A local core scan can be electronically quiet yet poorly repeatable. Near a joint or edge, a small change in probe height, angle or trajectory can alter the measured response. If that variation is not quantified, a later scan may appear to show a changed defect when only the observation geometry changed.
Separate three comparisons: repeated acquisition without moving the sensor, repeated positioning at one location, and repeated traversal of the full track. Each reveals a different part of the measurement variability.
Establish recoverable datums
Select physical references that survive the intended inspection sequence and can be accessed again. Record the track relative to those references. A photograph of the probe in approximately the same place is insufficient when the signal changes steeply with standoff.
Preserve sensor orientation, sensing volume and the registration method. For thermal imaging, include view geometry and surface condition. For electromagnetic scanning, include excitation and phase reference. The trajectory is only one part of the repeated state.
Fixed-position repeat
Estimate acquisition variation with the sensor stationary and the relevant source stable. Retain raw traces or frames. This describes noise and drift under one geometry; it does not include repositioning.
Repositioning repeat
Remove and replace the sensor or setup through the intended operational procedure. Compare results at the same datum. This captures mounting, height and orientation variability that the fixed-position test excludes.
Track perturbation
Make controlled, documented changes in height or lateral path on a suitable reference region. Determine how those changes affect the local pattern. The purpose is to identify sensitivity to realistic trajectory errors, not to select whichever path produces the clearest indication.
Report uncertainty at the feature
A single global repeatability number may hide a region with particularly steep sensitivity. Map repeat variability near the feature being interpreted. Compare that variability with the difference between reference and suspect states.
If confidence intervals or detection probabilities are reported, state the data, model and sampling assumptions used to obtain them. Do not attach a percentage to a handful of visually similar scans. Measurement uncertainty and classification performance are related but distinct quantities.
Keep repeatability from becoming false accuracy
Repeated scans can reproduce the same systematic bias, reflection or background pickup. Include known reference comparisons and capability checks as well as repeated trajectories. A stable image can still represent the wrong physical quantity.
The inspection record should show fixed-position noise, remounting variation and track sensitivity separately. It can then justify whether an observed local difference exceeds the method’s repeatability limit. If not, the correct response is to improve registration or select another observation, rather than label a small scan difference as defect growth or successful repair.
Set the repeat-position budget from a lamination scale
For a 0.23 mm nominal 23ZH90 lamination [1], assume repositioning uncertainty equal to one sheet thickness, a local magnetic-signal gradient of 2.0 mV/mm, and fixed-position standard noise of 0.10 mV. Position contributes 0.46 mV; independent combined standard uncertainty is √(0.46² + 0.10²) = 0.471 mV [2]. An assumed 0.30 mV difference is unresolved on that basis. Improving repositioning to 0.05 mm reduces the combined value to 0.141 mV, making the same difference worth further controlled comparison. It does not establish a fault by itself; excitation drift and background bias remain separate terms.
Manufacturing controls. Joint-edge position, packet steps, duct access and restraint define the spatial gradients seen by a probe. Stable probe stops must not change the sheet contact or gap being investigated.
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.