Design can be broken down into three major activities. 

• Generate option space
• Determine outcome expected at the choice of each option
• Determine value system and rank outcome of each option

The DATLab focuses its efforts on the first step through a variety of activities including designer assistance tools for EGTs and carts; graph and string grammars; representations; AI planning approaches; search techniques.  Please refer to Publications page for citations of specific papers.

Graph grammars and HTN

Graph grammars have been used with carts components to construct carts with a specified amount of surface area for load carrying with a minimum weight.  Because the kits are not 'complete', an assembler has been developed using the assembly problem as a constraint satisfaction problem.  Work on the cart assembly problem, and also generative design, has progressed to HTN planning.  The process works, but is it realistic and creative?

Designer assistance tools for EGTs

Graph grammars are used for the structure synthesis of general mechanisms.  The key is to populate design spaces with non-isomorphic designs.  This project is also in coordination with the NIST Design Repository.  The challenge remains in developing a concept selection module for multi-objective, multi-domain, and multi-decision making tasks.

Design representation

Recent work in design representation have incorporated a consolidation of the many representation views.  Form, function, and behavior are becoming more integral to each other and design representation as a whole. While form and function have achieved a high level of maturity, behavior-based modelers are deficient.  Developing a behavior-based modeler and integrating it with form and function systems is the next step. 

Search techniques with simulated annealing

Generative design requires efficient search techniques over discrete design spaces.  One search technique involves simulated annealing.  This technique has been improved upon in recent years using statistical process control techniques and further improvement in Detecting Productive Search (DPS) is currently being worked on.