Data being used to understand today's mechanical design spaces is both large and complex. Design performance requirements continue to increase, while budgets to produce said performance are being diminished. This has resulted in increasingly higher levels of design complexity, leaving designers with a great challenge in actually under- standing what they are designing. Mentally grasping high-dimensional data, both in general and in mechanical design, is thus becoming increasingly important to industry.

This thesis builds on prior work that introduced a novel way of looking at high- dimensional design spaces using a single contextual self-organizing map (CSOM). Self-organizing maps are first leveraged to find relationships amongst high-dimensional design data. Techniques of the contextual self-organizing map are then leveraged to add a visual interpretation of representative statistical properties of nodes residing within the SOM. The first major contribution of this thesis then breaks up prior work's single map into four separate maps, each representing a different mathematical property. The new four-map visual raises the level of understanding of these visualizations, and does away with the difficulty and potential error in interpretation of the coloring scheme previously used.

Additional updates to the application's user interface add further ease of use with the software, and enhance the directed exploration of resulting maps. An investigator can now filter or brush node properties to focus in on only those nodes matching preferences. Supporting multiple resulting performance metrics additionally allows the investigation of the broader impact of design space variables on design performance metrics.

Using CSOM results in transition to detailed investigation of data was difficult. There was no mapping between the high-dimensional SOM space to physically interpretable, three-dimensional space. Investigators desired the ability to "dive in" to resulting nodes of the CSOM to make this mapping possible. The CSOM application alone did not allow this. The second major contribution of this thesis, an immersive point cloud application, provides this desired mapping.

The point cloud application was developed as a standalone, immersive data inves- tigation tool. Using this tool, raw data sets can be visualized and interacted with on hardware ranging from an immersive CAVETMenvironment to a single desktop monitor. Additionally, a TCP-based server architecture was built into the core of the application to support control "hooks" from external applications. An XML-based command struc- ture was developed as the communication protocol between applications, and opens up the point cloud to external control over a TCP connection. Using the developed com- mand structure, a CSOM user is now able to "drive" any connected instances of the point cloud application, thus allowing interaction with and exploration of data residing within resulting nodes of the CSOM application.

Analyzing high-dimensional nodes of the CSOM in three-dimensional space of the point cloud has shown to open up additional insights over the somewhat abstract CSOM visualization alone. Through the use of the software environment developed in this thesis, designers and investigators may now be capable of greatly reducing the dimensional complexity of a given high-dimensional design space, furthering their understanding and facilitating more robust designs.