COMPUTATIONALLY SIMPLE XEROGRAPHIC SUBSYSTEM MODELS FOR SYSTEM MODEL SYNTHESIS USING MARS

摘要

This paper presents the application of the MARS (Multivariate AdaptivernRegression Splines) (Friedman [1])algorithm as an approach for developingrncomputationally simple input-output models of subsystems. The algorithm willrnbe used to construct models from experimental data as well as to developrnsimpler approximate models from computationally expensive finite elementrnmodels over a range of design variables of interest. It will be demonstrated thatrnthese simplified models can then be used to construct more elaborate systemrnmodels that consist of a large number of interacting subsystems. The systemrnmodel can be used for performing a large number of what-if studies andrnstudying the interactions between the subsystems. The examples presented inrnthis paper will be from the xerographic design domain although this approachrnto modeling can be applied to other engineering domains as well. Specificrnadvantages of this approach will be highlighted including its ability to handlernmoderately large data-sets in a semi-automated manner.