On this page
A magnified ceramic circuit image can look impressive while still failing to distinguish the defect that matters. The first calculation is the object-space distance represented by a pixel. The final decision is whether the complete inspection setup can consistently separate acceptable and unacceptable features under the actual surface, lighting and positioning conditions. Those are different questions. Use the sampling calculation to screen candidate fields of view, then challenge the selected optical setup with relevant conductor gaps, overglaze edges or ceramic chips before assigning an inspection limit.
Key design decisions
- Specify the defect's width, length, material contrast and location rather than requesting a camera with more megapixels.
- Use the image dimensions actually recorded by the inspection workflow, including cropping, binning and exported resolution.
- Demonstrate visibility across the field and working-height range; an attractive central image is not sufficient.
Define the defect and the decision
A conductor bridge, an open trace, a glaze-window intrusion and an edge chip require different visual evidence. A bridge is judged by continuity across a gap; an open trace requires separation through the conductor; an edge chip is measured relative to a functional boundary. Stating only a minimum defect dimension loses these distinctions. Record which dimension determines rejection and which surrounding features establish its location.
Separate detection from dimensional measurement. An inspector may notice a contrast change without measuring its width reliably. Conversely, software may return a repeatable edge coordinate for the wrong material boundary. For each inspection task, specify whether the required output is presence, classification, width, remaining clearance or a position relative to a datum. The image acceptance challenge should test that exact output.
Calculate the scale at the ceramic surface
For a uniformly sampled image, divide the object-space field width by the number of active horizontal pixels. Calculate the vertical direction separately when the fields or pixel counts differ. The result is a sampling distance, not a lens-resolution guarantee. Lens blur, lighting, focus and contrast can prevent neighboring samples from carrying independently useful detail.
Use the recorded image size, not the camera's advertised maximum. Hardware binning, downsampling during transfer or a reduced-resolution export can change the effective sampling. A digital crop that retains native pixels changes the field of view and pixel count proportionally; it does not improve the original sampling. Enlarging the cropped image on a monitor makes its pixels larger without adding object detail.
sx = Fx / Nx; sy = Fy / Ny; px = d / sx
- Fx and Fy: measured horizontal and vertical fields of view at the specimen plane, in mm.
- Nx and Ny: corresponding recorded pixel counts, excluding display interpolation.
- sx and sy: object-space sampling distances, in mm per pixel.
- d: defect dimension along the horizontal measurement direction, in mm; px: number of sampled pixels spanning that dimension.
A calibrated planar specimen at the stated working height, with local distortion either sufficiently small or corrected. Pixel count screens sampling only; it does not certify defect detection.
Compare field-of-view choices with one defect
Assume a 24 mm horizontal field recorded across 2,448 active pixels. The scale is 24/2448 = 0.009804 mm per pixel, or approximately 9.80 micrometers per pixel. A hypothetical 40 micrometer gap spans about 4.08 pixels. If the field is widened to 48 mm without changing pixel count, the same gap spans about 2.04 pixels. That comparison exposes a sampling penalty, not an automatic pass or fail.
Suppose the inspection engineer selects six pixels across the gap as an initial development target. That is an explicit project assumption, not a universal optical rule. The corresponding maximum field is 0.040 × 2448/6 = 16.32 mm. A larger circuit could then require several images. Before accepting the added image count, verify that the lens and lighting retain useful contrast at the chosen scale and that stitching is unnecessary for the actual decision.
| Horizontal field | Recorded width | Object-space scale | Pixels across feature | Development implication |
|---|---|---|---|---|
| 48 mm | 2,448 px | 19.61 micrometers/px | 2.04 | Little sampling margin; perform a visibility challenge |
| 24 mm | 2,448 px | 9.80 micrometers/px | 4.08 | More sampling, still dependent on optics and contrast |
| 16.32 mm | 2,448 px | 6.67 micrometers/px | 6.00 | Meets the assumed sampling target, not a detection approval |
| 24 mm exported at half width | 1,224 px | 19.61 micrometers/px | 2.04 | Export has discarded the original sampling advantage |
Check material contrast before adding pixels
A dark resistor on light ceramic may be easy to distinguish, while a thin residue beside a reflective conductor can disappear under the same exposure. Translucent green overglaze can reveal underlying metal and create more than one apparent edge. Identify the physical boundary used for acceptance, then choose illumination that makes that boundary stable when the part moves through its permitted orientation range.
Increasing exposure can brighten the image but saturate a conductor, spreading its apparent boundary into an adjacent gap. Aggressive sharpening may create a false dark outline at the glaze edge. Retain a minimally processed inspection image and record the processing settings used for the decision. Compare edge positions under the approved lighting conditions rather than accepting whichever image looks the sharpest on a screen.
Challenge the corners and height range
Move the same known feature to the center, edges and corners of the field while keeping the inspection settings fixed. This separates field-dependent blur or distortion from variation between ceramic parts. Repeat at the allowed specimen heights, including fixture seating variation and the relevant printed-layer topography. A system focused on bare ceramic may not give the same edge response at a raised conductor or component.
Record which positions were actually challenged. If a critical pad is always near one corner, averaging performance across the whole field can hide its weakest condition. Where an optical correction is used, verify it with an independent feature rather than the same pattern used to fit the correction. A mathematically corrected coordinate does not restore contrast that the lens never resolved.
Include image capture and part seating
A static bench image does not establish performance during routine handling. For a moving specimen, exposure time and object speed determine how far an edge travels during capture. As a simple screening calculation, motion of 20 mm/s during a 0.5 ms exposure produces 0.010 mm of travel. At 9.80 micrometers per pixel, that is approximately one pixel of movement, potentially important for a narrow gap.
The calculation assumes uniform motion and does not model vibration or rolling-shutter geometry. Challenge the actual capture sequence after the fixture stops, including any settling delay. Repeatedly remove and reseat the same circuit to expose tilt and focus changes. If the method works only after manual adjustment for every sample, that adjustment is part of the method and must not be omitted from its expected inspection time.
Use a blinded acceptance challenge
Build a challenge set containing the actual defect types, positions and surface states near the acceptance boundary. Preserve a trustworthy independent characterization of each sample. Mix their order and avoid labels that reveal the expected answer. Evaluate missed defects and false rejections separately; a method that rejects every uncertain image can appear sensitive while being unsuitable for production disposition.
Include acceptable features that resemble defects, such as a harmless surface mark near an edge or a correctly opened overglaze window with variable reflectivity. The useful result is a decision record by defect class and condition, not one overall accuracy percentage hiding rare but important failures. If there are too few independently characterized defects, record the limited challenge coverage and do not extrapolate a detection probability to unseen conditions.
Retain the measurement chain and operating settings
For dimensional decisions, retain the scale calibration, reference artifact information, working height, lens settings and uncertainty evaluation relevant to the measured dimension. A calibration certificate for one artifact does not by itself validate the complete imaging method. Position repeatability, edge-selection behavior, specimen condition and correction residuals can still contribute to the final result.
The handoff should distinguish the calculated scale, the demonstrated measurement uncertainty and the tested defect-visibility envelope. These quantities answer different questions and should remain separately reviewable when a camera, lens, light, image export or fixture changes. Re-run the affected challenge after a material change; keeping the same nominal magnification is not enough to establish that the new image contains equivalent decision information.
Send the inspection feature and image requirements
Share the actual defect decision so the optical field and validation challenge can be matched to the circuit rather than to a camera specification alone.
- Circuit outline and critical conductor, resistor, glaze-window or edge dimensions with their acceptance boundaries.
- Unprocessed sample images, allowed specimen orientations, fixture height range and the recorded pixel dimensions.
- Defect types and independently characterized acceptable and unacceptable samples where available.
- Required output, permitted false-reject handling, dimensional uncertainty requirement and expected capture sequence.
The drawing-upload form loads as you reach this section.