Mon, Dec 04, 2017 @ 11:00 AM - 12:00 PM
Aerospace and Mechanical Engineering
Conferences, Lectures, & Seminars
Speaker: Gianmarco Mengaldo, Senior Postdoctoral Scholar/California Institute of Technology
Talk Title: To Be (High-Order), Or Not To Be: A Perspective on Next-Generation Computational Tools for Engineering Applications
Abstract: Advanced computational tools in applied science are becoming increasingly crucial for the analysis, design and decision-making processes commonly required to drive technological and societal innovation. In engineering, high-fidelity simulations constitute an essential mean to provide otherwise unreachable insights that can significantly improve the understanding of complex systems. An area of particular interest is computational fluid dynamics (CFD), where the adoption of high-fidelity simulation technologies, namely large-eddy simulation (LES) is considered of paramount importance to advance the fields where a detailed understanding of flow physics is critical (CFD Vision 2030 Study, Slotnick et al. 2014). LES can provide substantial advantages over more commonly used steady-state tailored techniques. In fact, LES can significantly extend the simulation predictive skills to off-design conditions that include unsteady separated flows, thus providing high-resolution data that can be used in the analysis and design processes as well as to devise reduced-order models.
The key enabler to develop such high-fidelity CFD tools is the underlying numerical discretization, as it drives the accuracy, robustness and time-to-solution of the simulations. In this talk, I will provide an overview of the challenges that CFD is facing in the near future focusing on key aspects that the community needs to consider to push the current boundaries, namely (i) the development of numerical discretizations that can accurately handle complex geometries and that are competitive in terms of time-to-solution and robustness, (ii) the better understanding of numerical vs. physical dissipation to improve LES models and (iii) the co-design of software and hardware to achieve extreme computational performance on heterogeneous computing architectures. The discussion will primarily debate whether high-order finite element methods (also referred to as spectral element methods (SEM)), are a valid choice to build high-fidelity tools for next generation CFD or low-order alternatives are also an adequate option. From this perspective, I will show how SEMs are extremely competitive to describe unsteady separated flows over complex geometries for applications where accuracy is of paramount importance. On the other hand, I will emphasize the role of low order methods - e.g. mimetic finite-volume immersed boundary formulations - in effectively tackling problems with moving surfaces, where body-fitted discretizations may struggle, and for applications where resolution and accuracy are weaker constraints. The talk will also include some ideas from the weather community, where similar issues are being discussed. While the initial question might not have a definitive answer, I hope to provide a clearer view on its implications and on how "to be high-order" can be extremely competitive for certain applications and less for others.
Biography: Gianmarco Mengaldo is a senior postdoctoral scholar at the California Institute of Technology and he actively collaborates with Imperial College London, the Massachusetts Institute of Technology, University of Cologne and the European Centre for Medium-Range Weather Forecast (ECMWF), the world leader in numerical weather prediction. Gianmarco graduated among the top of his class from Politecnico di Milano, one of the most prestigious Italian universities, with a master of science in aerospace engineering. He went on to obtain a PhD from Imperial College London in aeronautical engineering, where he worked on novel approximation strategies for partial differential equations, including discontinuous Galerkin and flux reconstruction approaches, using the spectral element library Nektar++. During the PhD, he joined McLaren Racing for an internship in the Formula 1 R&D department where he worked on the aerodynamic design of the competing car. After the PhD, Gianmarco worked for one year at ECMWF leading the technical side of a project, ESCAPE, devoted to test several numerical algorithms for weather and climate simulations on emerging computing technologies and he contributed building the new data-structure for handling different numerical discretization for massively parallel weather applications. Currently, he is working at the California Institute of Technology, where he is developing next generation numerical algorithms for multi-scale and multi-physics problems, with applications that include energy harvesting, bio-inspired micro aerial vehicles and drones, among others. He is also an active senior developer of Nektar++, where he leads the development of discontinuous spectral element discretizations for compressible flow problems and continues his collaboration with ECMWF on a range of topics, including the evaluation of different numerical discretization strategies for weather and climate applications on next generation hardware.
Host: Department of Aerospace and Mechanical Engineering
Audiences: Everyone Is Invited
Contact: Ashleen Knutsen