Crack detection

[ricko_uk]

Hi,
we have a differential (bridge type) eddy-current sensor for metal crack sensing in large totally flat sheets of various metals. But we have a number of issues and we came to the conclusion that it might be easier to redesign the front end interfacing with the sensor.

Attached is the basic high level block diagram showing the concept. The two coils are arranged in a cross arrangement (one inside the other at 90 degrees to each other and both at 90 degrees to the surface being scanned). When there is no defect on the metal surface underneath the coils the voltage across the coils is the same. But when there is a defect there is a difference between the voltages across the two coils. Standard bridge measurement.

All we need is to have a DC output proportional to the difference in the voltages across the coils.

The issues are:
   - the voltage difference is very tiny, microvolts at best
   - we need to keep noise at a minimum
   - cost is NOT an issue (within reason ).

Questions:
   - How can we maximise the voltage difference between coils (for any given defect under the coils) while keeping the noise down. We tried changing the frequency but it does not make any difference.
   - Ideal parts to use at the front end
   - We googled but could not find any schematics. Does anybody have any willing to share? Or even just a picture of a schematic on a napkin would be appreciated
   - Does the bridge driver need to have any special requirements such as be a current source (perhaps to help thermal stability on the short term - the system is used for a maximum of 8 hours at a time then it is reset)?

Thank you again in advance

[Kleinstein]

I don't see a way, why there should be a DC / LF signal at the coils, unless the cracks have some kind of diode characteristics. Normally the bridge should give an AC signal with the same frequency as the excitation. To simplify the detection I would consider to have differential drive (e.g. trough a small transformer and the output amplifier to ground. So no differential amplifier needed. Chances are the most sensitive amplifier would be a discrete JFET circuit.
For the bridge I would consider a configuration with the 2 sense coils in series in one arm of the bridge and a transformer to as the other bridge. A transformer can be pretty stable compared to resistors and could also be used to drive the bridge differentially.

I am not sure if the coil use in a bridge is really the best way to go. It could be better to have one coil as a excite and the other as a detect coil. However this would not be sensitive to cracks in all directions. It would not detect a crack in parallel to one of the coils. So it would need kind of 2 runs (with the sensor rotated by some 45 degree) or some kind of 2 sensors combined.

AC detection would be most sensitive with phase sensitive detection in one kind or the other. Today this could be relatively fast sampling (e.g. 4 samples per period) with an ADC in sync with the exciting signal. The rest would be done by math. One could also use analog mixing, though it's likely more trouble. ADCs for the 500-1000 kHz range are pretty common these days. With suitable analog filtering 8-12 Bit resolution should be enough.

Different frequencies would effect the depth - lower frequency would go deeper.

[RandallMcRee]

Hi,
...Standard bridge measurement....
   - We googled but could not find any schematics. Does anybody have any willing to share? Or even just a picture of a schematic on a napkin would be appreciated

Thank you again in advance

I'm not sure why you were not able to find schematics? What search terms did you try? Bridge measurement is discussed virtually everywhere. Here are a few pointers
Low noise IN amp: https://www.analog.com/en/products/ad8428.html
Very low noise IN amp: https://www.analog.com/en/analog-dialogue/articles/low-noise-inamp-nanovolt-sensitivity.html
Bridge measurements by Jim Williams: https://www.analog.com/media/en/technical-documentation/application-notes/an43f.pdf
Bridge reversal circuit: https://www.analog.com/en/design-center/reference-designs/circuit-collections/bridge-reversal-eliminates-1-f-noise-and-offset-drift-of-a-low-noise-non-autozeroed-bipolar.html#cc-overview

Etc. Etc. Just scratching the surface here. In order for us (eevblog readers) to be more specific you need to provide some detailed specifications (like answering Kleinsteins questions for a start!).

[Marco]

I don't see a way, why there should be a DC / LF signal at the coils,

Yeah, a simple difference in AC impedance of the coil seems far more likely. That also simplifies things a bit, because at least DC offset becomes irrelevant. Can just AC couple the output of the inamp.

For phase detection, why not just amplify both signals, square them up, XOR them and low pass the result? If you peak detect the output of the inamp to see a change in amplitude difference, you end up with two low frequency signal.

[rhb]

What is the objective, determine if there is a crack, its location or both?

Also why not use ultrasonic methods?

The geometry is fairly important in any case.  As noted earlier, certain crack and sensors geometries will be very insensitive.  So you will need multiple sensors to get reliable results.

You really need to specify a use case.

Reg

[ricko_uk]

Thank you Reg,
The objective is to just detect the presence of a crack and a relative indication of depth (i.e. indicate that a crack is deeper than another).

- NO precise depth measurement
- NO shape of the crack
- NOT its location

It has to just detect that when the probe passes above it there is a crack under it. Any suggestions?

Thanks

[ricko_uk]

Thank you Kleinstein and Marco,
could you please elaborate on both of your points?

1) if I understand correctly you suggest to just place an AC coupling capacitor on the IN-AMP output? Then the INA output signal signal would be a AC coupled sinwave of amplitude equal to the voltage difference between the two nodes at the centre of the bridge amplifier. Correct? If not could you please explain further.

2) why the phase detection? Is that as an alternative approach or complementary to the first part? Or is it for detection of the probe's lift-off from the surface (someone suggested something similar for lift-off detection)?

You mentioned "If you peak detect the output of the inamp to see a change in amplitude difference, you end up with two low frequency signal.". Because I don't understand your second point (about the phase detection), if is it is at all relevant we don't need to sample the variation any faster than 70 Hz. Not sure if this changes anything.

Thank you

[Marco]

2) why the phase detection? Is that as an alternative approach or complementary to the first part? Or is it for detection of the probe's lift-off from the surface (someone suggested something similar for lift-off detection)?

The phase difference wasn't my point, but Kleinstein's. The R/L divider will cause both an amplitude change if L changes and a phase change, you can measure either and they will provide the same information.

The signal out of an inamp will be an AC signal at the driven frequency, of which common mode leakage will make up a significant part and which at 100-200 kHz is fairly high frequency. That said, subsampling is probably the most appropriate way to deal with it instead of trying to do peak/phase detection in analog. If you ADC it you can do a single frequency Fourier transform, which will give you much better noise resistance.

PS. you don't really need to generate a sine, driving the coils with a low passed square wave will work too ... the fourier transform won't care about the harmonics.

[rhb]

Thank you Reg,
The objective is to just detect the presence of a crack and a relative indication of depth (i.e. indicate that a crack is deeper than another).

- NO precise depth measurement
- NO shape of the crack
- NOT its location

It has to just detect that when the probe passes above it there is a crack under it. Any suggestions?

Thanks

A use case would be a huge help.  I'm a retired research level reflection seismologist.  So very heavy multichannel DSP and remote sensing background.  I'm also very familiar with every imaginable manufacturing process.  Lifelong fascination with how things are made.

Ultrasonic and x-ray are the standard methods for testing welds.  I can't say for certain as I've not done a mathematical analysis, but I don't think that eddy current sensing will detect a horizontal crack in a sheet coming out of a rolling machine to make up an example.  Ultrasonic certainly would, but I don't think x-ray would.  I cannot think of anything, but ultrasonic that would work.

There are obvious difficulties with ultrasonic sensing of say a hot roll line output, but one can use a high power laser to create impulses and  a laser interferometer to sense the reflection response which allows you to avoid the contact problem.  John Scales at the Colorado School of Mines was able to rotate a rock core on a turntable, create impulses with a high power CO2 laser and measure all 36  coefficients of the elastic tensor with an interferometer.  That was over 20 years ago.   To detect cracks which are at an angle to the surface one would require more than one sensor, but that could be implemented  by scanning and mirrors.

A good use case is important as inspecting a single part is very different from inspecting product from a processing line.  To continue with my hot roll example, the NRE cost would be in the low 7 digit range, but I see no fundamental barrier to continuous inspection of rolling line output for cracks in real time with very high reliability.  Not quite nuclear grade inspection, but close.  Unit cost would be quite low.

On the other hand a cold roll line would be much easier and cheaper to design the system for as you could use contact sensors.

If you don't want to discuss details publicly, send me a PM.

Reg

Edit:  I had a look in my library.   Get a copy of:

Electrical and Magnetic Methods of Nondestructive Testing
Jack Blitz
Adam Hilger 1991

it has analytic solutions for inspecting tubing for defects.

[martinr33]

I have some experience with eddy sensors. As you have discovered, the signal is small compared to the drive and so difficult to detect.

Commercial eddy sensor ICs use a different approach. They set the coil to oscillate with a damping factor on the edge of shutdown. When a metal object is in range, oscillation stops.

In your case, you would measure this setting by scanning the damping range. A crack would allow the sensor to operate at a lower damping level.

I think that TI has a dual sensor IC that would let you scan two coils at once.

[ebclr]

This commercial line, may give you some insight

https://www.ibgndt.ch/?gclid=Cj0KCQiAw4jvBRCJARIsAHYewPN1F2pSPUPHUeSXd7suZyZwM8bc5cXLn07xprPRX6QffGp_QY-L6t8aAoBPEALw_wcB

[ricko_uk]

Thank you Reg  ,
we do have to stick to coils/Eddy-currents  type solutions/sensing for a number of reasons. The equipment has to be integrated into other testing equipment/assemblies and the space is limited. We did look into it over a year ago but we discarded it from the beginning. Also in one of the product configuration the sensors are placed in a tight grid arrangement and that also would not allow the ultrasonic sensor to be placed close enough to each other, while with the coils we have much more flexibility in reducing their size. In terms of detecting cracks in sheets coming out of rolling machines, that is we we have the 2 coils arrangements.

Thank you for the suggestions and input! 

[ricko_uk]

Thank you Martin ,
that's very interesting (!!) I thought of that few days ago and researched it but could not find any such solution (perhaps wrong search terms?) and was worried that it might have some drawbacks I could not think about. So thank you very much for that info!!

UPDATE:
I just wired it up as shown in the attached schematic and driven the circuit with a signal generator at the resonant frequency. That is that I had in mind, is that what you suggested or is your suggestion different?

1) As I move the coils close to the metal the voltages across the two resonant circuits in the bridge move of the same amount (20mV variation in ampliture in BOTH). And you can see that in the attached screenshots. That I was expecting although A) I was hoping in a bit more than 20mV... and B) the traces are not exactly superimposing, the probes are calibrated so I assume that tiny offset comes down to the components tolerances?

BUT...

2) If I slide the coils along the metal surface and above a crack, nothing happens. There is NO variation whatsoever across either one of the resonant circuits. That I was not expecting at all...

Any suggestions?

Thank you again for all help and suggestions!!

[Marco]

You were only expecting uV I thought? You can't see uV on the scope. You'd have to do a Fourier transform to get sufficient dynamic range to see uV changes in the amplitude of the difference signal (which at the moment is dominated simply by the variation in the base values of the resistors and inductors).

[martinr33]

The problem is, the signals are very, very small.

Here's another solution.
http://www.ti.com/product/LDC1612

It isn't based on eddy current, instead it measures inductance.  The trick here is it has 28-bit resolution - which might let you see the effects of a crack. Plus, it has dual channels.

Eval board:
http://www.ti.com/tool/LDC1612EVM?jktype=tools_software

[ricko_uk]

Thank you Martin and Marco.

Martin, I actually have that eval kit from a while back and have been playing with it since reading your previous suggestion about it.

When I plug my sensor coils into it (80uH) they do provide a variation but it is very tiny exactly as you said. But also very erratic especially as I move the probe along the surface (i.e. even a slight tilt seems to have the same variation of a small crack). I have been tweaking all day trying different settings but not getting any decent results. I have been reading the datasheet but quite a bit of it is beyond my expertise. Just one example is the Rp setting...

Have you used that IC or others of that family before? Do you have any pointers to have a much larger swing?

Thank you again!

[Marco]

But also very erratic especially as I move the probe along the surface (i.e. even a slight tilt seems to have the same variation of a small crack).

That's always going to be a problem with a small effect magnitude ... how do you know the concept even works on principle? Are there existing products which use it or papers/patents which describe its use?

[martinr33]

I think you need to tightly control the distance between the sample and the sensor. Your variables are going to be coil geometry; frequency; coil positioning. You'll need a rig that holds the coils precisely in place so you are not measuring distance variations.

The TI chip looks like a good part for what you are trying to do.

[ricko_uk]

Thank you Kleinstein,
today I was also thinking about using a PLL for phase detection?
More precisely inputting into the PLL phase comparator 1) the original signal that drives the bridge and 2) the amplified signal from the bridge center taps. Then use the low pass filter output as a linear representation of the phase difference?
Any thoughts/feedback?
Thank you

[Marco]

It will require far more time to get the noise down than ADC/Fourier solution ... a phase detection circuit only gets information at the transitions, a Fourier transform gets information from the entire signal.

Similarly peak detection is much worse than Fourier for detecting amplitude.

[RoGeorge]

Did you tried using lock-in amplifiers?

Since the original signal (from the generator) is available, any noise can be easily filtered out by a lock-in amplifier (AKA homodyne detector, synchronous detector, etc.)
https://en.wikipedia.org/wiki/Homodyne_detection
https://en.wikipedia.org/wiki/Lock-in_amplifier
https://youtu.be/rzzliN_vTKs?t=252

[ricko_uk]

Thank you George!
I thought of it but have no experience with lock-in amplifiers. Only basic knowledge.

Are there any lock-in amp ICs?
The only one that comes up in seraches is the AD630 but it is 18 GBP in quantities. Cost is not "much of a constraint" but the system would use 20 (1 x channel - because of the modular design and requirements) which makes just those ICs more expensive than all the boards in the entire system together.

If not are there any "simple" schematics for it? The ones I came across just now are massive and I think they include much more stuff than I need but wouldn't know which part is relevant.

Finally, is the idea to then measure the amplitude variation? If so Marco (please see previous post) mentioned it might be an issue. Any suggestions to help mitigate that?

Thank you

[chickenHeadKnob]

How is what you are attempting different than Magnafluxing? That has been standard in the Aviation world for ages. Off the shelf available.
https://www.magnaflux.com/Magnaflux

[ricko_uk]

Hi chickenHeadKnob
No different from that. We just have to develop it ourselves as part of a larger electronic system/PCB with plenty of other metal inspection features and control. It all has to be enclosed in a single box together with automatic motion control etc. So we cannot use any external third party systems. Not to mention the boss wants the entire system to be as compact as possible.
Do you have any reference schematic or any other resources? Any infos is very much appreciated!!
Thank you

[Marco]

I wonder if magnetics is really the best way here, it's useful with the magnetic particle method because it's so low tech and precise ... you instantly see disturbances across a large area.

The moment you start doing it purely electronically though, why not just use resistance measurement with capacitive electrodes? Due to skin effect a surface crack will generate a large effect in resistance across it.