Thermal Imagers for PCB testing

Not sure offhand what's best, but I will offer this: the simplest / cheapest variants won't cut it for our kind of work. With fixed focus and

160x100 (or lower!) resolution, the PSF at any distance is just too blobby to resolve chip components.

Those are all sub-$1k products. So, there should be good offerings in the $1-6k range.

Tim

-- Seven Transistor Labs Electrical Engineering Consultation Website:

The germanium lens on our FLIR costs a couple of kilobucks.

I'd make sure to try out a loaner before you buy. Cheap imagers are good for tracing pipes in the wall but likely won't have the close-up resolution for imaging electronics.

I'd expect that there is some decent stuff in the $6K range these days.

This is what the $12K FLIR E45 can do:

--

John Larkin         Highland Technology, Inc 
picosecond timing   precision measurement  

jlarkin att highlandtechnology dott com 
http://www.highlandtechnology.com

Germanium is a really nice optical material at 8-12 micron, but the stuff itself is costly, and it has such a high index of refraction that it needs some expensive anti-reflective coatings. So it's not like you're being gouged or anything.

Do you know if your imager is made in Sweden or in the US? I know that for a while they were all being sourced out of Sweden, but FLIR bought some US companies that could do those thingies about the time that I was leaving for real.

--

Tim Wescott 
Wescott Design Services 
http://www.wescottdesign.com

Yeah, the lens looks like a purplish mirror-finish golf ball.

I know that the software was done in Sweden, and it's terrible.

It's USB but enumerates as a network device, not a memory stick or camera. It has a weird USB cable. It only works in the USB port that it was installed for... you have to always plug it into the same connector or hub port. That caused a lot of confusion.

But the optics is great.

--

John Larkin         Highland Technology, Inc 
picosecond timing   precision measurement  

jlarkin att highlandtechnology dott com 
http://www.highlandtechnology.com

Ideas:

1. I wonder if there are any point source bolometers or thermal detectors that can focus down to a small diameter. Put one on a scanning x-y table and beat a k or k FLIR for a hundred bucks or so. Maybe a bit slower, but does that matter?

1. How about a tiny thermistor on a x-y-z table and arrange to keep it 10 mils or so above the surface. Insulate the leads so convection currents don't mess up the reading.

2. Who needs a germanium lens. Get a pseudo-parabolic reflector that focuses to a small spot and put the thermstor at the focal point.

1. DIY X-Y tables are very cheap these days. It doesn't need 1 mil accuracy. Probably 10 mils would be fine. It could also be used for other applications, such as milling copper traces for prototype pcbs. I already want one for that reason, adding a cheap FLIR scanner just makes it more interesting.

It's called an elliptical reflector. Here is a nice demo that illustrates how the spot changes as the focus is moved:

Just run it backwards and put the thermistor at the focal point.

FLIRs use MEMS microbolometers, which are air-insulated. Thermal conduction slows down dramatically with increasing size--any vaguely macroscopic device coupled only by radiation will have a response time of at least a second or two, and that's being generous. Multiply that by 100k points per image, and your frame time is measured in days.

It isn't convective transport to the leads that's the problem, it's the leads themselves. Radiative transport near room temperature is very weak--it's comparable to conduction through a few millimetres of air. Copper is roughly 15,000 times as thermally conductive as air, so it doesn't take much lead cross-section to dominate.

Doesn't work well, because the image falls apart very rapidly as you move off axis. To get any signal, you need a large numerical aperture (i.e. wide cone angle) at the detector side. To get any field of view at close range, you need a fairly wide field angle (i.e. the angle the axis of the light cone makes with the axis of the lens).

Lenses, especially very high index ones like germanium, handle this combination much, much better than mirrors. Cheaper IR cameras use ZnSe lenses, but they still don't grow on trees.

Cheers

Phil Hobbs

--
Dr Philip C D Hobbs 
Principal Consultant 
ElectroOptical Innovations LLC 
Optics, Electro-optics, Photonics, Analog Electronics 

160 North State Road #203 
Briarcliff Manor NY 10510 

hobbs at electrooptical dot net 
http://electrooptical.net

I'm just starting to read about sensors, there are zillions of kinds available. Some have response times of a millisecond or less.

I was thinking of a minimum spot size of 10 mils, or 100 points per inch. Using a scan of 5 inches gives 500 points per line. With 500 lines, we would have 250k points.

But the scan does not have to be start-stop. The temperature gradients will probably be fairly smooth, so the sensor will not see step changes in temperature, but rather more gradual changes. So all we need to do is match the scan rate to the maximum temperature slope and scan slower than that. Very much like adjusting the smoothing on a sampling scope. This could drastically increase the scan rate over start-stop operation.

The x-y scan does not have to return to the starting side like the old tv raster scan. It can simply increment the y dimension then reverse direction and scan the next line in the opposite direction from the previous scan. This means the backlash in the x-y table has to be kept fairly low, but it should be possible to accomplish this at reasonable cost. For example, a cheap digital caliper could be used to measure the position in each direction and correct for backlash and nonlinearity errors.

Thanks. You have mentioned the problem with leads previously, but I believe this is the first time you mentioned hard numbers. That is very useful information.

Some of the sensors I have read about mount the sensing element on standoffs and run in a hard vacuuum. I don't know the cost, how hard it would be to couple into them, or how much the glass would attenuate the long wavelengths at room temperature. But there is a wealth of information available, and many different kinds of sensing elements, so there might be some combination that would serve the purpose. If nothing else, it might be worthwhile building a system just to explore the problems and see if it is worth investing more time and effort.

I would think we want the image to fall apart rapidly as we move off axis. We want to measure the temperature at a small spot, not over a large area.

I was thinking of examining the element in an ordinary IR thermometer to see how the thermal energy is coupled to the element, and how the ambient temperature is measured to eliminate drift.

The IR thermometers can have a ratio of distance to spot diameter of 20 to several humdred or so. A large ratio means little energy is reaching the sensor, but it is still capable of giving accurate temperature measurements. This could significantly reduce the requirements on the reflector, as well as give a larger standoff distance.

One of the big problems I think about is how to keep the sensor in focus. The pcb will have components with vertical edges, such as a heat sink or TO-220 transistor. It will go out of focus when we hit the edge. We can use knowledge of the pcb to try to figure out which direction to go to regain focus. For example, if we are at the surface of the pcb and suddenly lose focus, most likely the object is higher than the surface of the pcb so we would search in the positive z axis.

We could also use the 3D projection of the pcb with components such as generated from Eagle to locate the components and estimate how far to go in the z direction to regain focus.

Actually determining when the system is in focus is another major problem. A conventional IR thermometer does not need focus control - it's only when the sensor is coupled to a lens. So this needs more study.

Yes. That is why it may be interesting to explore elliptical reflectors. I understand they can be coated to reduce the emissivity, perhaps with gold or some less expensive material.

Thank you for your time and excellent suggestions.

That's irrelevant. You won't get independent measurements unless the dwell time per pixel is a few times the thermal time constant of the sensor. You can speed that up afterwards if your sensor is well enough insulated (see

, specifically the paper at )

Nope. Overheating components will stick out like the proverbial sore thumb.

I'm not sure what you mean by that. If you want the measured slope to be accurate, you need decent resolution. Otherwise you might just as well measure two corners of the board and draw a straight line through them.

Nope, sorry. It's just another fig leaf for trading off resolution. You might as well just defocus and use fewer points.

Doesn't help, I'm afraid. The scan efficiency of a mechanical scanner is better with a triangular raster, but the simple calculation I gave takes account of that. The retrace time of even a stepper motor + drawer slide scanner is only a couple of seconds, which is comparable to the pixel time.

Sure, but that isn't the primary problem.

Doesn't help much. Vacuum and air are much of a muchness in the few-millimetre length scale.

Calculating the theoretical limits is a much better use of your time, IMO. I always do that before attempting any serious development work on a project.

But as I said, that approach is doomed due to the time it takes.

"Eliminate" is a very optimistic way of putting it.

In your dreams. Sad but true.

It's all about the numerical aperture on the sensor side. If your optical system works at f/8, say, you get 1% coupling at best, with a perfectly insulated sensor.

With the amount of development work you're contemplating, you'd probably be better off flipping burgers at McD's to save up for a FLIR. ;)

IIRC germanium lenses are pretty nearly lossless in the thermal IR, unless they get hot. Free carrier absorption starts to be an issue at longer wavelengths and higher temperature.

Cheers

Phil Hobbs

--
Dr Philip C D Hobbs 
Principal Consultant 
ElectroOptical Innovations LLC 
Optics, Electro-optics, Photonics, Analog Electronics 

160 North State Road #203 
Briarcliff Manor NY 10510 

hobbs at electrooptical dot net 
http://electrooptical.net

Flir_E45_WA_Lens.JPG

Ahh, that evokes such a feeling of happy nostalgia.

Nostalgia, because that sort of thing seemed to happen all the time -- often about 9 months after some bigwig decided that his gonads were Really Big and imposed a schedule on a software group.

Happy, because I HAD NOTHING TO DO WITH IT! HAHAHAHAHAHAHA!

That sounds like FLIR, too.

--
www.wescottdesign.com

On Thu, 23 Oct 2014 19:24:29 GMT, Tom Swift Gave us:

Our bolometers focused their target image onto a 2mm spot inside the transducer. Usually through a 2mm silicon window. With a small optic in front of that, one could easily put it onto a gantry set-up above a 'device under test'. And then zip it around above it in a pattern, recording 'sample' readings.

One could also pipe a fiber optic probe tip onto the gantry and leave the bolometer in a more controlled arrangement..

You could, in fact, develop a more precise profile than the (a normal) imager does with this method. I believe you have something here.

And the data gathered could be used to allow a touch display to allow a user to select a specific "grid point" from a rendered still shot. Overlay a visible light edge definition mask, and you could erase all the fuzzy focus image issues we currently have. (the imaging guys have already gotten really good imagery)(but as you say... price)

Not real time, but hey... that hot spot *will* show up in the processed data.

Flir E4 units could be modified to give E8 resolution (I'm not sure if Flir have patched this loophole) and ZnSe closeup lenses can be had for $20 on ebay.

Flir has stopped this hack. See EEVblog for details.

The lens on the E4 can be adjusted to move the focus closer. Make a mark on the lens barrel so you can return it to normal then unscrew a turn. It will now focus close up. To do this, you might make a plastic tool to fit the notches in the lens.