Using multilayer neural networks in determination of pressure on the surface of supersonic flying vehicles

The paper considers a system of artificial neural networks based on the generator-discriminator model for classifying the elements of surfaces in supersonic flying vehicle bodies and restoring the coefficients of these surfaces canonical equations. The first network of the system appears to be a classifier, which after making a decision on the class of surfaces presents the precedent to one of two networks specializing in restoring regression of the canonical equation coefficients. As a precedent, a minimal shield of four adjacent elements of the surface grid is provided built at nine points. Neural network learning is based on back propagation of the error algorithm. Extended precision formula is used to determine pressure on target and reconstructed surfaces. Besides, pressure on the target surfaces was determined with the Newton method. Results were obtained for two classes of surfaces, i.e. ellipsoid and hyperboloid surfaces. Comparison of the results for both classes of surfaces is provided. Pressure on target and reconstructed surfaces does is not differing, despite the noticeable in some cases discrepancy in the coefficients of equations for the target and the restored surfaces. Constructed neural network could be used to quickly find the surface parameters and precisely calculate the pressure of bodies characterized by complex geometry.