Autodeductive Modeling and Optimization in Chemometrics

Abstract

The average chemist in academic and especially in industrial environment still considers computers and software as tools reserved only for selected specialists. Our experience in interacting with chemical, pharmaceutical and car industry within a global research project in Computer Chemistry [1,2] allows to understand the essential demands regarding experimentation, modeling and optimization. In general, highly redundant measurements are carried out; in most cases there is a lack of deterministic equations describing the physics of the experiment, and data appear to contain much noise. However, it seems sometimes difficult to convince people to involve computer support due to a certain difficulty in interpreting results of programs showing a highly mathematical layout. The average chemist prefers undoubtedly a more semantic, colloquial access to a computer session. This is reflected in the actual trend of creating thinking programs, like the expert systems, which emulate human logic and are used mostly for problem diagnostics. A central point in industrial experimentation is optimization.