Argonne VERIFI researchers applying GSA to investigate combustion engine parameters; seeking cleaner and more efficient engines
Researchers at Argonne National Laboratory, as part of the new Virtual Engine Research Institute and Fuels Initiative (VERIFI) (earlier post), are using global sensitivity analysis (GSA)—a specific form of uncertainty analysis which breaks down the uncertainty into constitute parts—to investigate a number of parameters in the internal combustion process. By gaining a better understanding of how these parameter uncertainties affect outcomes, the VERIFI researchers, along with colleagues at the University of Connecticut, are seeking to create cleaner and more efficient engines.
The parameters being investigated include the relationships between the diameter of the nozzle in the fuel injector; the dynamics of the fuel spray; the proportion of fuel to air in the combustion chamber; and the exhaust products. In an SAE paper presented at the World Congress this year, the researchers described the results of the first demonstration of GSA for engine simulations.
Global Sensitivity Analysis (GSA) is conducted for a diesel engine simulation to understand the sensitivities of various modeling constants and boundary conditions in a global manner with regards to multi-target functions such as liquid length, ignition delays, combustion phasing, and emissions. The traditional local sensitivity analysis approach, which involves sequential perturbation of model constants, does not provide a complete picture since all the parameters can be uncertain. However, this approach has been studied extensively and is advantageous from a computational point of view.
The GSA simultaneously incorporates the uncertainty information for all the relevant boundary conditions, modeling constants, and other simulation parameters. A global analysis is particularly useful to address the important parameters in a model where the response of the targets to the values of the variables is highly non-linear.—Pei et al.
The baseline in that study was a three-dimensional closed-cycle engine simulation in a 60-degree sector mesh under moderate speed-load conditions. The study first quantified the uncertainties for key model parameters, initial and boundary conditions—a total of more than 30 parameters. They ran 100 simulations by simultaneously varying those parameters, and then calculated multiple targets.
They then applied GSA as a screening method to highlight those parameters the accuracy and adjustments of which were most likely to influence the predictions of a computational model. The parameters with high sensitivities with regards to multi-target functions were identified and a detailed analysis of the important parameters was presented to different target functions.
There are lots of unknowns that are involved. We’re using sensitivity analysis to understand how they all affect overall uncertainty. If we can find a way to understand how uncertainty effects our simulations, we can take a step toward developing a more predictive simulation.—Sibendu Som, Argonne National Laboratory (ANL)
Overall, Som and Argonne mechanical engineer Yuanjiang Pei and chemist Michael Davis have investigated 32 different parameters simultaneously, trying to establish how the uncertainties vary under different conditions.
Building on several decades of work by chemists, statisticians, and applied mathematicians, Argonne chemists have developed the tools to apply GSA to large chemical models in collaboration with their colleagues at the University of Colorado and the University of Leeds.
These techniques were further refined in the last two years to allow their efficient application to engine simulations, leading to the present study, which involves the collaboration with the University of Connecticut.
These new methods demonstrate the benefits of close collaboration between basic and applied research, the researchers said.
This is the first time we’ve applied these methods in such a complicated system. We have demonstrated that GSA can be used in a systematic way for something as complex as an engine simulation.—Doug Longman, ANL
VERIFI researchers are taking an iterative approach in which data gathered from the simulations can be fed back to both engine modelers and combustion chemists to reduce uncertainty further and to create more predictive engine simulations.
What’s unique about VERIFI is the way we’ve refined the tools to create engine simulations that are more reliable and applied high-performance computing resources to run simulations faster and more intensively than ever before.—Sibendu Som
By taking advantage of the computational power available today, the VERIFI team can identify the most important engine and fuel parameters and develop unique engine simulations and analyses to enable optimized engine combustion in the presence of uncertainty at any operating condition. In the near future, the VERIFI team plans to run diesel engine simulations of unprecedented scale on Mira, Argonne’s 10-petaflop IBM Blue Gene/Q supercomputer.
VERIFI is the first and only source in the world for high-fidelity, three-dimensional, end-to-end combustion engine simulation/visualization and simultaneous powertrain and fuel simulation, with uncertainty analysis.
Pei, Y., Shan, R., Som, S., Lu, T. et al. (2014) “Global Sensitivity Analysis of a Diesel Engine Simulation with Multi-Target Functions,” SAE Technical Paper 2014-01-1117 doi: 10.4271/2014-01-1117