Abstract
Conventional CAD systems excel at detailed modelling but often fail to support the early conceptual stages of design. This paper identifies a "representation gap" where designers must prematurely commit to precise dimensions. We propose a framework for vague geometric modelling that allows for incomplete and approximate information, enabling a natural evolution from conceptual sketches to definite designs.
Introduction: The Representation Gap
In engineering, the early conceptual phase determines approximately 70% of a product's life cycle cost, yet only a fraction of the total project effort is spent here. Designers typically work with "fuzzy" ideas—vague shapes, approximate sizes, and relative positions. However, standard CAD systems require exact coordinates and rigid parameters. This forces designers to either over-specify their models too early or abandon digital tools until the design is nearly finalized.
To bridge this gap, we introduce a computational scheme based on the "minimum commitment" principle, allowing designs to evolve incrementally without forcing unnatural decisions.
A Taxonomy of Geometric Vagueness
The core of our approach is a classification of geometric information based on its level of definition. We distinguish between two primary types of vagueness:
- Incomplete Information: Elements are known to exist, but their specific values are not yet defined (e.g., an object has a width, but the value is unknown).
- Approximate Information: Values are known but imprecise, often represented as ranges or intervals (e.g., "height is roughly 15mm").
As shown in the taxonomy above, geometric characteristics like shape, size, orientation, and location can be categorized by whether they are complete/incomplete or precise/approximate.
Manipulation Classes for Vague Models
To support the design process, we identify three functional categories of operations required to manipulate these vague models:
- Synthesis: Operations to construct, modify, and evolve the model. This includes abstraction (simplifying geometry via skeletonising or packing) and concretisation (adding detail as the design matures).
- Analysis: Tools for displaying, calculating, and explaining the properties of the current model state.
- Re-use: Mechanisms for recording and retrieving previous design solutions to inform new projects.
Implementation: The GEMCON Prototype
These concepts were implemented in GEMCON, a prototype system developed in Common Lisp. GEMCON utilizes a hierarchical tree structure of "geom" objects, allowing for multi-level modeling where configurations can be modified at any level of detail.
Constraint-Based Reasoning
Instead of single numeric values, GEMCON represents parameters as intervals (lower and upper bounds). This allows the system to maintain a "region of possible positions" rather than a fixed point. Furthermore, it supports spatial logic, allowing designers to define locations using relative terms like "above" or "behind" rather than absolute X, Y, Z coordinates.
Non-Sequential Design
A key feature of the system is its support for non-sequential workflows. Designers can define any part of the geometry at any stage. The system does not break if a parameter is missing; it simply maintains the vagueness until more information is provided.
Conclusion and Future Directions
The GEMCON prototype demonstrates that incremental refinement of vague models is computationally feasible. By allowing designers to work with "roughly" and "between," we can better align CAD tools with the human design process.
Future work will focus on expanding the system's ability to handle complex vague shapes beyond simple primitives and exploring how these conceptual models can be seamlessly integrated into industry-standard detailed CAD kernels.
Summary updated