Reconstruction of turbulence statistics in a dump combustor using neural networks

Artificial Neural networks are utilized to predict flow properties of a confined, isothermal, and swirling flowfield in an axisymmetric sudden expansion combustor using a two-component laser Doppler velocimetry capable of measuring the mean velocity components and their statistics. Generalized feedforward, radial basis function, and coactive neuro-fuzzy inference system neural networks are tested and the results are compared in the reconstruction of the axial, tangential velocity profiles, and their root mean squares of their fluctuating velocity components. The results showed that generalized feed forward networks give the best prediction with the highest correlation coefficients for most of the flow profiles.