Author Topic: Gaussian resistors redux.  (Read 13807 times)

0 Members and 1 Guest are viewing this topic.

Offline JonnyBoats

  • Regular Contributor
  • *
  • Posts: 141
  • Country: us
    • BitsConnect
Re: Gaussian resistors redux.
« Reply #25 on: November 15, 2011, 04:56:07 pm »
I really liked this video, particularly in the beginning when you spoke about designing test jigs.

As this video demonstrated, using a multimeter for data collection can be problematic. I would love to see a video where you discussed automated data collection and testing, perhaps using a microcontroller like an Arduino to provide for greater flexibility.

Also how would you design a test system for a production environment, either for a manufacturer or for incoming acceptance testing?
 

Offline Bored@Work

  • Super Contributor
  • ***
  • Posts: 3932
  • Country: 00
Re: Gaussian resistors redux.
« Reply #26 on: November 15, 2011, 05:41:04 pm »
When normal theory fails, make a conspiracy theory: Manufacturers do this to make it harder to find a nominal value by selection. Instead you need to buy the more expensive resistors with less tolerance from them if you need to get closer to the nominal value.
I delete PMs unread. If you have something to say, say it in public.
For all else: Profile->[Modify Profile]Buddies/Ignore List->Edit Ignore List
 

Offline EEVblog

  • Administrator
  • *****
  • Posts: 28913
  • Country: au
    • EEVblog
Re: Gaussian resistors redux.
« Reply #27 on: November 15, 2011, 05:51:48 pm »
Also how would you design a test system for a production environment, either for a manufacturer or for incoming acceptance testing?

Usually some form of PC based system off-the-shelf system with GPIB instruments or some form of National Instruments card and LabView/LabWindows CVI or some such.

Only if you needed something that wasn't possible with off-the-shelf cards like the NI one's would you resort to designing a custom micro controlled jig.
Arduino or some micro sounds all cute until you realise that making a user interface is more difficult compared to UI's like Labview etc, and you might have to do a custom PCB and talk to real instruments etc. Switch boxes are common for automated jigs, and one again, you'd use an off-the-shelf solution before you'd consider a custom solution.
In business, paying $1000 for an off-the-shelf NI card or system that "just works" is the way to go, for many reasons.

Dave.
 

alm

  • Guest
Re: Gaussian resistors redux.
« Reply #28 on: November 16, 2011, 12:28:02 am »
Another issue is the analog side. I doubt that you'll achieve the 0.1% or so accuracy necessary for this test with the standard Arduino ADC, so you need to design a precision current source and external ADC. Then you need to calibrate this somehow, which gets interesting if you need traceable calibration. Much easier to just connect a DMM to a computer (eg. the HP 3478A if Dave wasn't too cheap to buy a GPIB adapter ;)). For hobby, this could even be a cheap Uni-T meter with RS-232 option, but you may need to provide your own software, I don't think they ship instrument drivers like Labview VIs.
 

Offline amspire

  • Super Contributor
  • ***
  • Posts: 3632
  • Country: au
Re: Gaussian resistors redux.
« Reply #29 on: November 16, 2011, 01:40:50 am »
Another issue is the analog side. I doubt that you'll achieve the 0.1% or so accuracy necessary for this test with the standard Arduino ADC, so you need to design a precision current source and external ADC.

You don't need the Arduino to have great accuracy.

If you were measuring a whole batch of say 1k resistors, you would just make a bridge with another calibrated 1K 0.01% adjusted resistor, a DC source (a battery is fine) and a precise  2:1 divider. The Arduino only has to measure the difference, so if you set it up to be able to measure +/- 5% with a suitable amplifier in front of it, then it can accurately resolve down to 0.01% errors.

The range can be adjusted to be as tight as you want, so you could set the range to +/- 0.1% and resolve errors down to +/- 0.0002% - as long as the reference resistor and the 2:1 divider were accurate enough.

Once the rig is set up, all you need is to change the reference resistor for whatever value you want to test. Even if the reference resistor is of unknown accuracy, if it is stable then the Arduino can accurately measure the variation of resistor values. So you could get an accurate distribution curve - you just wouldn't be sure exactly what the center of the curve is.

Calibrating 2:1 dividers to any accuracy you want is easy - doesn't need any precision equipment.

The one particular thing you have to watch out for in a bridge like this is the reference resistor will gradually warm up, and if it is just s standard cheap resistor, it will drift.  So you would want to test a resistor, test a whole bunch of other resistors, then return to measure the first again, and see how much the drift is.

The rig would work just as well for capacitors and inductors.

Richard.

« Last Edit: November 16, 2011, 01:56:44 am by amspire »
 

alm

  • Guest
Re: Gaussian resistors redux.
« Reply #30 on: November 16, 2011, 02:32:11 am »
A bridge should work fine as long as you have something that can act as an accurate null detector. I'm curious what you do calibrate a 2:1 divider to any accuracy without any precision equipment, surely you need stable resistors and a 'null detector' with enough accuracy and sensitivity? How would I calibrate the divider to say 1 ppm without any precision equipment?

Heating up of resistors is a real issue, I remember trying to use 1/4W resistors to compare and getting frustrated because their value would increase while measuring, sometimes by as much as 0.5%. High value (Mohm) resistors seem worse in this regard. I just put a 10M resistor on a DMM. Initial reading was 9.866M. After a few minutes it had increased to 9.925M. Designing the circuit for constant current through the resistor and warming up before use should help with this issue.

Of course in a commercial setting it's much easier to specify that you used an Agilent or NI model X with a guaranteed uncertainty of Y ppm than convince your supplier/customer that your rig is correct based on circuit theory and in-house validation.
 

Offline robrenz

  • Super Contributor
  • ***
  • Posts: 3035
  • Country: us
  • Real Machinist, Wannabe EE
Re: Gaussian resistors redux.
« Reply #31 on: November 16, 2011, 02:44:14 am »
A bridge should work fine as long as you have something that can act as an accurate null detector. I'm curious what you do calibrate a 2:1 divider to any accuracy without any precision equipment, surely you need stable resistors and a 'null detector' with enough accuracy and sensitivity? How would I calibrate the divider to say 1 ppm without any precision equipment?

See Conrad Hoffmans null detector  second article http://www.conradhoffman.com/mini_metro_lab.html
This thread has a lot about it  https://www.eevblog.com/forum/index.php?topic=5442.30

alm

  • Guest
Re: Gaussian resistors redux.
« Reply #32 on: November 16, 2011, 03:06:56 am »
Sure, you can build your own precision equipment :). With low value resistors the voltage you can use across the bridge while keeping heating down probably becomes limiting, together with the finite resolution of the null detector. Will Vishay believe you if you tell them their latest batch of precision resistors fails to meet spec based on these results, however?
 

Offline robrenz

  • Super Contributor
  • ***
  • Posts: 3035
  • Country: us
  • Real Machinist, Wannabe EE
Re: Gaussian resistors redux.
« Reply #33 on: November 16, 2011, 03:19:00 am »
Sure, you can build your own precision equipment :). With low value resistors the voltage you can use across the bridge while keeping heating down probably becomes limiting, together with the finite resolution of the null detector. Will Vishay believe you if you tell them their latest batch of precision resistors fails to meet spec based on these results, however?

I agree with you.  I recently sent back 500 Bourns 0.1% resistors because 17% were out of spec.  When the Bourns tech people called me about them, the first thing they wanted to know was what meter I was using to measure and what techniques. Fortunately I had a new fluke 8846A still in 90 day specs.  End of story, refund is on the way. That would not happen with a home made bridge no matter how accurate.

Offline amspire

  • Super Contributor
  • ***
  • Posts: 3632
  • Country: au
Re: Gaussian resistors redux.
« Reply #34 on: November 16, 2011, 09:44:00 am »
The easiest way to get a stable 2:1 divider is to start with a divider network like this for about $5:

http://www.irctt.com/file.aspx?product_id=411&file_type=datasheet

In a 2:1 divider , both resistors dissipate identical power, and in these network, the resistors track to 2ppm. If you make up two 2:1 dividers, then with the aid of a toggle switch, a battery,  a low offset opamp amplifier and any multimeter, you can adjust the two dividers to within 1ppm at the current room temperature easily.

So with a few external components, and if you calibrate the divider just before testing, accuracies of 10ppm should be achievable from the divider.

Alternately if you can get two precision metal film resistors of the same batch, they should do a very good job.

Sure it is good to have a known accurate meter when you are arguing with suppliers, but we were talking here about gathering statistics. If I started measuring batches of resistors and was concerned about parts not meeting specs, I would confirm it against a known meter, and i wouldn't even mention my test rig to the suppliers - none of their business! The bridge method by the way is a very standard way of accurately measuring component against a standard resistor. It is the way it was done for most of the 20th century.

Richard.
 

Offline IanB

  • Super Contributor
  • ***
  • Posts: 9449
  • Country: us
Re: Gaussian resistors redux.
« Reply #35 on: November 16, 2011, 10:16:03 am »
In a 2:1 divider , both resistors dissipate identical power...
I've been scratching my head and puzzling in your last post and this one about how to use a 2:1 divider for test purposes, and I finally think I've got it. You don't mean a 2:1 divider, you mean a 1:1 or equal parts divider! In other words a divider where each branch has an identical resistance, giving a divide-by-two function. A 2:1 divider (e.g. 2k:1k) would give a two-thirds/one-third function.
I'm not an EE--what am I doing here?
 

Offline amspire

  • Super Contributor
  • ***
  • Posts: 3632
  • Country: au
Re: Gaussian resistors redux.
« Reply #36 on: November 16, 2011, 11:08:59 am »
In a 2:1 divider , both resistors dissipate identical power...
I've been scratching my head and puzzling in your last post and this one about how to use a 2:1 divider for test purposes, and I finally think I've got it. You don't mean a 2:1 divider, you mean a 1:1 or equal parts divider! In other words a divider where each branch has an identical resistance, giving a divide-by-two function. A 2:1 divider (e.g. 2k:1k) would give a two-thirds/one-third function.

Yes.  But if I had said a 10:1 divider, you probably would have assumed that if you applied 10V, you get 1V out.  I meant a divider that if you apply 2V across the divider, you will get 1V out across any leg of the divider. So I guess the less confusing description would have been to say a 1:1 ratio resistor network as Vishay do in their data sheets rather then call it a 2:1 divider. None of the divider data sheets seem to use the division factor in their specs probably to avoid this confusion. The IRC data sheets don't talk about ratios at all, they just say something like 50K + 50K resistor.

So sorry for confusing everyone.

Richard.
 

Offline SgtRock

  • Supporter
  • ****
  • Posts: 1200
  • Country: us
Re: Gaussian resistors redux.
« Reply #37 on: November 17, 2011, 04:50:26 am »
Greetings EEVBees:

--I can think of several reasons why the manufacturers would possibly err on the the low side more than the high side:
1) Readings below 1000 Ohms, would tend to look more correct without the use of the relative function.
2)Less metal film means less cost. Marginal I agree.
3) To allow for increase in resistance due to aging.
4) To allow for increase in resistance to to temperature; the temperature variance is more likely to vary in the up direction, than the down direction, for a number of reasons.
5)The floor managers at the factories are using OHL (One Hung Low) meters. Or maybe they omit using the Relative function to save time, and confusion, KISS.

"Does not squirrel crack nuts on bough of oak tree." Lao Fu 1410 1620

Best Regards
Clear Ether
 

alm

  • Guest
Re: Gaussian resistors redux.
« Reply #38 on: November 17, 2011, 05:30:20 am »
2)Less metal film means less cost. Marginal I agree.
A spiral is cut in the metal film, so higher resistance just means more cutting.

4) To allow for increase in resistance to to temperature; the temperature variance is more likely to vary in the up direction, than the down direction, for a number of reasons.
Depends on the material. Metal film has a positive tempco, carbon negative.
 


Share me

Digg  Facebook  SlashDot  Delicious  Technorati  Twitter  Google  Yahoo
Smf