Knowledge discovery process for scientific and engineering data

摘要

Scientists and engineers are often confronted with the problem of modeling the physical laws that govern complex processes and systems. This task may generally be accomplished following traditional modeling procedures. However, when dealing with multivariate problems and/or huge quantities of experimental data, the modeling problem can easily become unmanageable. In such cases, knowledge discovery techniques may help to address this problem. Current knowledge discovery methods however rely mainly on inductive data mining techniques and do not make use of the structural properties of the specific physical context. Hence, they are not yet the ideal process solution for discovering functional models in science and engineering. This paper discusses a knowledge discovery process, which combines deductive and inductive reasoning techniques to find out mathematical models of physical systems. In the supplementary deductive process, the technique of dimensional analysis is used. This allows the incorporation of background knowledge of the involved domain to enrich the general process of knowledge discovery. The background knowledge forms hereby the specific context for a knowledge discovery process for concrete scientific data. As an example, the introduced method is used to find out the expression of the drag force that a viscous fluid exerts on a submersed and uniformly moving solid. The various issues that arise in the development and implementation of such a knowledge discovery system based on the method of dimensional analysis are analyzed and discussed.