Mathematically Reduced Chemical Reaction Mechanism Using Neural Networks

This is an annual technical report for the work done over the last year (period ending 9/30/2005) on the project titled ''Mathematically Reduced Chemical Reaction Mechanism Using Neural Networks''. The aim of the project is to develop an efficient chemistry model for combustion simulations. The reduced chemistry model will be developed mathematically without the need of having extensive knowledge of the chemistry involved. To aid in the development of the model, Neural Networks (NN) will be used via a new network topology know as Non-linear Principal Components Analysis (NPCA). We report on the significant development made in developing a truly meshfree computational fluid dynamics (CFD) flow solver to be coupled to NPCA. First, the procedure of obtaining nearly analytic accurate first order derivatives using the complex step method (CSM) is extended to include computation of accurate meshfree second order derivatives via a theorem described in this report. Next, boosted generalized regression neural network (BGRNN), described in our previous report is combined with CSM and used to obtain complete solution of a hard to solve wave dominated sample second order partial differential equation (PDE): the cubic Schrodinger equation. The resulting algorithm is a significant improvement of the meshfree technique of smooth particle hydrodynamics method (SPH). It is suggested that the demonstrated meshfree technique be termed boosted smooth particle hydrodynamics method (BSPH). Some of the advantages of BSPH over other meshfree methods include; it is of higher order accuracy than SPH; compared to other meshfree methods, it is completely meshfree and does not require any background meshes; It does not involve any construction of shape function with their associated solution of possibly ill conditioned matrix equations; compared to some SPH techniques, no equation for the smoothing parameter is required; finally it is easy to program