Fault Injection for Robustness Testing of Satellite On-Board Image Processing

Abstract

In this Bachelors thesis, we developed an automated fault injection approach for satellite on-board image processing and applied it to the case study of the Fine Guidance System (FGS) of the PLATO mission. Due to the enormous input range of the FGS, we used an equivalence class test to determine the robustness of the FGS. Based on a literature review, we selected the fault classes and made a domain-specific selection in consultation with experts from the DLR. We considered a wide range of faults, from electronics noise to optical de-focus blur to cosmic rays. Then, we modeled these fault classes and integrated them into our custom-developed injection framework. In addition, we thoroughly evaluated the equivalence classes and test suites with various coverage. Thereby we were able to identify which faults in the input data the FGS are robust against in various settings. Furthermore, we were able to identify cases where the FGS should be improved. The general tendency is that the FGS is robust against faults with changes in the width or shape of the stars, including de-focus blur and brighter-fatter noise. In contrast, the FGS is not robust against punctual faults, such as hot and dead pixels.