Parametric CAD: Modelling for the Change That Is Coming

The real deliverable

Every model gets changed. Requirements move, analysis returns an unwelcome result, manufacturing asks for a draft angle. The question is not whether the model will change but whether it will survive the change.

A brittle model is one where altering a dimension breaks fifteen downstream features and takes an afternoon to repair. A robust model absorbs the change and rebuilds. The difference is almost entirely decided in the first hour of modelling, before any complex geometry exists.

Design intent is the thing being encoded

The feature tree records not just the shape but the reasoning. A hole positioned 20 mm from an edge is a different statement from a hole positioned at the centre of a face — they may coincide today, and they diverge the moment the part gets wider.

Before creating a feature, ask what it is dimensioned from, and whether that relationship should survive a size change. Getting this right is what allows a design to be driven from a small set of meaningful parameters rather than edited everywhere by hand.

Reference geometry over model geometry

The most common cause of fragile models is sketching on faces created by earlier features. When those faces move, split or vanish, everything built on them fails — the notorious rebuild error cascade.

Anchor sketches to datum planes and axes established near the top of the tree instead. Datums do not disappear when a fillet changes. This single habit prevents more rebuild failures than any other.

Related: avoid selecting edges and faces for fillets and chamfers where a robust alternative exists, because those references are the ones most likely to be invalidated by upstream edits.

Structure the tree deliberately

A tree of ninety features in creation order is unreadable. Structure it:

• Datums and skeleton first — the geometric framework everything references
• Primary shape — the base solid that establishes the envelope
• Functional features — holes, bosses, pockets that carry design intent
• Cosmetic and manufacturing last — fillets, chamfers, draft

Placing fillets late matters practically as well as tidily: they are the features most likely to fail on rebuild, and they interfere with downstream selections when placed early. It is also far easier to suppress a block of late fillets to produce a simplified model for analysis.

Name features meaningfully. Cut-Extrude17 tells the next engineer nothing; Bolt_Clearance_Pattern tells them what it is for and whether they may change it.

Constrain sketches fully

An under-constrained sketch will move unpredictably when something upstream changes. Fully define every sketch — most systems indicate this — and prefer geometric relations such as coincident, tangent and symmetric over dimensions where the relationship is genuinely geometric.

Symmetric parts should be driven from a centreline with a mirror, not modelled twice. Two independent halves will eventually disagree.

Analysis needs a different model

The model that manufactures is rarely the model that analyses well. Small fillets, cosmetic text and fastener threads force enormous element counts for no useful return.

Plan for this by building a configuration or simplified variant with cosmetic detail suppressed, rather than deleting features and losing them. Simplify deliberately and record what was removed — a defeatured model that quietly dropped a stiffening rib produces a confident, wrong analysis.

Assemblies

Constrain components to datums and to a skeleton, not to each other in a long chain. A chain of part-to-part mates means moving one component shifts everything downstream of it, and a circular reference is only one careless mate away.

A skeleton model — a sketch or surface body holding the key interfaces and dimensions, referenced by every part — keeps the assembly consistent and gives one place to change a critical dimension.

The test

Take a finished model and change its principal dimension by thirty percent. If it rebuilds cleanly and still expresses the design intent, it was built well. If it fails, the model documented a shape when it should have documented a design.