Study notes for Ian Stroud, Boundary Representation Modelling Techniques (Springer, 2006).

Feature modelling is where CAD becomes design intent rather than just geometry editing. This chapter explains both the data structures for features and the workflow for feature-based creation and recognition.

1) Feature datastructures

The chapter introduces structures that bind “a feature” to parts of the B-Rep:

  • feature facesets: groups of faces/edges that constitute the feature’s geometric footprint
  • feature frames: reference frames / datum context used to define the feature parameters

In modern terms, this is the bridge from:

  • parametric definitions → generated boundary model
  • boundary model → recognized features

2) Manual feature acquisition

Before automated recognition, a system must support:

  • selecting faces/edges that represent a feature
  • validating the selection against feature type constraints
  • generating feature parameters from geometry (inference)

This is critical for edit workflows (redefine a feature from geometry) and for imported models.

3) Design by features

Feature-based creation becomes a pipeline:

  • create a feature definition (sketch, dimensions, references, parameters)
  • generate geometry
  • apply a modelling operator (e.g., cut, boss, fillet)
  • store mapping and ownership so the feature can be edited later

4) Feature recognition and verification

Feature recognition is treated as:

  • identify candidate regions in the B-Rep
  • classify shapes/topologies that match known patterns
  • verify against geometric constraints and tolerances
  • attach a feature object back onto the model

Recognition is never perfect; robust verification is essential.

Chapter outline (from the book)

Major sections

  • 9.1 Feature Datastructures
  • 9.3 Design By Features
  • 9.4 Feature Recognition
  • 9.6 Structuring Features

Selected subsections

  • 9.1 Feature datastructures
  • 9.1.1 Feature facesets
  • 9.1.2 Feature frames
  • 9.2 Manual feature acquisition
  • 9.3 Design by features
  • 9.3.1 Adding features to a shape
  • 9.3.2 Building objects from isolated features
  • 9.4 Feature recognition
  • 9.4.1 Classic feature recognition

Implementation takeaways

  • Design a feature object model that can:
    • own topology subsets
    • regenerate geometry/topology from parameters
    • rebuild ownership after editing (mapping)
  • Build recognition in layers: 1) simple primitives (holes, pockets, bosses) 2) blends (fillets/chamfers) 3) interacting features (patterns, multi-step)
  • Keep feature verification separate from recognition heuristics:
    • recognition suggests
    • verification proves (or rejects)

Feature modelling: three contracts

A robust feature system must satisfy:

  1. Creation contract: parameters + references → geometry/topology.
  2. Edit contract: change parameters → regen, with stable mappings.
  3. Query contract: geometry/topology → recover feature meaning (recognition/inference).

Most systems nail (1) but struggle with (2) and (3). The chapter highlights why (2)/(3) need explicit datastructures.

Practical exercises

  • Define a minimal feature object with:
    • definition parameters
    • references (datum/edges/faces)
    • owned topology set
    • mapping after regeneration