Phase Optimized Skeletal Mechanisms for Engine Simulations, Blurock, Edward S., Tuner, Martin, Mauss, Fabian, Combustion Theory and Modeling,14:295-313 (2010).
This is an application combining multiple machine learning methods, numeric algorithms and software technology to automatically produce a set of reduced sub-mechanisms (meaning they are computationally faster) to mimic a single full mechanisms. The sub-mechanisms are determined automatically. The functionalities within REACTION and ANALYSIS were used to carry out these tasks.
Artificial Intelligence: The core to the automation of this application is the use of conceptual clustering (COBWEB) on fuzzy logic predicates to automatically determine phases of a reactive process. The parameter on which the fuzzy logic predicates are based is directly connected to the quantification of whether the species is needed in the description. Fuzzy logic was key in this analysis giving essentially the information yes-no-maybe with “maybe” being the fuzzy transition between the species absolutely needs to be in the mechanism or absolutely does not need to be in the mechanisms. This use of the parameter accounts for cases where the parameter maybe does not give a proper answer. Thus the sub-mechanisms, represented by the clusters, can be reduced. A decision tree regression tree analysis, a generalization of the ID3 method with fuzzy logic, is used for the real-time decision structure.
Algorithms and data structures: The automated process, from numerical analysis of the original mechanism, division into clusters, the determination of a decision algorithm and finally the conversion to a working FORTRAN program, is a complex set of algorithms, software technical methods and data structures. Interface tools are available to guide the parameters of the process.
Another key publication in this area:
Phase Optimized Skeleton Mechanisms for Stochastic Reactor Models for Engine Simulation, Tunèr, M., Blurock, Edward S. and Mauss, F.; SAE 2005-01-3813 (2005).