Fast Prediction of HCCI Combustion with an Artificial Neural Network Linked to a Fluid Mechanics Code 2006-01-3298
We have developed an artificial neural network (ANN) based combustion model and have integrated it into a fluid mechanics code (KIVA3V) to produce a new analysis tool (titled KIVA3V-ANN) that can yield accurate HCCI predictions at very low computational cost. The neural network predicts ignition delay as a function of operating parameters (temperature, pressure, equivalence ratio and residual gas fraction). KIVA3V-ANN keeps track of the time history of the ignition delay during the engine cycle to evaluate the ignition integral and predict ignition for each computational cell. After a cell ignites, chemistry becomes active, and a two-step chemical kinetic mechanism predicts composition and heat generation in the ignited cells.
KIVA3V-ANN has been validated by comparison with isooctane HCCI experiments in two different engines. The neural network provides reasonable predictions for HCCI combustion and emissions that, although typically not as good as obtained with the more physically representative multi-zone model, are obtained at a much reduced computational cost. KIVA3V-ANN can perform reasonably accurate HCCI calculations while requiring only 10% more computational effort than a motored KIVA3V run. It is therefore considered a valuable tool for evaluation of engine maps or other performance analysis tasks requiring multiple individual runs.
Citation: Aceves, S., Flowers, D., Chen, J., and Babajimopoulos, A., "Fast Prediction of HCCI Combustion with an Artificial Neural Network Linked to a Fluid Mechanics Code," SAE Technical Paper 2006-01-3298, 2006, https://doi.org/10.4271/2006-01-3298. Download Citation
Salvador M. Aceves, Daniel L. Flowers, J.-Y. Chen, Aristotelis Babajimopoulos
Lawrence Livermore National Laboratory, University of California Berkeley, University of Michigan
Powertrain & Fluid Systems Conference and Exhibition