A photograph can locate a suspicious joint opening quickly. Turning it into a dimension requires more information. Perspective changes the apparent scale across the image, and a reference ruler at a different depth may not share the gap’s magnification.
Before measuring, define which two steel-body edges bound the gap and the plane in which their separation is required. A dark line may be a shadow, coating contrast or an opening viewed obliquely. The image must reveal the intended boundaries.
Put the reference in the measurement plane
For an approximately planar joint face, place qualified scale references in that plane and include their known geometry. A planar perspective correction can map the image to that plane when sufficient independent reference points are available. It cannot correct hidden three-dimensional geometry from one view.
Lens distortion is a separate contribution. A perspective transform alone does not remove it. Use a qualified camera calibration or restrict the measurement to a region and configuration whose residual distortion has been assessed.
Check 1: Reference agreement
After correction, compare known reference distances not used to establish the transform. Residual disagreement indicates that calibration, reference positioning or the planar assumption is insufficient for the task.
Check 2: Edge visibility
Inspect both bounding edges at the measurement location. Glare, burrs and shadow can shift an apparent edge. Compare another lighting arrangement or view where appropriate, without altering the physical joint.
Check 3: Resolution capability
Let s be the local scale in millimetres per pixel, and let each edge coordinate carry uncertainty u pixels. Under a simplified independent-error model, the separation uncertainty from the two edges is approximately s√2u millimetres. Add calibration and geometric contributions separately. This estimate does not validate a particular edge algorithm.
If the apparent gap spans very few pixels, a visually plausible line can still have uncertainty comparable with its width. Enlarging the image on screen adds display pixels, not new measurement information.
State when the photograph cannot decide
Do not use the image for dimensional acceptance if its reference is at an unknown depth, its edge is obscured, its correction fails an independent distance check, or its uncertainty cannot resolve the requirement. Preserve it as a location record and obtain a suitable direct measurement.
For gaps that vary along a joint, record measurement positions and report the profile rather than selecting the widest-looking point without a consistent rule. Keep the original image and correction parameters so the result can be reproduced.
A photograph becomes useful dimensional evidence when the scale and geometry are tied to the actual gap. Otherwise its best contribution is identifying where to inspect, which is valuable without assigning an unjustified decimal measurement.
Resolve an optical gap beside a nominal-thickness sheet
For scale planning, 23ZH90 nominal sheet thickness is 0.23 mm [1]; it is not a calibrated image ruler. Assume a qualified in-plane image scale of 0.010 mm/pixel, a 10-pixel gap and independent edge standard uncertainties of 0.5 pixel each. The gap is 0.100 mm; u = 0.010√(0.5² + 0.5²) = 0.0071 mm and expanded edge-only uncertainty at k = 2 is 0.0141 mm [2]. The nominal sheet thickness would span 23 pixels. An assumed 0.110 mm maximum gap cannot be accepted from 0.100 + 0.0141 = 0.1141 mm; improve the measurement and include scale, perspective and distortion terms.
Manufacturing controls. Cut-edge straightness, burrs, overlap offsets and sheet seating set the physical gap. Separate a raised edge from the steel-body boundary; changing lighting can improve visibility but cannot correct an actual joint opening.
References
- Nippon Steel Corporation, Grain-Oriented Electrical Steel Sheets, catalogue D004je_03_202004, 2020, Specifications; Size and Form Tolerances.
- JCGM, JCGM 100:2008, Evaluation of measurement data — Guide to the expression of uncertainty in measurement, 2008.