Application of Genetic Programming and Artificial Neural Network Approaches for Reconstruction of Turbulent Jet Flow Fields

Two Machine Learning (ML) methods are considered for reconstruction of turbulet signals corresponding to
the Large Eddy Simulation database obtained by application of the high-resolution CABARET method accelerated on GPU cards for flow solutions of NASA Small Hot Jet Acoustic Rig (SHJAR) jets. The first method is the Feedforward Neural Networks technique, which was successfully implemented for a turbulent flow over a plunging aerofoil in (Lui and Wolf, 2019). The second method is based on the application of Genetic Programming, which is well-known in optimisation research, but has not been applied for turbulent flow reconstruction before. The reconstruction of local flow velocity and pressure signals as well as timedependent principle coefficients of the Spectral Proper Orthogonal Decomposition of turbulent pressure fluctuations are considered. Stability and dependency of the ML algorithms on the smoothness property and the sampling rate of the underlying turbulent flow signals are discussed.