Harris Bang


I am a second year Ph.D student in mechanical engineering. I work on human-computer interaction in system architecture design. I am developing a GUI and explanation facilities for system architecting tool, which is a type of knowledge-based expert system that is capable of generating, evaluating, and improving candidate designs. I am also interested in functional allocation of design tasks between the human and the computer.

Many modern engineering systems such as aircrafts, nuclear power plants, and satellite systems are extremely complex, and their scale and scope can be immense. When designing such systems, it is difficult for human designers to track all relevant aspects of a design problem. Therefore, we rely on computation tools to explore the design space and find optimal designs. While the recent efforts have been put on increasing the level of autonomy of these trade-space exploration tools, not much consideration has been taken on how humans interact with these tools. My goal is to come up with a good communication strategy between humans and computers to enable effective collaboration in system architecture design process.

Driving Feature Identification and Surrogate Model Based on Classification Tree

Driving features are sets of decision variables that contribute the most to designs being located at certain regions (usually chosen to be “good” region) of the tradespace. This information is useful because it shows the dynamics of the complex model. These driving features can help the user to better understand the major trade-offs and gain useful architectural insights.

Identifying driving features is an interactive process. The user first has to specify what designs are considered to be good, based on their locations in the objective space. Then association rule mining and feature selection algorithms are used to generate candidate design features. Such information can also be used to build a surrogate model based on classification tree. By observing the surrogate model, the user can gain some insights on what could be the compact form of the driving features. The user can suggest new candidate driving features and test it by repeating the previous steps again.

Driving feature identification process. GUI is implemented using HTML and javascript

Related Paper:

  • Bang, H., and Selva, D. (2016). IFEED: Interactive feature extraction for engineering design, Proceedings of the ASME 2016 International Design Engineering Technical Conferences, Submitted. (pdf)