Identification of the Bombardier CRJ-700 Stall Dynamics Model Using Neural Networks

Abstract

This paper aims to present a new methodology to model the aerodynamic coefficients and predict the aircraft dynamics under stall conditions, including the hysteresis cycle, using neural networks. The aerodynamic coefficients variations required for the identification process were estimated from flight data collected during different stall maneuvers. Then, a level-D-qualified Bombardier CRJ-700 virtual research equipment simulator (VRESIM) developed by CAE, Inc. and Bombardier was used to gather flight data in both linear and nonlinear stall phases. According to the Federal Aviation Administration (FAA), level D is the highest qualification level for flight dynamics and propulsion models. Multilayer perceptron (MLP) and recurrent neural networks were trained for the aerodynamic coefficients learning and their correlation with flight parameters. A new methodology for tuning the neural network parameters, such as the optimal number of layers and neurons, was developed. The resulting models were validated by comparing predicted flight data with experimental data obtained from the level D Bombardier CRJ-700 VRESIM by considering the same pilot inputs. The models developed using the proposed methodology were able to predict the CRJ-700 flight dynamics in both static and dynamic stall conditions, with very good precision, within the tolerances of the FAA.