Abstract
PROMPT is a design system that moves beyond simple "knob-turning" parameter optimization to perform genuine structural innovation. By combining a "Graph of Models" for physics reasoning with specialized "Modification Operators," the system can "invent" new component geometries—like transforming a solid rod into a hollow tube—to satisfy conflicting design constraints that traditional CAD systems cannot meet.
Introduction: Bridging the Creativity Gap
In the late 1980s, most automated design systems relied on hierarchical refinement: they selected components from a fixed library and tweaked their parameters (like length or radius) to meet constraints. However, if the library didn't contain a part that worked, the system failed.
The authors identify a "creativity gap." For example, if a design requires a beam that is both very stiff and very light, simply increasing the radius of a solid rod is insufficient because weight increases faster than stiffness. A human engineer would solve this by changing the topology—making the beam hollow or I-shaped. PROMPT was built to bridge this gap by allowing a system to reason about the underlying physics of a part and modify its actual structure, not just its dimensions.
Methodology: The Graph of Models
PROMPT employs a multi-level approach to design that balances efficiency with deep reasoning. The first line of defense is constraint propagation, which attempts to satisfy requirements by varying known parameters of existing prototypes.
If simple tuning fails, PROMPT utilizes a Graph of Models to represent physics knowledge. Each node in the graph is a mathematical model based on specific assumptions (e.g., "slender member," "elastic material"). Edges represent the adding or relaxing of these assumptions. This allows the system to find the right level of abstraction for analysis without getting bogged down in unnecessary complexity.
Modification Operators and Structural Innovation
The core innovation of PROMPT lies in its Modification Operators. These are "precompiled" heuristics that bridge the gap between expensive first-principles analysis and simple parameter changes. A Modification Operator consists of applicability conditions, analysis heuristics, and structural heuristics (e.g., "move mass from low-load areas to high-load areas").
A primary example is the Mass Redistribution Operator (MRO). By analyzing the stress distribution of a solid rod under torsion, the system realizes that the center of the rod carries almost no load.
The MRO suggests moving that "useless" mass to the outer edges where the load is highest. This effectively "invents" the concept of a hollow tube. Unlike systems that treat shape as a fixed attribute, PROMPT treats shape as a variable that can be elongated or hollowed out based on the direction of applied loads.
Results and Future Directions
The paper demonstrates PROMPT’s capabilities through qualitative case studies in beam design. When faced with conflicting weight and torsional stiffness constraints, PROMPT successfully modified a solid cylindrical prototype into a hollow tube. When an additional bending constraint was added, the system used a "Shape Modification Operator" to elongate the cross-section into an ellipse, increasing stiffness in the direction of the load without compromising torsional performance.
While powerful, the authors acknowledge that the "Analyze-Modify" loop is computationally heavy. Future work aims to expand the library of Modification Operators to different engineering domains and further integrate complex stress analysis to move toward truly "creative" autonomous design systems.
Summary updated