Modeling Of A Machine Learning-Based Virtual Copilot For Helicopters

Abstract

Key challenges in the aviation industry are mainly related to increasing safety and efficiency. With the rise of Artificial Intelligence (AI), new technologies enable on-board systems to adapt to new tasks and to perform them without programming for different scenarios. This research focuses on the design of a virtual copilot for helicopters, to assist the pilot and to reduce workload. Three critical use cases are investigated: predicting Vortex Ring State (VRS), detecting engine malfunctions, and aiding pilots during autorotation scenarios. Each of these presents unique challenges, so different machine learning techniques are analyzed to choose the best fit for each use cases: a Supervised Learning algorithm is developed to predict VRS, providing timely warnings for proactive mitigation; engine malfunctions are addressed using an Unsupervised Learning assis-tant to detect subtle performance deviations; Reinforcement Learning and Imitation Learning algorithms are explored to design a virtual assistant for supporting during autorotative descent.