Study on a neural network model for high speed turbulent boundary layer inducing optical distortions

A neural network (NN) model for high speed turbulent boundary layer inducing aero optical distortions is proposed to solve the spatial frequency problem in sparse measurements. The temperature fluctuation, a measurable flow state, is set as the NN model input, and the optical path difference (OPD) is outputted. The NN structure, its optimal inputting point number and neuron number of hidden layer are analyzed. Numerical simulation results for two flow cases show that the OPDs of simulated positions using only limited points have good agreements with the calculated ones by ray tracing from whole field. Their correlation coefficient is up to 0.95 and the relative error of OPD root mean square is within 7% for both cases, which validates this model. Besides, its accuracy performances for different cases are analyzed, and the results accord with the experimental knowledge. This research provides numerical basis for the practical application of NN in sparse measurements.