radiographs and sophisticated film noise modelling

Speaker:
Eckel, Sebastian; Imperial College; United Kingdom

Authors:
Eckel, S.; Imperial College London; United Kingdom
Huthwaite, P.; Imperial College London; United Kingdom
Schumm, A.; EDF; France

ID: ECNDT-0460-2018
Download: PDF
Session: X-ray & CT modeling 2
Room: G3
Date: 2018-06-14
Time: 10:50 - 11:10

The qualification of radiographic inspections is crucial to ensure operational reliability. Traditionally, this could be achieved by repeated experiments, which can be expensive and time consuming. To make the qualification process more efficient, computational tools are necessary to model the test setup including defects and simulate the inspection results. In radiography, these results are gained in image form. As opposed to other NDE methods, the qualification is then based on a subjective image quality measure regarding how well the included defect is visible by the inspector. The aim of this project is to automate that interpretation step, and thus pave the way to applications like model-assisted Probability of Detection (POD): A modelled computer observer should give an objective image quality measure to reduce the human impact and to reach a holistic simulation including the image interpretation.
To incorporate that interpretation, a Model Observer approach was chosen based on the so called Channelized Hotelling Observer. It was validated and outperforms other visibility models used by industry.
In order to improve the realistic appearance of the simulated X-ray images, a second part of this work aims to improve the noise model of the film based radiographs. Film noise shows specific characteristics like graininess influencing the defect visibility. To make the simulation results closer to reality, a new noise model was developed based on experimental results. The model allows to generate very realistic looking noise, which is specifically adapted to the film class, resolution and optical density.