High-fidelity numerical simulations have already contributed in a significant way to the emergence of state-of-the-art turbomachinery components but are still sparingly exploited due to their cost. It is hence relevant to continue the investigations on low-fidelity tools and their combination with higher fidelity methodologies especially when breakthrough technologies are studied. Multi-fidelity approaches are key in supporting turbomachinery engineers to quickly decide on the relevance of an innovative concept, and to accelerate optimisation processes.
This paper describes a low-noise propeller design methodology developed during the IRON project (Clean Sky 2). This methodology is driven by Calypso, a platform developed by Cenaero, which integrates machine learning strategies and multifidelity aerodynamic and acoustic solvers. Two key elements of this framework are detailed and validated in this paper. The first one is a low-fidelity tool that offers a fast estimation of the steady aerodynamic performance of a propeller blade using a lifting-surface method. The second one is a far-field tonal noise prediction solver, relying on an acoustic propagation performed with Farassat’s formulation 1A of the Ffowcs-Williams and Hawkings equation. Noise source can be either obtained after a CFD computation or after a low-fidelity aerodynamic evaluation.
The low and high-fidelity aero-acoustic methodologies have been applied on 600 samples. It is shown how data mining strategies available within Calypso pave the way to astutely combine both levels of fidelity within an optimisation to quickly and efficiently explore a large design space, and to identify the key design parameters as early as possible in the design process.