The predictive capability of the lattice-Boltzmann very large eddy simulation (LBM-VLES) methodology was evaluated for industrial gas turbine exhaust diffuser-collector applications. The effort focused on evaluating the accuracy of performance predictions against experimental rig data gathered at engine representative conditions including the Reynolds number, Mach number and inlet flow conditions. The performance prediction capability of LBM-VLES simulations was also compared against Reynolds-averaged Navier-Stokes (RANS) simulations. Evaluations were completed for two distinct exhaust rig geometries at conditions indicative of their respective design points, and the performance prediction capability was also evaluated at conditions depicting two off-design cases. The off-design conditions were represented by an increase in inlet swirl angle and associated incidence on the diffuser struts.
Overall, the authors found that LBM-VLES simulations were a suitable approach for predicting the performance of gas turbine exhaust diffuser-collector systems at both on- and off-design conditions. The LBM-VLES simulation accuracy was substantially better than that achieved by RANS simulations, with 66% lower RMS pressure recovery error between CFD and experiments for the four cases studied, and an 89% reduction in error for the furthest off-design case. The computational cost of the transient LBM-VLES simulations was roughly the same as the RANS simulations performed using best practices for that solver.