Heater Uniformity Masks: Quantify Registration and Edge Coverage

Measurement purpose

Evaluate geometric registration and finite-pixel coverage when transferring an approved heater acceptance region into a thermal image.

Specimens and conditions

Approved region

Retain drawing-defined boundaries, apertures and exclusions with current dimensions and datums.

Observable geometry

Identify stable registration features and preserve the heater's actual shape and viewing state.

Equipment and records required

• Thermal imaging: Provide raw numerical data, image validity and a characterized spatial response for boundary interpretation.
• Registration analysis: Save transformations, residuals and physical-to-detector mapping with independent check features.

Method sequence

1. Register

   Map the approved physical region using stable features and assess local residuals.

   Record: Transformation and check-feature results.

2. Classify coverage

   Separate required area, physical exclusions, invalid observations and partial boundary footprints.

   Record: Versioned masks and coverage areas.

3. Test sensitivity

   Recompute the fixed metric over justified registration perturbations.

   Record: Boundary-driven result interval and decision.

Decision and uncertainty

Do not remove a required edge to obtain a favorable result; unresolved geometric or radiometric coverage must remain in the acceptance limitation.

Registration, distortion, finite footprint and invalid boundary observations affect both included area and temperature interpretation.

The application owner approves the physical region; the metrology owner qualifies its transfer and measurement coverage.

Traceable outputs

Begin with an approved physical region, not an image crop

Use the application drawing to identify the region that already owns the uniformity requirement. Preserve holes, terminals and other exclusions with their geometric definitions. This page does not decide whether those exclusions are appropriate; it makes the approved boundary reproducible in the image.

Record the datum, units and drawing revision. A rectangular image crop is not equivalent to a shaped ceramic area with apertures. Keep a distinction between the required physical mask and the subset for which the current camera provides valid quantitative measurements. If the required area is not fully measurable, that gap must remain visible in the acceptance record.

Locate stable features without fitting to the temperature pattern

Identify physical corners, holes or other suitable fiducial features that can be located in the acquired view without altering the thermal function. Use their positions to establish the drawing-to-image transformation. Do not align images by shifting whichever region is hottest into the same location; that would make the measured outcome define its own registration.

Choose a transformation appropriate to the geometry and view. A planar heater observed obliquely may require more than translation and rotation. Lens distortion or a nonplanar assembly can leave local errors after a simple global fit. Preserve residuals at independent check features, especially near the consequential acceptance edges.

Count the geometric overlap of boundary pixels

A detector pixel's projected footprint can straddle the accepted region and an excluded hole or terminal. A center-in-mask rule, an any-overlap rule and an area-weighted rule include different boundary coverage. Choose the rule in advance and retain its definition with the metric.

Area weights describe how much projected physical area belongs to the mask; they do not undo optical mixing. A boundary pixel may receive radiation from both inside and outside the region through the optical response. Even perfect geometric weights cannot reconstruct two separate temperatures from one mixed observation without an additional validated model.

Estimate the area exposed to registration uncertainty

For a simple rectangular region of thirty by twenty millimeters, an inward uncertainty band of half a millimeter leaves a certain interior of twenty-nine by nineteen millimeters. The full area is 600 square millimeters and the interior is 551, leaving 49 square millimeters, about 8.17 percent, near the uncertain boundary. This example shows why a visually small coordinate uncertainty can affect a meaningful fraction of a small heater.

The uncertain band is not permission to remove that area from the acceptance requirement. It identifies where better registration or a bounded result is needed. Holes and multiple narrow features add boundary length and can increase the affected fraction. Use the actual geometry rather than applying the rectangular example to every product.

Recompute the same metric over plausible registrations

Use the estimated registration uncertainty to define a justified set of perturbed transformations. Reapply the unchanged physical mask and metric under those alternatives. Report the variation in the result and identify which edge locations drive it. This is a sensitivity analysis, not a search for the mask position that produces the smallest temperature spread.

For example, a fixed narrow hot feature beside an excluded terminal may enter or leave the accepted mask under a small shift. Preserve both classifications until the location is resolved. If the acceptance decision changes across plausible registrations, improve the geometric evidence or use the agreed uncertainty-aware decision rule rather than selecting one convenient alignment.

Separate physical exclusions from invalid measurement pixels

A hole excluded by the drawing, a saturated camera value and an unresolvable reflective surface are different categories. Store separate masks for required area, geometric exclusions and measurement validity. Combining them into one binary mask makes it impossible to tell whether an area was intentionally outside the requirement or simply not measured.

Do not fill invalid thermal pixels merely to complete a colorful image. Any interpolation used for display should remain separate from the data used in the acceptance calculation. A missing consequential edge requires another observation or an explicit limitation; its absence cannot count as evidence of an acceptable temperature.

Use a transparent boundary-decision table

The mask review should lead to a reproducible decision about geometry and coverage. It should not silently change the functional region or the temperature requirement when the image is difficult to analyze.

Release the mask as part of the measurement record

Save the physical polygon or equivalent drawing definition, transformation parameters, fiducial coordinates, residuals and image validity masks. Include the exact boundary-pixel rule, any geometric weights and the sensitivity results. Another reviewer should be able to reproduce which raw observations contributed to the reported statistic.

Requalify the mapping after a camera move, lens change, altered aperture or revised heater outline. A software filename is not sufficient version control when the physical mask has changed. The final report should distinguish the required area, the quantitatively observed area and the uncertain boundary so a professional acceptance decision remains tied to the actual product geometry.