Supersonic Transport
The work is divided into the following major areas.
The Model/Simulation Associate software modules control the computational simulations used to evaluate candidate designs, returning the evaluation to a Design Associate software module, which controls the search of a space of possible candidate designs.
Supersonic Transport
We recently completed and reported on the use of modeling constraints to mediate search control during the conceptual design of a supersonic transport. This work demonstrated an order of magnitude speedup in finding the best design for a given mission. Technical details of this work are web accessible.
Intelligently controlled multi-level design and design space decomposition have considerable potential for improving performance of an automated design system.
Four bar Exhaust Nozzle
We have recently completed experiments using multi-level design techniques to automatically design both the airframe and exhaust nozzle of a supersonic transport. This design was carried out by first searching a high-level (i.e., more abstract) design space and then using these results to seed the search of a lower-level design space. In this experiment the multi-level approach reduced the design cost by an order of magnitude. Technical details of this work are web accessible.
The Designer's Interface is a framework to standardize interaction between programs. Each program is categorized as either an information server (e.g. an analysis tool) or a client (e.g. an optimizer).
Clients may call server programs repeatedly, letting the Interface deal with execution details. This includes handling crashes or stopping runaway processes, as well as allowing parallel execution locally or on remote machines.
When new program modules are adapted to the Designer's Interface, they become immediately available for use with previously incorporated methods. For example, if a new simulator (server) is constructed, it can be used with all incorporated optimization methods. This grants easy access to a wide range of utilities, and facilitates code re-use. More information can be found at the Designer's Interface Overview.
A particular form of AI-augmented optimization is the use of problem reformulation to improve optimizer performance. Problem reformulation has long been an important area of theoretical AI research, and our work demonstrates that problem reformulation can have practical benefits as well.
Physical systems may be modeled at many levels of abstraction, and properly using different abstractions and approximations is a key component of automated designs.
In collaboration with colleagues at Rutgers (Doyle Knight and Gecheng Zha) and at United Technology Research Center (Marty Haas) we have applied AI-augmented optimization methodologies to the design of an inlet for an air breathing supersonic strike missile being considered by the Navy as a replacement for the Tomahawk. This work is described in a Design of Propulsion systems for aerospace vehicles.