Temperature prediction of the main radiators of Alpha Magnetic Spectrometer-02 by the artificial neural network method

The Alpha Magnetic Spectrometer-02 (AMS-02) is a particle physics detector of high precision installed at the International Space Station to search for the origin of dark matter, existence of antimatter, and origin and properties of cosmic rays. Analyzing and predicting the thermal condition of electric devices are of great importance to ensure them work under permitted temperature ranges. In this study, the orbital parameters of the AMS-02 and thermal data of the main radiators were analyzed. On this basis, artificial neural network models for the RAM and WAKE radiators were built using historical thermal data. The input neuron is the beta angle and the output neurons comprise the temperature of 28 sensors for the RAM radiator and 31 sensors in the case of the WAKE radiator. Comparisons between the prediction results and recorded data validate the models, and the errors are approximately 0.2 °C. The analysis of the selection of period time indicates that the model using the averaged data of one orbital period time is more accurate than that using others, such as half, one-third, or one-sixth period times. The models provide powerful tools to predict the variations in the thermal data of the main radiator with the beta angle.