Previous Software Systems
These software systems were web-based , meaning that the user interface and the programs which do the computations can be on different computers (anywhere in the world), and they are platform independent , meaning that they can be used on any computer, regardless of the operating system of the computer. This was done to facilitate international cooperation. The JAVA user interface can be found on numerous public domain web-sites. The source code for the systems is currently available on SourceForge (though not 100% user friendly in its installation, due to the many computer languages, C, C++, JAVA, XML and some web technologies, that used in its implementation). It should said that since a great deal of my financing has been through industrial projects, the output concentrated on concrete deliverables and not publications. This is particularly true of my time at RISC in Austria.
This is a highly object-oriented system, complete with GUI interface, for the interaction of a palette of artificial intelligence, machine learning, statistics and fuzzy logic algorithms. This includes the fundamental techniques for the methods that I have developed for mechanism reduction. It was originally developed for the REFORM Esprit project for the analysis and modeling of quality control data in Steel production. Its generality lends now to its use in mechanism generation and reduction.
The machine learning implementations were done is a general way, especially to be able to include generalized fuzzy logic (not restricted to any particular formulation). The/my interpretation of the fuzzy logic in terms of machine learning was probabilistic in nature. I also experimented (with implementations) with a generalized genetic algorithm and generalized rule-based reasoning system. All the implementations, whether algorithms or data types, were done is as general and as extensible as possible.
One of the principle philosophies of this software was not to base analysis on a single method, but to be able to efficiently combine algorithms and communicate data structures. To this end, a complete sub-system for the management of algorithms and the establishment of a goal-oriented structure was created. In addition, the data structures themselves were designed to be configurable. A typical example is the fuzzy logic predicate (derived from a general function class). This could be configured (not hard coded, but by a configuration table) to use an arbitrary function to be the basis of the “fuzzy”. For example, ramping function (for greater-than, less-than) or trapezoid or pyramid (for equal-to/not equal-to) derived from piece-wise linear functions (where the “piece-wise” could also be externally configured).
This is built on top of the ANALYSIS++ system for the analysis and manipulation of chemical data. My main use of it now, though other capabilities exist, is the generation and analysis of complex kinetic mechanisms in combustion. This has been successfully used in the generation of diesel and soot mechanisms for several European Community and Swedish projects.
This is a numeric package for the large scale tabulation of kinetic data for use in computational fluid dynamic calculations. This software is the basis for the Swedish Science Ministry (Vetenskapsrådet) project.
This is software I have developed with the group of Frédérique Battin-Leclerc for the determination of thermodynamic data from basically the Lewis structure description (or, said another way, 2D-graphical structure description). This is the level of chemical bonding that is used in automatic detailed combustion mechanism generation. It has been integrated into my own REACTION system and the EXGAS system of Battin-Leclerc.