Engineering practice is nowadays inconceivable without the presence of computational tools. Within this context, Computational Fluid Dynamics (CFD) are an essential tool for fluid-based machine design, such as heat exchangers, turbines, cooling processes or aerodynamic performance of vehicles. Among the simulation capabilities of modern softwares, Reynolds-Averaged Navier Stokes (RANS) simulations are the most popular industrial approach, due to the decent computation elapsed times and accuracy for a vast range of applications. However, some engineering applications that simulate complex flows may exhibit certain discrepancies as a consequence of neglected sources of uncertainty. The effect of uncertainty can be even increased when the effect of different sources of inaccuracy are combined in the simulation.
Once a reliable computational model is achieved, further designs can be explored. One advantage of CFD is that prototyping costs can be reduced by performing optimisation via simulation. This allows to obtain a large number of data at lower cost than experimental testing. Thus, such data can be further used to train Machine Learning algorithms that may improve or speed up the optimisation process.
In this presentation, the aforesaid concepts will be shown. Different examples of uncertainty propagation in CFD simulations of engineering applications will be illustrated. Finally, a successful case of Machine Learning aided optimisation of a mechanical micro heat exchanger/mixer will be presented. This research is supported by the UMA18-FEDERJA-184 funding.