Turbomachinery flows are characterized by complex multiscale phenomena, including transition, turbulence, large-scale unsteadiness and threedimensional effects. For turbomachinery working in the compressible regime, the complexity can be exacerbated by the presence of shock waves interacting with the surrounding boundary layers and wakes, and by real gas effects.
In this state of matter, so-called high-fidelity simulations, which resolve all or a large range of turbulent scales, are an essential tool to improve our understanding of the underlying physical mechanisms and to develop improved low-fidelity models. An attractive option consists in discovering such models from high fidelity data by using modern machine learning techniques. In my talk I will first present selected examples of high-fidelity simulations of turbomachinery flows, with or without real-gas effects. Subsequently I will present a machine learning approach for discovering improved models. Finally, I will discuss how machine learning can help fusing together low and high-fidelity data to obtain more reliable turbomachinery designs.