Grantee Research Project Results

Since the start of the year, an automated mechanism reduction algorithm and supporting code has been developed in python with mechanism testing programs developed in Matlab.

The core algorithm is adapted from the directed relation graph (DRG) method developed by Lu and Law. Adjustments were made to this algorithm to account for the specific goals and parameters of this project. This algorithm maintains chemical pathways between isoprene and all species that are deemed important as the mechanism is reduced. In addition, three different weighting schemes are used for the mechanism reduction that are similar to but distinct from the weightings originally published with the DRG method.

In addition, new methods of mechanism reduction have been developed which have significant promise for improving upon the DRG method, but these methods have not yet been fully implemented for comparison.

Mechanisms are reduced to a desired size, and a simple metric for mechanism complexity was developed to measure the size of mechanisms. In order to account for a range of input parameters, the algorithm can be run with multiple inputs to calculate multiple weighting schemes which are then averaged to produce one final mechanism.

The reduced mechanisms are converted to a F0AM readable format and input into Matlab. They are compared to the full mechanism they originated from in order to quantify their performance. An error metric was developed based on the difference in concentration of important species between the reduced and full mechanism. The error is averaged over multiple samples and species to determine an overall performance for the mechanism.

The full mechanism was adapted from the complete isoprene mechanism published by Wennberg et al. at CalTech. This mechanism did not include complete chemical pathways for some species, however, these pathways were included in the CalTech reduced mechanism. An updated mechanism was created by extrapolating chemistry from the reduced Caltech mechanism into the full Caltech mechanism. This updated mechanism is the current working mechanism for reduction. Further testing needs to be done to confirm that this updated mechanism matches measured data.

The automated reduced mechanisms tend to perform worse as the size of the mechanism decreases, which is the expected result. These mechanisms have comparable performance to other published mechanisms when compared to the full mechanism. These results suggest that the automated mechanisms are on the right track, however more extensive analysis needs to be done to fully assess their performance.