A user may click a single command to create a fillet, subtract one body from another, or extrude a sketch, but the operation performed underneath is rarely simple. A CAD kernel has to interpret geometric input, calculate new curves and surfaces, modify topology, resolve intersections, and verify that the resulting model is still valid. For developers of engineering software, these internal steps explain both the capabilities and the limitations of solid modeling systems.
The Kernel Starts with Mathematical Geometry
At its lowest level, a modeling system works with mathematical entities rather than rendered objects.
Points define positions. Curves may represent lines, circles, ellipses, or splines. Surfaces can include planes, cylinders, cones, tori, and freeform parametric forms. These entities provide exact or parametrically defined geometry that can be evaluated, intersected, trimmed, and transformed.
A geometric kernel supplies algorithms for working with these objects. A geometric modeling kernel extends this responsibility into the construction and modification of complete engineering models, where geometry must remain connected through a valid topological structure.
This distinction becomes especially important in solid modeling.
Topology Gives Geometry Structure
A collection of surfaces is not automatically a solid.
In a B-Rep model, topology records how geometric entities are connected. Vertices define topological points, edges connect vertices and usually refer to underlying curves, while faces reference surfaces bounded by loops of edges. Faces can form shells, and closed shells can define solid bodies.
The topology therefore describes relationships that pure geometry does not contain.
This structure is what allows a CAD application to identify a particular face, select an edge for filleting, or determine which regions belong to the inside of a body.
What Happens During a Boolean Subtraction
Consider a common operation: subtracting a cylinder from a block to create a hole.
The kernel first determines where the surfaces of the two bodies intersect. These intersection calculations generate curves that divide existing faces into separate regions.
The next step is classification. The system determines which portions of the original bodies lie inside or outside the other body. Regions that should disappear are discarded, while the remaining pieces are retained.
Topology must then be rebuilt. New edges, loops, and faces are created from the surviving geometry, and the resulting shell is checked to determine whether it represents a consistent solid.
A command that appears instantaneous in the interface may therefore involve surface-surface intersection, curve construction, trimming, classification, and topological reconstruction.
Fillets Create New Geometry
Filleting exposes a different internal process.
When an edge is rounded, the original neighboring faces cannot simply be moved. The kernel has to construct a transition surface that satisfies the required radius and geometric conditions.
That surface must be intersected with adjacent faces, trimmed to the correct boundaries, and inserted into the model. The original topology near the selected edge is replaced with new faces and edges.
Situations where several fillets meet at a corner are harder because the transition surfaces interact with one another. The kernel has to determine a geometrically consistent configuration while avoiding invalid or self-intersecting results.
Tolerances Affect Nearly Every Operation
CAD geometry is computed using finite numerical precision. Values that should theoretically be identical may differ slightly after transformations, imports, or repeated calculations.
For this reason, a CAD kernel uses tolerances when deciding whether points coincide, curves intersect, edges connect, or surfaces can be treated as adjacent.
These decisions are not minor implementation details. A gap between two faces may prevent a shell from becoming a valid solid. Nearly tangent surfaces can make intersection calculations difficult. Very short edges or extremely small features can create unstable topology.
Tolerance handling is therefore deeply connected to model validity.
From Kernel Results to Application Features
A CAD application usually accesses the modeling layer through an API or SDK. Higher-level application logic determines what the user or engineering workflow requires, while the kernel performs the corresponding geometric operations.
The resulting B-Rep or other model representation can then be used by visualization, CAM, CAE preprocessing, measurement, data exchange, or further editing tools.
This division of responsibilities is fundamental to CAD application development. The interface may expose commands such as extrude, shell, cut, blend, or offset, but the kernel is responsible for turning those commands into mathematically and topologically consistent 3D models.
Inside a CAD kernel, geometric modeling is therefore less about drawing shapes and more about preserving relationships. Every successful modeling operation has to produce geometry that not only looks correct, but can continue participating in future engineering calculations.