Automated Test Generation for Satellite On-Board Image Processing

Kurzfassung

On-board image processing technologies in the satellite domain are subject to extremely strict requirements with respect to reliability and accuracy in hard realtime. Due to their large input domain, it is infeasible to exhaustively execute all possible test cases. Furthermore, because of their complex computations, it is difficult to find specific test cases that provoke mission-critical behavior. To overcome these problems, we first define a test approach that efficiently and systematically captures the input domain of satellite on-board image processing applications. We present a dedicated partitioning into equivalence classes for each input parameter. As a result, our approach systematically reduces the number of test cases. Moreover, we define novel multidimensional coverage criteria to assess a given test suite for its coverage on the input domain. We present a test generation algorithm that automatically inserts missing test cases into the given test suite based on our multidimensional coverage criteria. This results in a reasonably small test suite that covers the whole input domain of satellite on-board image processing applications. Second, we define a test approach that automatically searches for test cases that are specifically tailored to provoke mission-critical behavior of satellite on-board image processing applications. For that, we present a novel genetic algorithm. We define a two-criteria fitness function that is based on the execution time and mathematical accuracy of the application under test. Therefore, our algorithm automatically selects test cases that provoke worse execution times and inaccurate results of the satellite on-board image processing application. We investigate the efficiency of our approaches on the PLAnetary Transits and Oscillation of stars (PLATO) Fine Guidance System (FGS) algorithm. This is a satellite on board algorithm that calculates the high-precision attitude of the spacecraft. The experimental results show that our first approach efficiently and systematically generates a test suite. This suite completely covers the input domain with respect to our multidimensional coverage criteria. This test suite has a higher error-detection capability than a randomly generated test suite. Therefore, our test approach increases the test efficiency and quality. Furthermore, our genetic algorithm automatically finds test cases that provoke longer execution times and less accurate results when using the generated test suite from the first approach as search space than using a randomly generated test suite. Hence, our genetic approach improves a given test suite to support robustness testing. As a summary, the combination of our approaches increases the efficiency and effectiveness of the test process for satellite image processing applications.