.png){kind=icon}
.png){kind=icon}
Return
Design space exploration using Self-Organizing Map based adaptive sampling
.png){kind=icon}
OA
AI
DOI：10.1016/j.asoc.2016.02.036.png)
.png){kind=icon}
.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
In engineering design, a set of potentially competitive designs is conceived in the early part of the design process. The purpose of this research is to help such a process by investigating algorithm that enables approximate identification of a set of inputs of real variables that return desired responses from a function or a computer simulation. We explore sequential or adaptive sampling methods based on Self-Organizing Maps (SOM). The proposed method does not rely on parameterized distributions, and can sample from multi-modal and non-convex distributions. Furthermore, the proposed merit function provides infill characteristics by favoring sampling points that lay farther from existing points. The results indicate that multiple feasible solutions can be efficiently obtained by the new adaptive sampling algorithm. The iterative use of the SOM in density learning to identify feasible or good designs is our new contribution and it shows a very rapid increase in number of feasible solutions to total number of function evaluation ratio. Application examples to planing hull designs (such as in powerboats and seaplanes) indicate the merits of the feasible region approach to observe trends and design rules. Additionally, the well distributed sampling points of the proposed method played favorable effect in improving the prediction performance of a classification problem learned by Support Vector Machine. (C) 2016 Elsevier B.V. All rights reserved.
Keywords:
Optimization
Simulated-annealing
Self-Organizing Map
Density learning
Feasible region
AI Summary
Key information extracted from the uploaded paper, including a brief overview, abstract, background, key highlights, visual analysis, and future outlook.
Journal
IF:
6.6
Papers:
1.4W
Citations:
4.8W