No Script, No Fear, All Opinion
RSS icon Home icon
  • EEVblog #216 – Gaussian Resistor Redux

    Posted on November 14th, 2011 EEVblog 50 comments


    Not content with 400 resistors, Dave ups the ante with several thousand genuine OneHungLow resistors, a semi-automated test jig, and some anger and love for the Agilent U1272A multimeter.

    Be Sociable, Share!
    • http://chrisvighagen.se Chris Vighagen

      So, has Agilent sent you a letter yet?

      I mean most of the stuff you critique here on the eevblog warrants a response, you were dead on, on the multimeter that failed criticaly, you were right a bout microchip, your suggestions for the Agilent Scope got corrected etc.

      Its obvious that someone listens to your blog and take the sugestions seriously.

      Though if I were you, I would not have given up after 1000 measurements. Come hell or high water Id get that data out of there. Im stubborn like that.

      Great episode, would love more like this and not just product reviews.

      Cheers!

      Chris

      • Geoff

        You could just open the CSV in notepad and do a search replace for ‘ m,’ with ‘,’ to get rid of the extra crap.

    • sqkybeaver

      i had to pause, i was laughing so hard when i saw the onehunglow brand sticker.

      bloody marvelous!

    • Zac Bailey

      Dave!

      You got frustrated and forgot the obvious question to be answered!

      Having done all of this testing will you now use a name brand resistor or a “OneHungLow” resistor for your future designs?

      Z-

      • http://www.eevblog.com EEVblog

        Sure, they were all well within the quoted 1%, and that’s the spec you’d usually design around.

    • http:thechipmunk.net Martin

      Dave, maybe the OL reading on the first one was causing a problem with the import? Is there anyway you can delete the first one and try again…

      • http://www.eevblog.com EEVblog

        Other ones I’ve uploaded had OL on the first sample too. Haven’t looked into deleting any yet.

    • Ondrej

      You can be even faster if you measure sample instead of entire population. Let’s assume 10% margin error (E = 0.1), 90% confidence level (c = 0.9) — you have to measure 64 resistors only (randomly selected out of 1000).

      x = Z(c)^2 * 0.5^2 (= 0.6724)
      n = (N * x)/((N-1)*E^2 + x) ( = 63.06 -> 64)

      Z is from this table:
      c z
      80% 1.28
      85% 1.44
      90% 1.64
      95% 1.96
      98% 2.33
      99% 2.58

      • http://www.eevblog.com EEVblog

        The whole idea is to measure every one of them and see what the results are!

      • JRR

        That would defeat the entire purpose of what he’s trying to accomplish here. He’s trying to ESTABLISH whether sampling would be effective, and what the population curve looks like. Unless the entire population is sampled you can’t really be sure.

    • Ondrej

      Could be that average around 0.4% is caused by test jig resistance? (0.4% -> 4ohm)

      • http://www.eevblog.com EEVblog

        No, it was zeroed out, and it’s not that high anyway, and in the wrong direction!

    • George Graves

      I wonder if those values change after they been heated up to soldering temperature? And by how much?

      Part 3? ;) (I bet you’re done with resistors for a while! Ha!)

      • http://www.eevblog.com EEVblog

        More to come!

        • George Graves

          HA! Love it! Can’t wait to see what else you come up with.

          • Jay Ts

            After seeing this, I won’t be happy until Dave designs a machine that automatically tests a full reel of SMD parts while he sits back with his feet up! (It would be just like watching a movie. hehe.)

    • Eric Hill

      We have seen problems with USB to COM port adapters similar to what you’re seeing, although not with that specific meter. We usually clear it up with a reboot of the computer while the device is attached. We’ve been requesting FTDI chips in all of our newer test equipment for just that reason.

      • f4eru

        I second that.
        Had this issue with a rohde&schwartz spectrum analyzer(FSH3) which has an usb-uart integrated. After some failure, and a lot of tests, we found that it didn’t work due to COM port problems, especially on “some” computers.
        After extensive communication with RnS, an official response was that the cable driver was not working on some dual core computers (new at the time). And that we had to purchase (!!!) a new cable with a chip using another USB-UART. Basically, the chip manufacturer offered no more support !!! crazy.

        My suggestion : simply test it on another computer. Probably a driver problem on these always unstable converters.

    • http://blog.mikucki.eu Mikuslaw

      Thanks Dave for the video, I was very curious how you will solve the problem of automating the process. It seems that the simples ideas work the best. It’s a shame that the second batch didn’t download. It is really interesting how the production process is organized that such a shift occurs. Just by luck that the material was less resistive during production of this series, or maybe it’s easier to get far positive outliers and they want to be on the safe side to always stay in range.

    • Zak k

      Cannot believe that this has not been stated, but isn’t that an anti-static mat? AKA a surface that is going to reduce resistance if you happened to press your probed wired right into it?

      might want to check that out. you should be able to get and absolute error by testing the jig directly on the mat.

      thanks for the video mate

      • Zak k

        or If that is in fact negligible, what if the piece you used to zero out the jig wasn’t actually ~0.0ohm?

        from what I saw during the few values flashed from the second run they were all low also. since these are two different manufacturing

        • http://www.eevblog.com EEVblog

          It was zero ohms.
          People keep tryign to explain the offset as some form of measurement error, I can assure you all it’s not the ESD mat, it’s not the bandoleer, it’s not the temperature, it’s not the zero offset, it’s the resistor themselves.

      • http://www.eevblog.com EEVblog

        It’s not thew ESD mat, it is dissipitive, not conductive. And YES, I’ve measured it. The mat reads OL with the Agilent meter, and is in fact in the Gohm range.

    • Mike

      I consider cheap chinese resistors OK to use. Electrolytic capacitors are another story after a few months (in circuit) cheap ones often show how crappy they are.

      Great show, very entertaining.

    • Ian K Rolfe

      The readings were consistantly low because during the cultural revolution the Communists eliminated .3% of the resistance .,,,, :D

      • http://www.eevblog.com EEVblog

        LOL! I’ll have to pay that one!

    • Magicmushroom666

      Wow that agilent meter really seems to handle the data logging well, just as you’d expect! very impressed. Just a shame on the PC software side, however i’ve found that to always be an area where companies skimp on the time spent. I guess as not many end users need it.

      • Bill Clay

        I agree, especially in respect to Fluke meters. The data logging software they have looks like it is from the 1980′s.

    • Wonko The Sane

      OneHungLow !!!!

      What did you do, did you use OneHungLow as you order number – so they put it on the box ???

      LOL…

    • Wonko The Sane

      .3% offset is 3 Ohms, so your mean is 997 Ohms

      if you put 330,000 Ohms across a 1,000 Ohm resistor you get 997 Ohms…

      What Value do you get measuring you ESD mat with your Probe ?

      • http://www.eevblog.com EEVblog

        The resistor measures the exact same value either in or out of the bandoleer. The ESD mat is effectively infinite.

    • http://sven.killig.de Sven Killig

      To salvage the data, you could do a bit SCPI yourself, see
      http://www.eevblog.com/forum/index.php?topic=2423.msg43750#msg43750
      If I remember correctly, a small gotcha is the line ending to delimit a command, at least on my U1253B.

    • Rasto

      I think You have unofficial answer here from Eric Hill above. If You google, You’ll find email address pointing to agilent.

      I also have bad experience with USB-COM adapters which Eric points to be the issue here.

    • rasz

      theamphour host died , cant listen to last episode :(((((

      • pete

        you are talking about the server, yes?!

    • Stephen1424

      I agree that if possible, deleting the first 2 entries may be a solution if it is possible.

      Hope you can get it working, I am curious to see the next set of data.

      Also, I love the rig, great idea !

    • Bill Clay

      Dave,

      Even though you tested the Agilent with a .01% resistor, couldn’t the offset be due to the fact that you used your bench meter the first time? The Agilent meter itself may be offset compared to the other. Care to repeat the 100 resistor test with the Agilent?

      PS I like the test jig. Brilliant!

      • http://www.eevblog.com EEVblog

        The bench meter has nothing to do with this series of tests on the new resistors. The U1272A is absolute accurate to at least 0.01%, and we see a 0.3% offset.
        The bench meter would read the same.

        Dave.

    • ThunderSqueak

      automate these types of measurements with legos?

      The OneHungLow Brand made me giggle :>

    • ac

      Did you test the thermal coefficient of the OneHungLow vs the brand ones?

      I have found that eg. ebay “metal film 1%” brand has very much different coefficient. It says metal film one the sticker but the coefficient is carbon film like.
      The resistors are colored like as if they were metal film.

      At room temp they are all within 1%. I got my money back from the ebay seller after mentioning this.

    • Bob

      Oh my god, you’re the nerdist guy I have ever seen!

    • salec

      You may want to see this little USB to GPIB project Google dug up: http://cluster.physik.uni-freiburg.de/~kuhnen/pic/pic_usbgpib/

    • Rubi

      Hi

      I also use the 3478 to test resistance.
      IO bought a Gpib to Usb Adapter from Sparkfun.

      http://www.sparkfun.com/products/549

      I use this adapter also for my TDS 210 scope.

      There is also a project from Elektor with free source code.
      http://www.elektor.com/magazines/2011/april/gpib-to-usb-converter.1738967.lynkx

      Cheers
      Rubi

    • alex

      Page 7 of this cheapo resistor datasheet shows a tolerance bias
      http://www.rapidonline.com/pdf/62-0712e.pdf

      • http://www.eevblog.com EEVblog

        Thanks. Interesting, more so for the similar +/-0.5% range I think…

    • Ling SM

      Zero with the same pressure as measurement on pogo test pins, and check the resistance of a R in the strip and out of the strip. Any diff?

    • Chris

      Love your videos Dave. I’m a mechatronics engineer so have uni level experience and strong interest in the sort of things you do but a lot of my professional work lands in the Mechanical/Process sphere. I’ve been doing some contract work lately with a large Australian building product company and statistical process control is certainly an area I’ve learned a lot about in the time there.

      There are some great measures to get a handle on the numbers – in particular Cp and Cpk – which can give you a feel for how ‘good’ the process would be if it were centred (eg your onehunglow resistors – fairly tight bell curve but not centred – they’d have a strong Cpk but weak Cp).

      http://en.wikipedia.org/wiki/Process_capability_index

      Keep up the good work mate, I’ll keep watching :)

    • http://www.cliftonlaboratories.com Jack

      I don’t know if this is true, but it certainly sounds plausible … and it’s consistent with measurements I made a couple years ago of 5% Xicon carbon film resistors.

      If you remove the epoxy film from the carbon film part, you will find a spiral cut into the resistor, down to the ceramic tube substrait. I’ve been told that the resistor manufacturing process is to screen and fire the carbon film so that the resistance is below the target value. A diamond tool or laser then cuts the spiral until the resistance increases and the part falls into tolerance.

      In order to save time in the trimming process, the machinery stops when the part is “close enough” and hence all the resistors trimmed this way will be lower than the nominal value. If the tolerance is 5%, and as a safety measure to assure all parts shipped meet the spec, the algorithm might be to trim until the resistor measures -2.5% from target and then stop trim. To trim to 0% of nominal wastes machine time.

      I looked at a few old carbon composition parts to see how much they had drifted, and also a small set of 1% and 5% new parts. Discussion and plots are at

      http://www.cliftonlaboratories.com/carbon_composition_resistors.htm

      Jack

      • http://www.cliftonlaboratories.com Jack

        “I don’t know if this is true, but it certainly sounds plausible …”

        By this I meant the story about trimming the resistors and stopping when they were in tolerance from the low side in order to save machine time.

        Jack

    • Pingback: A machine to measure resistors...