A Scale Separation Approach: A Machine-Learned Surrogate Model for an Injector Coupled with Spray Simulations
DescriptionIn this work, we use machine learning to develop a surrogate model for an expensive portion of a computational fluid dynamics simulation. In particular, we learn a surrogate model for fuel injection, predicting the flowfields at the exit of an automotive injector. The output of the surrogate model is then the boundary conditions for reacting spray simulations. Static coupling between the injector flow and external spray via a spatiotemporal boundary condition reduces the range of scales that need to be resolved for the spray domain. The surrogate model includes uncertainty quantification and is able to conserve the injected mass flow rate. The surrogate model results in an O(10^6) speedup for predicting internal flow fields for a new test case, and an O(10) speedup for the full simulation.
TimeMonday, June 2614:00 - 14:30 CEST
Event Type
Computer Science, Machine Learning, and Applied Mathematics