Designers of large and complex engineered systems are constantly in need of decisionmaking aids to sift through the enormous amounts of data produced through simulation and experimentation. However, understanding the design space of high-dimensional systems in a comprehensive manner is extremely difficult when using conventional methods. Visualization techniques, such as self-organizing maps (SOM), offer powerful methods to portray these systems with the goal to produce simple to understand representations that quickly highlight trends to support decision-making. A drawback to using SOMs is the clustering of promising points with predominately less desirable data. This paper applies a cluster analysis technique to SOMs to segment a high-dimensional dataset into "meta-clusters". The visual tool, generated by the created cluster analysis technique, can not only highlight the optimal designs in terms of the desired output, but as well reveal respective designs containing similar characteristics. The paper will describe the algorithm created to establish these meta-clusters through the development of several computational metrics involving inter and intra cluster densities. A case study of a satellite design problem is presented using this algorithm to show how optimal designs can be easily located within the visualization for aiding in decisionmaking.
展开▼
