Steffen Brückner, Stephan Rudolph
Knowledge Discovery in Technical Data
in: Proc. of SPIE Vol 4730, Data Mining and Knowledge Discovery: Theory, Tools, and Technology IV
Orlando, FL, USA, 1-4 April 2002

Abstract. In many engineering applications numerical software has reached a more or less satisfactory quality of predicting the system behaviour. A major disadvantage with this kind of software is that it can only be used in a later step in the engi-neering design process since it requires a detailed system model, such as a finite element simulation model for structural mechanics. Finite element software can only give satisfying results when the complete geometry and all material pa-rameters are specified. However, despite all the parameter definitions in such simulation models, still a severe valida-tion effort with experiments is needed to investigate the model abstractions. Especially in the early conceptual design phase, a need for simplified modelling and the prediction of the system behaviour using only little knowledge about a new design exists. This kind of conceptual knowledge can be given e.g. in simple algebraic equations. These equations can either be derived from first principles or from knowledge discovery in data of previous designs, the latter being the topic of this work. Huge amounts of experimental data have been recorded and stored by industry especially in the past ten years with microcomputers being available throughout the companies. Additionally the engineering domain, other than e.g. the business domain, has the advantage that at least a small number of planned experiments can be conducted to enhance the data quantity and quality and to validate the knowledge discovery results. This paper emphasizes the need for a modified knowledge discovery process for engineering (and other scientific domains) and shows the differ-ences to the traditional knowledge discovery in data bases.

Keywords: Keywords: knowledge discovery in scientific data, data mining, dimensional analysis

back