Abstract
Interpreting engineering drawings is a complex abductive task where geometric shapes can represent various functional components depending on context. This paper presents a method for simplifying the computationally expensive process of visual analogy by leveraging functional knowledge. By mapping shapes from a known source drawing to an unknown target drawing using a hierarchical model, the system significantly reduces the search space for analogical mapping.
Introduction
When human engineers view a technical drawing, they perceive functional assemblies like pistons and crankshafts rather than mere collections of lines and circles. For computational systems, however, this interpretation is an "abductive" problem: a circle does not inherently signify a crankshaft except within a specific design context.
The primary challenge lies in analogical mapping. If a system understands a source drawing, how can it transfer that knowledge to a similar target drawing? Mathematically, this is often modeled as finding the maximum common subgraph between two sets of shapes and relations—a problem known to be NP-hard. Without heuristic guidance, a computer may waste significant resources attempting to map every trivial geometric feature.
The DSSBF Framework
To address this, the authors propose the Drawing Shape Structure Behavior Function (DSSBF) model. This framework extends the traditional Structure-Behavior-Function (SBF) model used in artificial intelligence by grounding it in geometric primitives.
As shown in the hierarchy above, the model consists of several layers:
- Function: The high-level purpose of the device.
- Behavior: The causal processes and movements.
- Structure: The physical components and their connections.
- Shapes and Spatial Relations: The geometric primitives (lines, circles) and their relative positions.
- Drawing: The actual visual representation.
By linking these layers, the system can identify "important" shapes based on their functional relevance. For instance, if a device's function involves a crankshaft, the circle representing that crankshaft is prioritized during the mapping process.
Analogical Mapping and the Archytas System
The researchers implemented this approach in a system called Archytas. The system uses the Bron-Kerbosch algorithm to identify maximal cliques—consistent sets of matches—between source and target drawings.
The mapping process is guided by functional importance. By focusing on shapes that are critical to the device's behavior, Archytas drastically prunes the search space. Once a mapping is established, the system "transfers" structural labels from the source model to the target drawing, effectively interpreting the new design by analogy.
Evaluation and Results
The approach was demonstrated using kinematics assemblies, such as piston and crankshaft systems. In these tests, the system successfully mapped a source drawing to a target drawing where components had moved (e.g., a piston moving to "top dead center").
Key results include:
- Search Space Reduction: Ranking shapes by functional importance allowed the system to focus its search, making the NP-hard graph matching problem computationally manageable.
- Structural Inference: Archytas successfully hypothesized the structural model of new drawings, identifying components and connections by analogy with known designs.
Conclusion
This research demonstrates that "function guides form" in the context of automated design interpretation. By treating interpretation as a mapping problem rather than a raw recognition task, and by using high-level knowledge to prune low-level geometric searches, the system achieves efficient and accurate interpretation of 2D engineering drawings. Future work will focus on expanding the system's capability to handle more complex geometries and refining the adaptation of models to significantly different target geometries.
Summary updated