Reduced order modeling for supersonic and hypersonic aerodynamic flows via probabilistic machine learning
PI: Ashwin Renganathan (Aerospace)
The tuition for the student selected for this award will be covered from my grants.
The goal of this project is to develop a novel, data-driven reduced order modeling methodology and associated software, applicable to next-generation aerospace and defense applications. Specifically, we will focus on supersonic and hypersonic aerodynamic flows, which are highly convection-dominated, leading to “shocks” which are difficult to model and emulate with existing reduced order modeling techniques. This project will develop strategies, founded on computational science, machine learning, optimization, and software development, to overcome these limitations. If successful, our tool can drastically reduce turn-around times and cost for design of revolutionary next-generation aerospace systems, thereby contributing to US leadership in this domain.
Background. Reduced order modeling is a mathematical method to reduce the computational complexity of numerical solutions to ordinary and partial differential equations (PDEs). PDEs find widespread application in aerospace engineering, including modeling the aerodynamics of commercial, combat, and space exploration vehicles, amongst others. For practical applications, PDEs are solved on complex spatio-temporal domains which are discretized into finite elements or volumes—the computational cost of solving the PDE scales cubically as the number of elements/volumes and hence is not scalable. Reduced order modeling provides a mathematically sound way of reducing the number of finite elements/volumes by a drastic amount for a typically negligible trade in accuracy. As a result, reduced order modeling can offer a paradigm shift in the way advanced aerospace concepts are designed by reducing design costs and improving design reliability.
List of expected skills. Numerical analysis (solving PDEs), Python programming, highperformance computing (usage of SLURM, MPI4Py), PyTorch, GPyTorch, and BoTorch.
List of specific objectives for work supported by this call. The graduate student will spend 50% of their time working on this project. This includes (i) conducting research, (ii) service to ICDS, and (iii) involvement in collaboration with other ICDS faculty. We will use this effort to generate preliminary data that will then be used to submit proposals to DOD agencies (AFOSR and ONR) as well as manuscripts for publication.
Connection to ICDS mission. We will develop probabilistic AI/ML methods to reduce, interpret, and learn data. This project will include both large-scale data generation by running f inite-volume based multiphysics codes on Roar Collab, as well as developing AI/ML methods on that data with GPU acceleration.