Adding error bars to simulation results

Date: 2022/06/17 - 2022/06/17

Academic Seminar: Adding error bars to simulation results

Speaker: Noah Van Dam, Assistant Professor, Francis College of Engineering, University of Massachusetts Lowell

Time: 10:00 - 11:30, June 17, 2022 (Beijing Time)



Simulations have become an important complement to experiments in science and engineering. However, in order to draw robust conclusions from the simulation results the simulation errors must be understood and quantified. This presentation will cover several projects within the Multiphase and Reacting Flows Laboratory at University of Massachusetts Lowell related to quantifying different simulation errors. The first is established the fundamental error expressions for stochastic Lagrangian-Eulerian methods. These methods, which use Lagrangian parcels to represent dispersed liquid fuel drops in the Eulerian gas field, are widely used to simulate fuel sprays. The coupling between Lagrangian and Eulerian frames of reference creates a new source of error that has yet to be fully quantified. Initial expressions have now been derived for Lagrangian-to-Eulerian coupling for general coupling kernels. The derivation of these new expressions will be shared, along with empirical tests demonstrating their accuracy. The next project is an Extended Taylor approach to calculating discretization error for specific simulations. Classical methods all rely on an assumption of asymptotic error behavior, which requires fine grids be used and in practice can be very difficult to achieve for engineering application simulations. The Extended Taylor approach is proposed to enable error estimation for so-called coarse simulations that do not have asymptotic error behavior. Preliminary results will be presented for simple functions to compare the methods performance against the current standard, the Grid Convergence Index. Errors are not just in the numerical methods used, but also simulation set-up. The third project discussed will be a recent study into boundary condition and initialization procedures for a gasoline spray. Different options for injector geometry, injection location, and whether experiment or simulation results are used for rate-of-injection data are compared.


Prof. Noah Van Dam is an assistant professor at the University of Massachusetts Lowell where he leads the Multiphase and Reacting Flows Laboratory. Prior to joining UMass Lowell, he received his Ph.D. from the University of Wisconsin-Madison in 2015 and then worked at Argonne National Laboratory in the Energy Systems division as a postdoctoral associate. His research focuses on high-fidelity, computational fluid dynamics modeling of complex flows, focused primarily on energy applications. Topics of interest include Lagrangian-Eulerian methods for multi-phase flows, machine learning for physics model development, and verification, validation and uncertainty quantification (VVUQ) of numerical simulations. Applications include internal combustion engines, gas turbines, and general multi-phase/spray systems.