Abstract

Performance models used in the aircraft development process are dependent on the assumptions and approximations associated with the engineering equations used to produce them. The design and implementation of these highly complex engineering models are typically associated with a longer development process. This study proposes a non-deterministic approach where machine learning techniques using Artificial Neural Networks are used to predict specific aircraft parameters using available data. The approach yields results that are independent of the equations used in conventional aircraft performance modeling methods and rely on stochastic data and its distribution to extract useful patterns. To test the viability of the approach, a case study is performed comparing a conventional performance model describing the takeoff ground roll distance with the values generated from a neural network using readily-available flight data. The neural network receives as input, and is trained using, aircraft performance parameters including atmospheric conditions (air temperature, air pressure, air density), performance characteristics (flap configuration, thrust setting, MTOW, etc.) and runway conditions (wet, dry, slope angle, etc.). The proposed predictive modeling approach can be tailored for use with a wider range of flight mission profiles such as climb, cruise, descent and landing.