Manufacturing & engineering

Deviation analysis

Definition

Deviation analysis measures and displays the distance between reference geometry and another curve, surface, or mesh. In scan-to-CAD work it is used to show where a rebuilt model sits above or below the measured scan and whether selected regions fall within declared bounds.

Also known as: deviation map, mesh deviation analysis, surface-to-mesh deviation

Updated

What can be compared#

SOLIDWORKS Design 2026 supports surface-to-mesh, curve-to-mesh, and mesh-to-mesh comparisons. A scan-to-CAD workflow can therefore check a fitted surface against the measured mesh, inspect a reconstructed section curve, or confirm that mesh cleanup has not moved the working data beyond a chosen limit.

The software samples distances and displays them as colours or spines. Upper and lower bounds control the displayed range, and a report can retain numerical results. The comparison still depends on the alignment between the two datasets and on which regions were selected.

Interpreting the colour map#

A deviation map needs a stated sign convention, bounds, units, coordinate system, and model revision. A mostly green plot means the sampled distances fall inside the configured central band. It does not prove that the band matches a product tolerance or that the underlying scan is accurate.

Review critical interfaces separately from cosmetic or freeform regions. A bearing seat, sealing face, or mounting pattern may need a tighter fit and an independent dimensional check. A styled outer surface may use wider bounds while still requiring surface continuity review.

What deviation cannot prove#

Deviation analysis compares two representations. It does not identify which one is nominal, recover design intent, validate scanner calibration, or account automatically for measurement uncertainty, wear, coatings, or deformation in the physical sample.

Use the result with the original point cloud, alignment record, fitted-feature assumptions, and applicable tolerance. The scan-to-CAD guide shows how that evidence supports an editable model without pretending one used part is a perfect master.

Sources

All glossary terms