Gaps in a Clear Conformal Coating Show Up Under UV Fluorescence

LIGHTING

In reflected light a transparent coating cannot be told apart from bare board; shine 365 nm on it and keep only the light the film gives off, and the missing spots turn into black holes.

For a 5 MP camera viewing a 120 mm wide board, I calculated the blocking depth of a filter pair, the boundary that separates thin coating, and the width of the shadow beside tall parts.

About 15 min read

Put a PCB with conformal coating¹ under white light, and the image shows only green solder mask and silver solder. The coating is a thin, clear resin that passes light almost unchanged, and its surface is so smooth that it flashes only where the light catches it. That is why some lines still have an operator tilting each board under oblique light and following the edge of the film by eye.

Where the film is missing, the board lies open to moisture and ionic contamination, and after passing outgoing inspection it shows up later as leakage current or corrosion as temperature and humidity cycle. The opposite failure happens when the film spreads into an area that must stay bare, such as connector contacts, and causes contact faults. Both start from a film that is hard to see, so the spread between one inspector’s eye and another’s passes straight into the field failure rate.

The question comes down to one thing. How do you make a camera recognize where an invisible film is missing? My answer is to give up looking for the film by reflection and change the setup so it captures the light the film gives off by itself. The calculations assume a 5 MP mono camera with 2,448 × 2,048 px and 3.45 μm pixels, fitted with a 16 mm lens and viewing a board 120 mm across. One pixel then covers 49.0 μm, and the distance from the lens principal plane to the board, the WD², is about 243 mm.

Look for the coating by reflection and it disappears

One line of arithmetic shows why the coating vanishes under white light. At normal incidence, the boundary between air and a material with a refractive index of 1.5 sends back 4 % of the incoming light by Fresnel reflection³. Take the refractive index of both an acrylic conformal coating and the solder mask as about 1.5, and the top of a coated area and the top of bare solder mask both reflect the same 4 %, while the boundary where coating meets mask reflects almost nothing. The brightness difference between the two areas in the image is close to zero.

Tilt the board under oblique light and there is a moment when the film seems to appear. Where the edge of the film rises in a rounded lip, the reflection changes direction and a thin line flashes, but that line marks only the boundary of the film and says nothing about where a pinhole sits inside it. The rim of a hole glints for the same reason, yet it vanishes as soon as the angle drifts a little. A person keeps turning the board to collect these brief glints, which is a method that does not suit one camera fixed in place.

To separate two surfaces that reflectance cannot separate, you have to make them do different things. Many conformal coating materials contain a fluorescence⁴ tracer that glows under ultraviolet light, so only the places with film can be made to shine. From then on the decision is simply whether light is there or not, with no fine brightness differences to read.

The light the coating gives off has a different wavelength

Fluorescence is what happens when a molecule absorbs light of a short wavelength, loses part of the energy as heat and gives off light of a longer wavelength. The gap between the absorbed and emitted wavelengths is called the Stokes shift⁵, and thanks to this gap the camera can take in the light that was shone and the light the film made separately, by wavelength. It is the same principle that makes hidden patterns float up in a banknote under the violet lamp of a counterfeit detector. The figure below shows how far apart the two kinds of light sit on the wavelength axis.

In the figure, the narrow peak on the left is the UV-A excitation light near 365 nm, and the broad peak shifted to the right is the emission of the tracer. Assume the tracer glows in the blue band: coated solder mask then glows an even blue, while pinholes and skipped areas without film stay as black holes. In this setup, where one pixel is 49.0 μm, a 300 μm hole spans 6.12 pixels and a film that has spread 250 μm toward a connector spans 5.10 pixels. My standard is that a defect must span at least three pixels, so the pixel resolution⁶ here leaves room to spare.

The emission band and its strength vary with the coating material and the tracer concentration, so the blue assumption stays an assumption until a spectrometer measures the actual emission peak. Solder mask and component bodies also give off a weak autofluorescence⁷ of their own, so one image of the board taken before coating, under the same conditions, belongs in the setup as well. The excitation light cannot be managed in lx. The lx is weighted by the sensitivity curve of the human eye, so at 365 nm, which the eye barely sees, the number loses its meaning, and the light is measured and logged as irradiance⁸ in mW/cm².

With one long-pass filter the solder glows like coating

Fluorescence is far weaker than the light that excites it. Assume that the 365 nm reflection bouncing toward the lens from shiny surfaces such as solder fillets and metal pads is 1,000 times brighter than the coating fluorescence, and the problem becomes a number. Silicon sensors keep some sensitivity at 365 nm, so a long-pass filter⁹ that blocks the excitation light and passes only the emission has to sit in front of the lens.

In the earlier figure, the solid line on the right is the long-pass filter. If it is an OD¹⁰ 4 filter that cuts light at 365 nm to one ten-thousandth, the reflection comes down to 0.1 times the fluorescence; with OD 3, which cuts it to one thousandth, 1.0 times remains, as bright as the fluorescence itself. Once a bare spot next to solder looks as bright as a coated area, the missing-coating decision flips on the spot. When I choose this filter, I look at its blocking depth at 365 nm first and its passband second.

One filter is not the end of it, and the reason lies in the LED. The light of a UV LED is no single line at 365 nm; it trails a tail toward longer wavelengths, and when that tail reaches the passband of the long-pass filter, the reflected tail goes straight through. Assume 0.1 % of the LED output leaks above 420 nm: multiply the 1,000-times-brighter reflection by 0.001 and 1.0 times remains, so the solder fillet rises to the same brightness as the coating. The size of the tail can be guessed from the spectrum plot in the LED datasheet, but on a plot with an ordinary linear axis a tail at the 0.1 % level is buried in the baseline, so it has to be replotted on a log axis or measured directly. Fit an OD 3 short-pass clean-up filter¹¹ in front of the LED to cut the tail, and the remainder drops to 0.001 times, leaving a gap, like the dashed line on the left of the figure, where the filter on the light and the filter on the lens do not overlap.

Side by side, the three numbers show that the detection floor of this setup is held by the filter pair more than by the camera pixels. Metal surfaces mix diffuse and specular reflection, though, so the share of reflection that actually enters the lens differs from board to board, and the 1,000-times ratio used in the calculation cannot be guaranteed before a sample test. A software threshold cannot fill this gap for you. The reflection that leaked in has already been added into the same pixel value as the fluorescence, and there is no way to pull it out afterward.

Fluorescence tracks thickness only where the film is thin

After missing coating come the thinly coated areas. The fraction of excitation light a film absorbs follows 1 − e^−αd for absorption coefficient α and thickness d, and the fluorescence broadly follows this absorbed amount. For a thin film with small αd it is nearly proportional to thickness, but a thicker film saturates and the brightness barely moves as thickness changes.

Say the intensity ratio is computed per pixel against a properly coated board, with a ratio below 0.3 called missing and 0.3 to 0.7 called thin. How much real thickness these boundaries stand for depends on the αd of the normal film. Assume a normal film with αd of 0.5, and a ratio of 0.7 means 64 % of normal thickness while 0.3 means 25 %; an area at half thickness reads 0.56 and falls inside the thin decision. Apply the same boundaries to a film with αd of 2, and a ratio of 0.7 is reached only when thickness falls to 46 %, while an area thinned to 60 % of normal reads 0.81 and passes as good.

So before fixing the numbers 0.3 and 0.7, I would shoot a step sample coated in several thicknesses and check where the thickness-to-brightness curve flattens. If the curve flattens early for a given material, the honest choice is to use fluorescence only as an alarm for missing coating and hand the thinness decision to a thickness gauge. Fluorescence falls short as a thickness gauge, yet it fits well as an alarm that points to thin spots.

A weak signal also means the outside light has to be kept out. The long-pass filter passes everything above 420 nm, so the blue part of factory lighting comes in too and shakes the reference map. My standard is that an image taken with the UV off and the door closed averages no more than 5 % of the brightness of a coated area, with the residual illuminance at the sample plane targeted below 1 lx. Light leaking through a door gap shows up at once as the difference between an image with the door open and one with it closed, so checking the shielding with the camera is quicker than with a lux meter. The light shield is also a safety device that keeps UV-A off eyes and skin, so it gets a door interlock and a UV-blocking viewing window, and the photobiological safety classification of the source is checked against IEC 62471.

The area next to tall parts needs its own handling. Assume the ring light sits 230 mm above the board with a 50 mm radius: its light arrives tilted 12.3° from vertical, and a 10 mm tall part partly blocks the excitation light for 2.17 mm to its side, a band 44 px wide in the image. Only one side of the ring is blocked there, so the area does not go fully black, but the intensity ratio drops and it reads as thin, which is why a separate mask of that width is drawn from the part heights in the design data.

One page for the setup that catches a 300 μm hole

Move the blocking depth of the filters, the thinness boundaries and the shadow width onto one spec sheet, and it looks like the table below. With the same camera and lens, each line records how many pixels a defect spans and how the light, filters and decision are matched to it.

Item	Value	Condition and basis	In plain words
① Minimum defect size · missing film	300 μm diameter = 6.12 px (49.0 μm/px)	120 mm across ÷ 2,448 px, report blobs of 6 px (294 μm) or more	A pinhole must cover six pixels to be caught
① Minimum defect size · spread into keep-out area	250 μm wide = 5.10 px	keep-out areas such as connector contacts from design data	Spread down to five pixels wide is checked
① Minimum defect size · thin coating	1 mm × 1 mm = 20.4 px square	intensity ratio 0.3 to 0.7, 25 to 64 % thickness when normal film αd is 0.5	Thin areas are caught by their area
② Optical setup · excitation light	UV-A 365 nm LED ring + OD 3 short-pass clean-up	managed as irradiance in mW/cm², 0.1 % LED tail assumed	Cut the long wavelengths out of the light you shine
② Optical setup · detection filter	long-pass, passing from about 420 nm, OD 4 or more at 365 nm	leakage 0.1 times with reflection assumed 1,000 times the fluorescence	Block the bounced UV in front of the lens
② Optical setup · lens and WD	5 MP mono (3.45 μm), 16 mm, f/5.6, WD about 243 mm (from principal plane)	depth of field about 16.7 mm (2-pixel circle of confusion), height for ring light, filter holder and shield secured inside the WD	Measure room for the light between lens and board first
② Optical setup · shielding	UV-off image average below 5 % of coating brightness, residual illuminance under 1 lx	door interlock, UV-blocking window, IEC 62471 check	Keep out outside light and leaking UV together
③ Algorithm parameters · background subtraction	subtract the autofluorescence image of the uncoated board per pixel	removes autofluorescence of solder mask and component bodies	Erase what already glowed
③ Algorithm parameters · ratio decision	below 0.3 missing, 0.3 to 0.7 thin	fixed after the thickness-brightness curve is checked on a step sample	The sample sets the boundary
③ Algorithm parameters · shadow mask	2.17 mm (44 px) beside a 10 mm tall part	ring radius 50 mm and height 230 mm assumed, part heights from design data	Areas beside tall parts are judged separately

Read the table top to bottom and it shows that what sets the detection floor of this setup is the filter pair and the shielding more than the pixel resolution. Even the smallest defect spans more than five pixels, so there is little reason to raise the resolution; measuring the 365 nm blocking depth, the clean-up filter and the residual light with the door closed comes ahead of that. If those three are off, doubling the pixel count still leaves the solder glowing like coating.

Some boards clearly favor the opposite approach. If the coating material has no fluorescent tracer, there is no emission at all and this method does not stand; you go back to methods that work with reflection, such as low-angle oblique light or cross-polarization. If film thickness must be guaranteed as an absolute value in μm, fluorescence is a relative indicator and falls short, and interference or confocal thickness measurement is the right fit. Where the autofluorescence of the board or parts overlaps the emission band of the coating, the contrast stays low even after background subtraction, so changing the excitation wavelength may come out cheaper, and given diffuse reflection from solder surfaces and emission strength that varies by material lot, which way is better cannot be guaranteed before a sample test.

Two things to measure with a spectrometer before choosing filters

If I were building a new setup that inspects conformal coating with UV fluorescence, I would book two spectral measurements ahead of the camera spec. The first is the emission peak of a cured coating sample, to see whether the cut-on wavelength of the long-pass filter can sit far enough on the short side of that peak. The second is the share of the tail leaking above 420 nm from the LED lit without any filter; if it is larger than 0.1 %, the blocking depth of the clean-up filter goes up by the same factor. Once these two numbers are set, the filter pair and the gap between them are set, and the remaining parts are chosen to fit.

The person who holds these measurements and filter calculations is the machine vision engineer in charge of inspection lighting on site. For someone just learning the field, the quickest way to see what a long-pass filter does is to shine a UV flashlight on paper marked with a highlighter and, keeping the flashlight pointed away from your eyes, hold a sheet of yellow cellophane in front of them. From the side that spends the equipment budget, the quote lines that decide the detection floor are the two filters and the light shield more than the camera pixels, so it pays to read the quote in that order too. This article explains technology and is not investment advice. See our Disclaimer.

Field Note

If solder fillets glow as bright as the coated areas in the fluorescence image, I suspect the LED tail first. For this setup, I would take one image with only the long-pass filter and one with the clean-up filter added, at the same exposure with the UV on, and see whether the fillet brightness drops from near 1.0 times the coating to near 0.001 times. If it does not drop, the 365 nm blocking depth falls short rather than the tail, so the OD of the long-pass goes from 3 to 4. In the same session I also check that the image with the door closed and the UV off averages no more than 5 % of the coating brightness.

Field Checkpoints

If the two images in the Field Note match the calculation, use the questions below to turn the setup into spec-sheet sentences.

• Is there height left inside the WD of about 243 mm (16 mm lens, from the principal plane) for the UV ring light, filter holder and light shield? (check the distance from the lens front on the lens drawing)
• Have you separated the surface materials and their reflectance first? (solder mask, solder fillets and metal pads reflect differently)
• Have you measured with a spectrometer whether the conformal coating material contains a fluorescent tracer and where its emission peak lies?
• Is the long-pass filter OD 4 or more at 365 nm, with a non-overlapping gap between it and the clean-up filter in front of the LED? (so the solder does not glow like coating)
• Does the image with the UV off and the door closed average below 5 % of the coating brightness? (target residual illuminance under 1 lx)
• Have you set the 0.3 and 0.7 boundaries from the thickness-to-brightness curve of a step sample? (if the curve flattens early, leave the thinness decision to another instrument)
• Are a door interlock and a UV-blocking viewing window in place, and has the photobiological safety classification of the source been checked against IEC 62471?

Frequently Asked Questions

Wouldn’t a color camera show the color of the fluorescence as well?

It would, at the cost of light. A color sensor has a color filter on every pixel, so blue fluorescence is properly received only by the blue pixels, one in four, and the other pixels throw most of that light away. With weak fluorescence, a mono camera with a long-pass filter gives a brighter image at the same exposure and keeps its full resolution. On boards where color is needed to tell autofluorescence from coating, color wins, and the way to separate color differences with a single wavelength is calculated separately in mono cameras and wavelength selection.

Do UV LEDs also get dimmer over time?

They do. The fluorescence follows the strength of the excitation light, so as a UV LED ages the brightness of normal film drops with it and the intensity-ratio baseline shifts. My approach would be to keep one properly coated sample as a reference, image it at each shift change, and correct the decision by its ratio to the value recorded when the reference map was made. If that ratio falls below 0.9, the LED gets replaced or its output raised, and the reference map is made again.

Glossary

• [1] Conformal coating conformal coating
  A thin protective resin film applied over a populated board. It keeps moisture and dust from reaching the circuit directly. ↩
• [2] WD working distance
  The distance between the lens and the inspected surface. The numbers here are thin-lens values measured from the lens principal plane, so the actual clearance from the lens front is checked again on the lens drawing. ↩
• [3] Fresnel reflection Fresnel reflection
  The reflection that sends part of the light back at the boundary between two materials with different refractive indices. Between air and glass, a little over 4 % of light at normal incidence comes back. ↩
• [4] Fluorescence fluorescence
  A molecule absorbs light of a short wavelength and promptly gives off light of a longer wavelength. Switch off the excitation and the glow stops almost at once. ↩
• [5] Stokes shift Stokes shift
  The wavelength gap between absorbed and emitted light. The wider the gap, the easier it is to separate the two with filters. ↩
• [6] Pixel resolution pixel resolution
  The length one camera pixel covers on the object. It is the field width divided by the horizontal pixel count. ↩
• [7] Autofluorescence autofluorescence
  The weak fluorescence a material gives off on its own, without any added tracer. Solder mask and component bodies show a little of it too. ↩
• [8] Irradiance irradiance
  The light energy reaching a surface per unit area, in units such as mW/cm². It measures the light itself, regardless of the sensitivity of the human eye. ↩
• [9] Long-pass filter long-pass filter
  A filter that passes only light longer than a set wavelength. In fluorescence inspection it sits in front of the lens and blocks the UV that was shone on the board. ↩
• [10] OD optical density
  The blocking depth of a filter written as a power of ten. OD 3 passes one thousandth of the light, OD 4 one ten-thousandth. ↩
• [11] Short-pass clean-up filter short-pass clean-up filter
  A filter placed in front of the light that cuts off wavelengths longer than a set limit. It keeps the tail of the source spectrum from getting through the detection filter. ↩

References

• Formulas: pixel resolution = field of view ÷ pixel count, magnification m = pixel size ÷ pixel resolution, object distance = f × (1 + 1/m), depth of field ≈ 2 × N × c × (m + 1) ÷ m², reflectance at normal incidence = ((n − 1) ÷ (n + 1))², absorbed fraction = 1 − e^−αd, filter leakage = reflected intensity × 10^−OD, shadow width = part height × (ring radius ÷ ring height)
• Assumptions: 5 MP mono camera (2,448 × 2,048 px, 3.45 μm), 16 mm lens, f/5.6, 120 mm horizontal field, refractive index 1.5, 365 nm reflection returning to the lens at 1,000 times the fluorescence, 0.1 % LED tail, normal film αd of 0.5 and 2, ring radius 50 mm and height 230 mm, part height 10 mm, circle of confusion of 2 pixels (6.9 μm)
• IEC 62471:2006, Photobiological safety of lamps and lamp systems, IEC, 2006-07-26. webstore.iec.ch/en/publication/7076

Related reading — An 850 nm Band-Pass Filter Erases Factory Lights but Leaves Sunlight · A Mono Camera Catches PCB Color Differences When You Choose the Wavelength · How Strongly Cross-Polarization Suppresses Glare Depends on the Lighting Angle