CADEngBench: It Looks Like CAD, but Does It Work? Evaluating Parametric Design, Assembly Reasoning, and Physics Simulation
2026-08-10 • Artificial Intelligence
Artificial IntelligenceComputer Vision and Pattern RecognitionMachine LearningRobotics
AI summaryⓘ
The authors explain that just because a CAD model looks right doesn't mean it works correctly for engineering. They made a benchmark called CADEngBench with two tests: one checks if individual parts behave properly when changed or edited, and the other checks if assembled parts connect correctly. They found that editing existing CAD models is easier than creating new ones, but making complex changes and verifying how parts behave under forces is still hard. The study shows it's important to evaluate CAD models based on how they actually perform, not just how they look.
CAD modelparametric partboundary-representation (B-Rep)design for manufacturing (DFM)finite element analysis (FEA)joint retrievalkinematic verificationstructural responsefunctional editingassembly
Authors
Harmanjot Singh, Abhra Dubey, Jorge Alejandro Amador Herrera
Abstract
A CAD model is not engineering-grade merely because it looks correct. It must satisfy design requirements, respond predictably to parameter changes, support controlled edits, match a reference structural response under a declared analysis, and connect to other parts through valid joints. We present CADEngBench, a two-track benchmark for these capabilities. CADEngBench-P evaluates 300 parametric parts, each used for one zero-to-CAD task and one functional-editing task (600 tasks in total), through boundary-representation (B-Rep) validity, engineering and DFM checks, parameter-family perturbations, functional editing, and matched linear-static FEA in CalculiX. CADEngBench-A evaluates 150 body pairs through ranked joint retrieval, exact face-and-edge grounding, joint-frame prediction, and kinematic verification. Across eight multimodal, code-capable models, editing supplied CAD is substantially easier than generating it, while complex edits and matched FEA remain difficult. Assembly predictions often locate the relevant region but fail to recover the recorded joint or mating entities. These results show that CAD evaluation must test engineering behavior rather than appearance alone.