CAD-to-VR framework for interactive assembly simulation in product design review

Industrial quality inspection encompasses tasks ranging from metrological verification and surface-defect detection to feature checking and assembly conformity assessment. The last of these still often depends on operators interpreting technical drawings and following checklist-based procedures, which can make the process slow, variable, and difficult to trace. This paper describes a CAD-driven augmented reality (AR) workflow for component-level conformity checking of industrial assemblies. CAD/MBD product data are converted into inspection content, component-specific regions of interest (ROIs), and runtime references for shop-floor use. A commercial AR tracker aligns the CAD model with the physical assembly. Custom Unity and OpenCV modules then isolate the selected component and compare CAD-derived and observed edge maps within the same ROI. Small residual registration errors are accommodated by a Gaussian-weighted distance-tolerant score, while an orientation-consistency cue helps interpret ambiguous cases. Each check generates a structured record containing component identity, similarity indicators, decision state, and traceability information. Product structure, operator-guided checking, visual evidence, and quality records thus remain connected throughout the workflow. In Quality 4.0 terms, the system provides a Detect and evidence-capture layer for missing, inconsistent, or uncertain component-level evidence. The evaluation covers 13 industrial assemblies and 927 component-level instances, using transparent CAD-to-image baselines, ROI ablation, degenerate-case and sensitivity analyses, and a controlled missing-component pilot. The results show that CAD-driven masking reduces off-component edge contamination, while conservative decision states identify weak or fragmented evidence that requires reacquisition or operator verification.