3D Model of the Month - Deviation Target: Optimizing CAD Models While Staying Within Tolerance
Table of Contents
Introduction
While CAD models are a great source for digital twins and product visualization, tessellation can often result in meshes with a very high face count, making them too demanding for many real-time applications.
In this Model of the Month, we will show how RapidPipeline's Decimate Action and Deviation Target settings work together to significantly reduce the face count while keeping the optimized mesh within a defined geometric tolerance. This makes it possible to create lightweight assets without exceeding the acceptable deviation from the original CAD model.
Create 3D digital twins for CGI & XR, faster than ever.
Turn your raw 3D or CAD into realistic digital twins.
Why does a Deviation Target matter?
Engineers regularly work with dimensional tolerances to define how much a manufactured part may deviate from its geometry while still meeting design requirements. The deviation target in RapidPipeline follows the same principle for mesh optimization.
Instead of reducing polygons blindly, the decimation process ensures that the generated mesh never exceeds the specified geometric deviation from the original model. This means the Deviation Target will automatically check that your output still represents the original CAD data within the required tolerance and re-run adjusting face targets if the changes were too aggressive.
For CAD users, this provides a predictable optimization process: rather than choosing an arbitrary face count and hoping important features remain intact, you define the maximum acceptable deviation and let the optimizer work within that limit.

Importing & Optimizing the model
For this example, we used a .3dm Camera model. Using the RapidPipeline plugin, we can tessellate and import it into Blender. The resulting tessellated mesh has about 625K polygons.

CAD Input Tessellated Result
To make the model more suitable for real time applications, the model’s face count needs be reduced, while preserving the details that matter visually.
To achieve this, we used RapidPipeline's Decimate Mesh action with a defined Deviation Target. The deviation target defines the maximum geometric deviation allowed between the optimized mesh and the original CAD geometry. Similar to a manufacturing tolerance, it sets an upper limit on how much the simplified model may differ from the source. In this example, a deviation target of 0.01 % was used. (*Note: The percetage refers to largest bbox edge length).
Different parts of the model were optimized using individual face targets while maintaining the same deviation target throughout the process. This resulted in a face reduction from approximately 625,000 polygons to just 160,000 polygons, while preserving the overall visual fidelity.

Adding Materials
With the geometry optimized, the model can now be further prepared for real-time applications. Using Blender's material system, materials were generated, applied and baked to achieve the model’s final appearance.

Create 3D digital twins for CGI & XR, faster than ever.
Turn your raw 3D or CAD into realistic digital twins.
Export
Once all the materials are ready, the model can now be exported. RapidPipeline’s export settings provide additional options to further optimize the model such as customizable material formats and geometry compression.
The result is now a 15MB asset with 160K faces, created form a 60MB high detail CAD file, while remaining within the specified deviation target. This provides a lightweight model for real-time applications without compromising the geometric accuracy defined by the original CAD tolerance.
You can explore the final version of the model using our interactive viewer below.
The model can also be viewed in AR for an even closer look.

Meet the Author

Daniel
Technical 3D Artist
Daniel is a 3D and VR artist and responsible for QA at DGG. Working with leading retailers, DGG is on a mission to automate 3D asset optimization workflows for Web, mobile and XR targets - for e-commerce, and beyond. Daniel completed his studies in Expanded Realities at the Hochschule Darmstadt.
You might be interested in


