Research and development projects involving complex systems are often executed in a collaboration between engineers and scientists of various institutions. In such a collaboration, different parts of the system are developed by different companies located in different physical locations. Whenever this happens, testing the system and its components is difficult, since the adjacent hardware required to test the communication channels may not be available on site. This is further problematic in the space industry, because replicating a spacecraft hardware is expensive and arises other technical difficulties as every hardware is a prototype. A solution to this problem is to use a hardware emulator application in the development cycle of the OBSW which can reflect the true hardware behaviour and can be shared easily between the distributed development teams. However, precise low-level hardware emulators are very tedious to build and require deep domain knowledge. A noble alternative solution to this problem is to learn a black-box model from the temporal data-space of the candidate hardware using a ML algorithm which can reflect the realistic hardware behaviour. RNN based models with the LSTM and GRU memory units are excellent in learning sequences and are able to capture long-range dependencies in the temporal dataset. In this study, different LSTM and GRU models are applied to the task of modelling a physical hardware system. Input and output signals from the real hardware are used to train the RNN models following a supervised learning method. The obtained result show that both types of RNN models are capable of simulating the realistic hardware behaviour with a considerable accuracy. The performance of the well configured GRU model is seen to be slightly better than that of the equivalent LSTM model. Nevertheless, the signals predicted by both RNN models showed more than 90% resemblance to the actual signals obtained from the hardware in their frequency spectrum.


    Zugriff

    Download


    Exportieren, teilen und zitieren



    Titel :

    Modelling and Simulation of Spacecraft Hardware based on Machine Learning Techniques


    Beteiligte:

    Erscheinungsdatum :

    2019


    Medientyp :

    Hochschulschrift


    Format :

    Elektronische Ressource


    Sprache :

    Englisch




    Modeling and Simulation of a Spacecraft Payload Hardware Using Machine Learning Techniques

    Nepal, Ayush Mani / Prat, Arnau / Höflinger, Kilian Johann et al. | AIAA | 2020


    Spacecraft Digital Hardware

    Sebestyen, George / Fujikawa, Steve / Galassi, Nicholas et al. | Springer Verlag | 2018


    SPACECRAFT CHECKOUT HARDWARE

    GORCHAKOVSKIJ ALEKSANDR ANTONOVICH / EVSTRAT KO VLADISLAV VLADIMIROVICH / MISHUROV ANDREJ VALERIEVICH et al. | Europäisches Patentamt | 2015

    Freier Zugriff

    Survey of machine learning techniques in spacecraft control design

    Shirobokov, Maksim / Trofimov, Sergey / Ovchinnikov, Mikhail | Elsevier | 2021


    MACHINE LEARNING TECHNIQUES FOR AUTONOMOUS SPACECRAFT GUIDANCE DURING PROXIMITY OPERATIONS

    Federici, Lorenzo / Benedikter, Boris / Zavoli, Alessandro | TIBKAT | 2021