{"id":90,"date":"2025-11-06T15:45:17","date_gmt":"2025-11-06T20:45:17","guid":{"rendered":"https:\/\/wp.stgeorges.bc.ca\/connork\/?p=90"},"modified":"2025-11-30T19:24:40","modified_gmt":"2025-12-01T00:24:40","slug":"neural-gear-network-in-onshape","status":"publish","type":"post","link":"https:\/\/wp.stgeorges.bc.ca\/connork\/2025\/11\/06\/neural-gear-network-in-onshape\/","title":{"rendered":"Neural Gear Network in Onshape"},"content":{"rendered":"\n

For this assignment, I took the learning-focused path instead of creating a full model from scratch. I wanted to understand the core workflow of CAD (how sketches, extrusions, assemblies, and drawings all connect) before I tried to make something ambitious. I followed the Onshape learning pathway and built a series of small, symbolic parts that relate to the structure of a transformer model. It became a way to understand both design principles and the way complex systems are built from simple, well-defined pieces.<\/p>\n\n\n\n

Sketching and Constraints<\/h4>\n\n\n\n

The first part I made was a simple 30\u00d730 mm square. It represented a single token<\/em>, the smallest building block in a transformer. I drew the square from the origin, added equal and horizontal\/vertical constraints, and made sure the lines turned black, meaning fully defined. This was my introduction to precision: the sketch only behaves predictably when every relationship is constrained.<\/p>\n\n\n\n

After finishing the sketch, I extruded the square by 4 mm and added a 1 mm fillet on the edges. That small step turned a flat, two-dimensional idea into something tangible. It felt like an \u201cembedding,\u201d a basic analogy for how data gets transformed from a sequence of numbers into something with depth and geometry.<\/p>\n\n\n\n

\n
\"\"
Token tile before extrusion<\/figcaption><\/figure>\n\n\n\n
\"\"
Token tile after extrusion and filleting<\/figcaption><\/figure>\n<\/figure>\n\n\n\n

Modeling the Attention Dial<\/h4>\n\n\n\n

Next, I learned to revolve and extrude circular shapes. I created a simple dial:<\/p>\n\n\n\n

A 24 mm diameter cylinder, 8 mm thick, with a 4 mm hole through its center. This part symbolized an attention head<\/em>. The hole became the pivot point for rotation, which I\u2019d later use in the assembly to experiment with motion.<\/p>\n\n\n\n

The process introduced me to more complex geometry, including concentric constraints and the use of reference dimensions to keep everything parametric. This part also made me appreciate how small design decisions (like leaving the center hole) make the assembly process smoother later on.<\/p>\n\n\n\n

\"\"
Attention dial with central bore.<\/em><\/figcaption><\/figure>\n\n\n\n

Assembly and Mate<\/h4>\n\n\n\n

Once I had a few small parts, I opened a new Assembly workspace. I added a flat base plate, fastened it to the world, and attached the dial using a revolute mate. That simple mate allowed the dial to spin freely while staying aligned with its axis.<\/p>\n\n\n\n

Seeing the dial rotate felt like the moment the whole process clicked. The revolute mate represents controlled movement; one degree of freedom. It perfectly mirrors the idea of an attention mechanism adjusting its focus. The difference between a fastened<\/em> mate (no movement) and a revolute<\/em> mate (single rotation) also helped me understand how constraints define motion in CAD.<\/p>\n\n\n\n