Automatic Characterization of Ignition Processes with Machine Learning Clustering Techniques, Blurock, Edward S.; International Journal of Chemical Kinetics, 38 (10): 621-633 (2006).
This paper used conceptual clustering (COBWEB) with fuzzy logic descriptions of the time-dependent concentration curves to characterize an ignition process. The fuzzy logic predicates represented view of the curves as humans see them, namely as maximum, minimum, inflection points, high, low, etc. The characterizations found, the reactive phases, mimics those known by modelers.
Artificial Intelligence: This is an application of machine learning and fuzzy logic to mimic how a human characterizes a reaction ignition process represented by the molecular species concentration over time. Conceptual clustering was used to find the classes representing the reactive phases. The concentration profiles at a specific time, represented by fuzzy logic predicates, were the objects clustered. Fuzzy logic played a key role in the more human-like characterization of the curves. First a modeler thinks in terms of high and low concentrations absolute concentrations. Fuzzy logic softens the transition between high and low. In addition, by using similar high-low or equal/not equal to zero fuzzy predicates on the first and second derivatives gives human-like information about the shape of the curve. Fuzzy logic plays a key role in describing the maximum of the curve by representing the times at and near the maximum (definition of near is determined by the slope of the fuzzy function).
Algorithms and data structures: Though fuzzy logic clustering was at the heart of this work, within the ANALYSIS system, many algorithms and data structures are needed to interface the input and output in a general way. The techniques used in this project were implemented in a general way for use in other similar analyses.
Characterizing Complex Reaction Mechanisms using Machine Learning Clustering Techniques, Blurock, Edward S.; International Journal of Chemical Kinetics, 36 (2), 107-118 (2004).
Use of Atomic and Bond Parameters in a Spectral Representation of a Molecule for Physical Property Determination, Blurock, Edward S.; J. Chem. Inf. Comput. Sci., (1998).