Speaker
Description
This project introduces a Machine Learning approach for anomaly and novelty detection in the HERA mission. The main objectives include evaluating diverse techniques, integrating them into the processor, and assessing their performance in terms of memory requirements. Additionally, we aim to run the model in real-time on the radiation-tolerant LEON 3 processor utilized in the mission's On-Board Computer.
During our investigation, we explored both supervised and unsupervised machine learning techniques for identifying anomalies and novelties. We carried out an in-depth analysis by utilizing data from the XMM and MEX subsystems, supplemented with artificially generated data. We successfully integrated the trained models into a C++ environment, specifically targeting big-endian processors like LEON 3. Moreover, we conducted benchmarking to evaluate the models' performance and their resource requirements, including imported libraries.
This research highlights the promising capabilities of machine learning, as it enables swift responses to unexpected events and plays a crucial role in ensuring the success of space operations.
