To cope with the increasing complexity during the engine calibration process, experimental design and model based methods have become widely used. For large numbers of engine parameters, however, test planning, data gathering and modelling can still be excessively time consuming. In this article, strategies are outlined to tackle the problems of test plan creation and modelling in high dimensional input spaces. An approach to reduce the number of free model parameters and, consequently, the number of measurements is demonstrated for a Diesel engine emission model with nine input parameters. Two methods are proposed that increase the efficiency and quality of the experimental design and the modelling process. The first method (DoE approach) provides a means to use the convex hull information for the definition of candidate test points. This increases the efficiency of the design algorithm as well as the volume of the design space. The second method (split-model approach) factorizes the test space according to independency criteria between model inputs. Consequently the number of polynomial interaction terms decreases which leads to a significant reduction of measurement test points. Both methods accomplish a better fit between model and physical system and increase hence the quality of the resulting model. The presented theoretical results are confirmed by the application of these methods to a Diesel engine modelling process.
Strategies for improving the process of test design and test plan computation for high dimensional designs
2011
12 Seiten, 12 Bilder, 3 Tabellen, 4 Quellen
Aufsatz (Konferenz)
Englisch
System design and effectiveness improving built-in-test designs
Tema Archiv | 1987
|Development Test Plan (Design) Dyna-Soar
NTIS | 1960
|NTRS | 2002
|NTIS | 1966