An Edge Computing Architecture for a Lunar Dust Recognition System

Lunar dust, also known as regolith, presents several challenges for rovers, including abrasive damage to mechanical parts, reduced traction, and impaired mobility. The accumulation of dust on solar panels and radiators diminishes power generation efficiency and disrupts thermal regulation, potentially leading to overheating of critical systems. Furthermore, contamination from lunar dust can compromise the integrity of scientific instruments and data, as electrostatically charged particles tend to adhere to surfaces, thereby impairing functionality. One precondition for mitigating the risks associated with lunar dust is its visual detection through computer vision algorithms. However, existing algorithms are computationally intensive and necessitate careful trade-offs regarding where the computation is conducted. This paper introduces an architecture for a lunar dust recognition system that addresses various computational trade-offs and initial validation results. The proposed approach leverages edge computing to establish an inter-satellite communication and computing system to cope with deep space mission requirements. To reduce bandwidth usage, machine learning techniques are employed to classify data into critical and non-critical categories, ensuring that only essential data is transmitted to Earth. However, scientists may benefit from raw data analysis, so opportunities for raw data downloading are also provided in the proposed design for analysis and model training, contingent on downlink capacity. The LunaLab at the University of Luxembourg, which replicates lunar surface conditions, has been utilized to validate the lunar dust recognition system.