Using Adaptive Response Surface Regression

1440 Words6 Pages
3 Methodology The developed optimisation routine makes use of adaptive response surface regression to use a limited initial amount of FE models to feed an optimisation routine which is specifically designed for general thermal problems where parameters linked to the general heat equation can be optimised or estimated using experimental input data. The algorithm uses a pan and zoom function to move through the design space and delivers faster predictions with fewer iterations than standard updating routines [35, 41]. 3.1 Adaptive response surface method The adaptive response surface optimisation routine is used to optimise numerical models with a lot of data points and the time reducing by the algorithm increases as the number of parameters rises [40]. The routine is designed to handle multiple-output time series data [35]. The optimisation procedure can be divided into following steps: 1. Starting reference simulation points are ran and a correct object function is built of the difference between the FE model and the target value (experiment or validation model). 2. The FE model is replaced by a meta-model of response surfaces to decrease the optimisation time but remains an accurate approximation. 3. The optimisation routine is run on a specific object function. It is possible to use multiple objective functions or build an objective function related to multiple outputs. 4. The estimated parameter values are used as input parameters for a new FE model that corrects the
Open Document