(ultra)acoustic in circuit component level crack detector?

[coppercone2]

also knowledge of impedance units and values in this area will help in designing any sort of test system that is mechanical

i do see merits for RF but if you want to measure like a inductor or diode or something its going to get really hairy, the mechancial method should work in circuit on anything even while powered I think

correct VNA and even ohm setup (range selection) always depends on measurement impedance.




the similar wave behavior of electrical and mechanical system should make this useful to know

[rhb]

I don't think that you could get satisfactory mechanical coupling with tweezers on one side of the board.  That also would not detect a solder joint crack between the board and the device.

The correct electrical analogy is a transmission line.  The presence of cracks will attenuate the signal transmitted through the part and the board.

A significant complication is applying enough force to get good coupling, but not so much force that it closes the cracks.  And when I look at the 0.15 mm thick xtals I wonder how one can mount them so they vibrate but not break them.

[coppercone2]

maybe rubber bands? a spring should be linear too, you can add one to tweezers

[Kleinstein]

Attaching the probes would be a big problem, as too little force would cause poor contact and possible false alarm. To much force could cause the transducers to wear out fast.

With many parts on a board I am afraid individual testing with a contact sensor would be not practical as it takes a long time with manual handling and much wear on the sensors. So if at all it would be something for very high requirements where they can pay something like $10 extra per part for the test.  The more practical test would likely be high resolution X-Ray.

The is some kind of ultrasonic microscopy available already:
https://www.ksisam.com/acousticmicroscopy-en/technology-transducer.php

Depending on the circuit one could also do an electrical test under mechanical load to the board. Some cracks would show up as kind of jumps / spikes.

[coppercone2]

hmm if i wanted to experiment further, how would i crack ceramics to make test samples?

[coppercone2]

What are the mechanical units for loss from friction and loss from radiation (like ohms and radiation resistance)?

[rhb]

I was stuck in a waiting room all day and  had a long think on the subject.  It's highly unlikely that the idea can be made to work.  The frequency would have to be so high that I don't think mechanical transducers exist that can do it.  The problem is the small size of parts and the size vs wavelength required for resolution.

Suppose the compressional velocity is 8333 ft/s (water is 5000 ft/s).  That's 100,000 inches per second. So at 1 MHz a wavelength is 0.1" and at 10 MHz it's 0.01".  To get 0.01" resolution we'd need a signal at 40 MHz or higher.  To make matters worse,  even at 13 MHz, the transducer is so small and fragile that putting a strong enough signal into the xtal will probably break it.

I'm also very doubtful about attenuation as it would be hard to distinguish poor coupling from a cracked part.

I'm going to do some experiments with the xtals I uncanned,  but the very narrow response at 13 MHz will make the signal processing very difficult for any type of reflection processing.

Experiments to test:

Response of the bare xtal to a minimum width impulse from my 33622A

Response with a small piece of PCB glued to the xtal

Response of a 2nd xtal glued to the PCB as a receiver.

It's very likely that the loading by the PCB will either completely kill the impulse response or shift the frequency so low that even the transmission delay time through the PCB will be unresolvable using an 8 bit DSO.

[coppercone2]

how many bits do you need? is it really that hard so long you don't need microwave level response or complex DSP or all that back end oscilloscope stuff? or do you mean triggering issues with memory storage? Can't you do some kind of predictive sampling triggering based on process time constants? or multiple filtered channels in sync? not sure i am understanding the problem can you make a drawing?

What kind of dynamic range and resolution would it need in terms of signal acquisition. It would be actually interesting for me because as of yet any kind of signal sampling higher then something like 500KSPS has been utterly uninteresting for me. I like the slow slower sensors related to thermodynamics and chemistry that don't really need a sample speed higher then like 100KSPS even with the filters and decimation.

i am not sure what exactly you are getting at because I am getting shit about lasers on google and its 4am but keep in mind there is literally no real time data accuracy requirement is, you could do various things to the signal and resend impulses with different reciever setups till you got something that looks right if the process is so complicated it requires ridiculous dynamic range or memory, like a SAR DAC or something.

[DaJMasta]

hmm if i wanted to experiment further, how would i crack ceramics to make test samples?

Actually probably can just mimic a common failure mode - solder a few down and then flex the board a bunch.  Could also try thermal shocking by soldering one end and then putting a drop of water on the other or something (though, definitely with eye protection).


Don't know enough about the physics to make any accuracy requirement estimate, but you can try enhanced resolution modes on the scope as a baseline, and then trace averaging for a bit more.  Even for impulse response, if you measure hundreds impulses or more you should be able to get lower effective noise.  Maybe you'll still need some bits to see the information you want beyond that, but once you start seeing an interesting signal you can look into higher resolution converters with that kind of bandwidth (something like the converters in an SA, most modern ones should be able to do realtime/time domain analysis in the 10s of MHz range).

[coppercone2]

do you think frequency of oscllation would change the crack profile?

I thought to make a PCB of particular mechanical lenght with the part in a particular place and use some kind of motor to jiggle the PCB. Like a rotational to linear converter.

[rhb]

I'd forgotten about lasers. You might be able to get the BW required using a laser pulse as the source and an interferometer as the receiver.  But the resulting system would be *very* complex.

There is no escape from the wavelength vs resolution problem.  That's the fundamentals of the wave equation at work.

[rhb]

This is *really* cool.  I am astonished at the experimental result.  Setup is a 5 nS pulse at 10 MHz repetition rate from my 33622A and my MSO2204EA.

The first image is the SG connected to the scope via a 50 ohm thru.  Visually it looks perfectly Gaussian and zero phase. DS0123.PNG

The next image is the SG pulse fed to the DSO with the FFT turned on.  We see a classic comb spike series every 10 MHz from DC to Nyquist at 500 MHz and a clear picture of the antialias filter amplitude response. DS0125.PNG

The 3rd image is with the SG and the scope connected to the bare xtal.  It has now has a minimum phase very broadband filter applied which results in the asymmetric waveform.  Bode wrote the classic paper on this. It's important to note that the ringing I expected at the fundamental of the xtal is not seen.  This was a real surprise for me and suggests that a 10 cent xtal may work as a piezo transducer to very high frequencies.  This is the shortest pulse I can produce.  I now have an excuse to buy one of Leo's 100 pS long pulse generators DS0116.PNG

I then attached a 0.55 mm thick piece of plastic between two xtals with super glue.  In this instance the source and receiver are the same xtal.  In accordance with theory, we see ringing as the pulse travels through the plastic disk and xtal, hits the air-xtal interface and is reflected with a negative reflection coefficient.  It then travels through the disk until it hits the opposite air-xtal interface and is reflected again. DS0130.PNG

The last image is the direct arrival through the disk detected by the 2nd xtal.  All the other stuff is reflection off the air-quartz interface at the back side of the xtals. DS0128.PNG

The single xtal reflection response suggests that a simple probe that one pressed against a part might actually work.

I'll repeat the tests later with my LeCroy, but I'm beginning to think that coppercone's idea might actually work. after all.

[coppercone2]

hmm I have a 50KHz 70ps wave generator (old tektronix unit).

Too bad I am on vacation or I would experiment with this. I also have a nice VNA meant for categorizing crystals as a matter of fact. what you did is clever. I would like to see this work.

I wonder how you can couple tweezers to a part. I wonder if a lossy deformable material would work better despite the attenuation then a hokey contact between hard tweezers and a hard part.

god the box of crystals I got off ebay a long time ago has been bugging me, I literally used 1 part from it. one of those purchases which made me question keeping components at all. salvation at last?

also may be of some interest about crystals

[coppercone2]

hopefully i can link the ode to joy within 2 days? 

[rhb]

I don't think tweezers are needed.  I think a pencil like probe that one pressed down on  top of the part is the most attractive option.  The resonance frequency of a cracked part will be higher than a part which is not cracked.  There's really not a lot more information to be gained from the transmission response.  In fact,  it was proved over 40 years ago that given the transmission response of a stack of layers you could derive the reflection response and vice versa.

However, this would only apply to cracks parallel to the PCB plane.  But it should make a very good way to locate cracked SMD solder joints.

Here's a photo of the DUT used.  It's just two 40 MHz 3rd overtone xtals with the cans removed and a piece of thin plastic glued between them.  I bought 100 for $10 long ago to use to build xtal filters.  The minimum phase response of the bare xtal assures me that the ringing with a single sensor is a measure of the travel time thickness of the sandwich.

I made a similar plot before I glued the 2nd sensor on, but failed to transfer it to the PC.  I find it interesting that a 1 Meg load on the 2nd xtal modifies the response of the first xtal.

[coppercone2]

I am actually not sure in which direction cracks in SMD parts form, I thought it was perpendicular to the PCB but I am sure both types are possible.

[rhb]

I'd expect mulitlayer capacitors to crack parallel to the layers.  So it would depend upon how they were mounted.

I've got an ~1/4" length of 3/8" aluminum rod with an xtal glued to it with hotmelt.  I can just barely make out the reflection around 137 nS and I've got a lot of ringing in the hotmelt layer.  A friend who spent his career doing this sort of stuff in a major oil company lab suggested I try honey to attach the xtal to the sample.

The signal is very faint as I can only apply 5 V drive, but at the moment I'm DC coupled, so I may be able to double that to 10 V.

It's still  a long way from what I've done so far to having a usable tool.  But just being able to check for cracked solder joints would be  very useful.  But I've got to figure out a way to reinforce the xtal without loading it too much.  The xtal is the thickness of 2 sheets of copy paper and very brittle.  The need to make electrical connections to both side of the xtal further complicates things.

I think the next test will be to sweep the xtal from 1 Hz to 120 MHz and see how flat the response is.  That would solve the limited power input, but if you have to sweep for an hour it won't be much use.

In any case, the DSP code to make a tool rather than a lab demo will require significant effort.  I'm very familiar with it, but it's very fiddly.

I'm going to look for some lower frequency xtals as they will be thicker.

I've always wanted to do this with rocks, so I'm having a blast even if this doesn't work out.  I never had a chance to actually do anything hands on.  I just got sheets of paper and did a bunch of calculations.  Just not as much fun.

Have a Merry Christmas.

[coppercone2]

a crazy option is to maybe use liquid metal that is adhered to something by surface tension to make a low mechanical resistance connection to the crystal. Like two droplets of liquid metal that are upside down that touch the crystal legs.

Otherwise an option might be to take incredibly fine wire (say from mutli strand silicone wire that is very fine) and make a weave (I noticed solder braid is very nimble). Not sure what weave structure would work but people make weird bracelets that are pretty nimble. you could assemble it with fine tweezers under magnification but it would be a japanese effort. I don't know what the finest gauge wire they sell on spools is but I have a hunch crazy high strand count conductor would be finer. You can split a wire with a exacto blade or a special jig to remove the insulation and access the fine strands.

glue dots, silicone putty (silly putty), double sided tape, pitch, tar, fun tack (they sell it at home depot), self bonding tape, liquid electrical tape (doubtful),  all come to mind as a inbetween to honey and hot glue.

I am also wondering about if you made a little gasket between the crystal and the tweezers, filled it with honey or whatever, then used some kind of rubberish zip-tie to hold it in place (they sell weird stuff like this from panduit to act as 'softer' couplers.

Also I have seen that gaskets are actually bonded with cyanoacrylate (if you wan to make a custom replacement gasket with a gasket kit). This might also mean you can make a custom sized rubber band by cutting an existing one precisely and using super glue to bond it.

and instead of honey there are also dielectric lubricant gels like the ones used for crimps but they might actually be worse for circuits if you follow nasa guidelines about glass formation on arcing relays as a result of silicone contamination. But in all seriousness honey could result in a life-form problem of some kind ranging from microbial to rat or maybe even bear sized if it warms up and smells nice.

[rhb]

Dried honey is a water soluble glue that will form a very thin film. The operative word is "thin".  It's also easy to handle which super glue is not.

Glue dots are a nightmare. Double sided tape might work, but it's still thicker than a film of honey.  The requirement is that the layer be extremely thin so that the resonance in the layer is very high.  If that is not the case, most of the energy is trapped in that layer.

[coppercone2]

oh so its not really a big gap filler, you are not trying to get a wider contact area, just a better contact area (in the edge)? I imagined a big glob.

yea the only other thing that comes to mind right now is cooling down the tweezers with a peltier to increase the viscosity of something inorganic

[rhb]

The ideal glue would have a density of  2.65 grams/cc and be water thin.  I'm going to look for a source of sodium silicate.

A single transducer as both source and receiver (e.g. an old style medical ultrasound)  is the only way I can see to make this work.  Otherwise you have to solve for the response of the tweezers and correct for it.  That's a major PhD dissertation project over the course of several years.

But it *might* be possible to get distinct reflections from the chip to board interface and the back of the board.  A strong reflection before the back of board reflection would indicate cracking.  But there is the problem of the reflection from the backside of the transducer support.

Last night I hit on the idea of a piece of thin sheet ( e.g. 0.040 aluminum or plastic) supported by a spring.

The requirement is that we get the reflections from the front and back of the xtal to arrive at different times.  If we can't do that, then it turns into another PhD project.

This whole thread is completely insane.  But I'm having a lot of fun playing so I'll continue until we succeed or it's no longer fun.

We are fighting very basic physics, so I expect we will lose.  But I never expected the clean pulse response I got from the bare xtal. 

Another experiment that I just thought of is to place two xtals as close together as possible, but not touching and examine that response.  The xtal may be operating in shear mode in which case there will be no response from the receiver. That changes the physics a good bit.

[coppercone2]

with the tweezers and reflection, can you make lopsided tweezers where one portion is really massive compared to the other one, so it acts like a anvil? Maybe dissimilar metals (use a dense steel for the anvil and a aluminum for the transducer or w/e? Just so you can do the measurement through the component in the direction where cracks are likely to form.

Sodium Silicate is cheap, I use it for making sand molds in lead casting that are hardened with carbon dioxide.

[rhb]

If the transducer is attached to a large mass it won't vibrate.  That's the first order problem that needs to be solved.

[coppercone2]

no I mean just do a reflection test but put a backer on the part so it mimics the PCB board.

[chris_leyson]

For crack detection you're going want to use pusled ultrasound to detect the distance to a crack or boundary and in some ways it's not too dissimilar to time domain reflectometry. You transmit a short burst of ultrasound into to device under test and then use the ultrasound transducer as a receiver and measure the time taken for the signal to return, if you know the propagation velocity in the material you can work out the distance to the fault or boundary. Given that the DUT is going to be small and the propagation velocity is quite high you will need to operate at some tens of MHz. I've seen 20MHz transducers used for medical Doppler at very short distances, just under the skin and 20MHz is probably the upper limit for medical ultrasound although I have seen 40MHz transducers.

There are a few companies that sell bare unmounted PZT, lead zinc titanate, transducers with metalized surfaces. Mounting them is a problem as you have to make electrical contact with both surfaces and for larger low frequency transducers used in medical applications you can get away with soldering small, say 38AWG, wire onto both sides. At 20MHz or 40MHz the disc transducers are much smaller and become difficult to mount and attach electrodes to. Small high frequency transducers are rather speciallized and expensive and not something I would contemplate building at home. These guys have a good range of PZT transducers, mostly low frequency though. https://www.steminc.com/ Maybe send them an email, you never know they might have something suitable.