Comparison of Selected Machine Learning Tools for Identification of Jet Engine Dynamics

This paper addresses an application of machine learning to identify dynamical model of a gas turbine engine. Fundamental aspects of component modeling and calculation of both the steady state and the transient operation are briefly described. Subsequently, models of spool speed dynamics were identified via the neural network, the sparse identification of nonlinear dynamics, and the Koopman operator theory. Concerning the second mentioned approach, the utilization of nonlinear parameters of basis functions for sparsity promotion is proposed. The results show similar performance of considered approaches, while their complexity and interpretability differ severely. Furthermore, control-oriented advantages of sparse identification of dynamics and Koopman model are discussed.