Neural networks based subgrid scale modeling in large eddy simulations F. Sarghini a, * , G. de Felice a , S. Santini b a Department of Energetics, Applied Thermofluid Dynamics and Environmental Controls, University of Naples Federico II, P.le Tecchio 80, 80125 Naples, Italy b Department of Computer and Systems Engineering, University of Naples Federico II, P.le Tecchio 80, 80125 Naples, Italy Accepted 26 June 2001 Abstract In this paper a multilayer feed-forward neural network (NN) is used as subgrid scale (SGS) model in a large eddy simulation (LES). The NN was previously off-line trained using numerical data generated by a LES of a channel flow at Re s ¼ 180 with Bardina’s scale similar (BFR) SGS model. Results show the ability of NNs to identify and reproduce the highly nonlinear behavior of the turbulent flows, and therefore the possibility of using NN techniques in numerical simulations of turbulent flows. Ó 2002 Elsevier Science Ltd. All rights reserved. 1. Introduction Numerical simulation of turbulent flows is still one of the most challenging task in computa- tional fluid dynamics, for difficulties involved in modeling and for computational requirements. A promising technique which became popular in the last years is the large eddy simulation (LES). While in direct numerical simulation (DNS) all length and time scales must be resolved, reducing dramatically the possibility of using this approach in flows of industrial interest, in LES only the large energy-carrying scales of motion are resolved, and the subgrid unresolved scales are mod- eled. Nonetheless, in the near wall region the computational cost is still very high, due to the ne- cessity of capturing small structures behavior in the viscous sublayer. One of the most important Computers & Fluids 32 (2003) 97–108 www.elsevier.com/locate/compfluid * Corresponding author. E-mail address: sarghini@unina.it (F. Sarghini). 0045-7930/03/$ - see front matter Ó 2002 Elsevier Science Ltd. All rights reserved. PII:S0045-7930(01)00098-6