A Machine Learning Approach For Data-Driven Enhancement Of Rotorcraft Development Emulation Tools

Abstract

First principles models, based on prior physical knowledge of the system of interest, are widely adopted in the aerospace industry, as they provide valuable emulation tools for core development and certification activities, such as requirements validation and verification, performance analyses and flight simulations. Incomplete representation of underlying physics, as well as simplifications driven by computational constraints, could lead to errors in model's prediction. Oftentimes, real-world measurement data of the system of interest are available, particularly for projects in an advanced stage of development. This data availability possibly represents a valuable resource to fill the gap between model and real-world data, thanks to dedicated machine learning techniques. This paper presents an autoregressive grey-box approach based on Gaussian Process regression, that exploits both prior knowledge and available data for the enhancement of existing first principles aircraft actuation models.