How reliable is a calibration?

[mikepa]

I just picked up a Agilent E3615A 20V 3A power supply from eBay and it came with a calibration certificate saying it was just calibrated a few days ago.  The company I got the supply from was Brltest and they did the calibration, so I was expecting the calibration certificate to mean something.

But on testing the output voltage, I see the power supply voltage indicator is reading 0.07V low at 10V and 0,22V low at 20V.  (This is with or without a 10 ohm load on the supply).  At first I thought it must be my Fluke 179 that was out of calibration (it has never been calibrated), but as I have two other DMMs (Fluke 77 & one from RS) for a total of three that all read within .02 volts of each other, it makes me think the power supply is the one out out of calibration.

Thoughts?  Is this expected or should I ask Brltest to re-calibrate - or should I just fix it myself?

[EEVblog]

The calibration itself is practically 100%
But remember that calibration is only valid for the instant that calibration was done.
The rest of the time has to do with the confidence you have in that instrument. This is built up with calibration history that gives you a confidence level in a particular instrument.
Technically, calibration labs also offer confidence level based on certain gear.
But it's all a statistical thing of course, you could have a complete dud.
This is why some critical things mandate the use of two meters, to ensure that one has not drifted or is not playing up.
Welcome to the interesting world of metrology.

[EEVblog]

For your actual problem, what is the PSU spec? Most PSU do not have tight specs.

[alm]

You should not expect the company to make any adjustments if it's well within specs. Many will make adjustments if it's off by more than for example 70% of its specified tolerance.

[mikepa]

E3615A meter accuracy is stated as ±(0.5% of output + 2 counts).  So reading 9.99V while output is actually 10.07V would be just out of spec.  Reading 20V while output is actually 20.28V is way out of spec.

[alm]

Either the supply drifted a lot the past few days (defect?), you're really unlucky with three DMMs off by the same amount, or someone botched the calibration. I would call them and see what they have to say.

[Galaxyrise]

Just another sanity check: How long did you leave the PSU on before testing the voltages?

[mikes]

It's eBay. Let me help:

For parts = someone put this one through a crusher.
Won't power up = this experienced a direct lightening strike
Powers up/can't test = something changes when you turn it on, but it's broken with valuable pieces removed.
Working = it's broken in some way, but the return shipping costs will keep you from returning it.
Calibrated = we checked a few things, and our China meter says they're close.

[vk6zgo]

0.7% out at 10volts is hardly the end of the world!

Neither is 1.1% at 20volts.

After all,the thing was on eBay,not brand new,so I assume it was cheap.

The easy answer is,just use your Fluke to set the voltages when they are critical.

There are a lot of other more important specs for power supplies than the readings on the front panel meters.
Have you checked for long term drift in settings,hum level,short term stability,etc?

By the way,I notice you changed the figures in your later posting--which is right?

[quantumvolt]

I second the last post. Those 2 1/2 digits meters are probably of the same quality that you get for a few dollar today. The quality of the supplied voltage/current is what matters.

Also we do not know the real voltage you are measuring (your other meters have unknown absolute inaccuracy and differ in readings), so the PSU meters might err less than you think.

Since both readings are low and the 20 V error is bigger, your meter could be quite linear but with a too low resistance to generate the voltage reading from the tapped current. You can look for a trimpot to lift the 10 V reading to 10 (or a tad over) and also at the same time lifting the 20 V reading closer to 20. Or you can put in an op amp with variable offset and some 1.005x trimmable amplification. What is clear though is that you will never get both readings accurate (because of unlinearity in the meter).

Or you can buy a 4 1/2 digit Voltmeter module on eBay for 10-20 USD (I got one, they claim informally around 0.3% accuracy, but mine is closer to 0.1% within the readings from an Agilent 34401A).

[eevblogfan]

hey

simply open the case and adjust the pot(s) right there , that's easy to do and takes no efforts ( unless you are looking for +- 1 LSD 0_0 )

your flukes are more then enough so feel free , 

if you will state your country , I am  sure there will be one eevbloger who might have the will to either calibrate or simply borrow you some meter say fluke 87V or any other goodies  

Cheers

[mikepa]

Thanks for the replies guys. 

I left the unit on all night set at 10.00v and it's still 10.00 volts this morning, so that's good, zero drift.  But the actual output voltage still is .07V high at 10V and 0.2V high at 20V.

Alm seems top have the best advice (call brltest), while Mikes seems a little cynical of eBay goods.  If an auction emphasizes a calibration is included, one expects the thing to be calibrated.  (Cosmetically the condition is like new, I don't know how old the unit is.)

I am tempted to just do it myself as Eevblogfan suggests, but when one has essentially paid for a professional calibration …   

I see very little noise on the output < 4mV.

[G0HZU]

I don't know about your Agilent supply but I have various Thurlby PL320 dual and triple PSUs here and some of them are 20 years old.

I just checked all the PSUs when set to display 20.00V and they all measured 20V within 0.015V.
Note I set each rail so the display 'just' clicked over to 20.00V from 19.99V to try and get some consistency in the way I set up each one.

I also have a Thandar dual supply that is a lot newer and it only reads to two decimal places up to 18.00V but both rails are within 0.015V at 18V.


I might have 'calibrated' the older supplies about 15 years ago but they still seem to be pretty accurate today. I didn't really expect them all to be this accurate but that's the results I got

So I would be a bit annoyed if I was sent an Agilent PSU that was supposed to have been calibrated and it was out by 0.20V at 20V.

However, I think the best option is to tweak it yourself

[branadic]

Or you can buy a 4 1/2 digit Voltmeter module on eBay for 10-20 USD (I got one, they claim informally around 0.3% accuracy, but mine is closer to 0.1% within the readings from an Agilent 34401A).

You mean 3 1/2 digit moduls, don't you?

[alm]

Agilent specs are usually realistic and quite conservative, even though the indicators are not exactly the most critical part. In general I would not buy calibrated equipment on eBay with the expectation that they performed a factory calibration, but that doesn't mean I wouldn't complain. The value of a cal certificate depends on the credibility of the company, and my impression of Britest is that they seem fairly credible. If they give you a piece of paper that it met specs (careful they didn't do a partial cal), then they should behind it or admit that the equipment drifts way more than specified.

[jucole]

A calibration should be reliable, but if the item was shipped a long distance, or it's old gear - i think you'd be unrealistic to assume it would be have to be anything other than spot on;  but the best thing is you can tweak it yourself! 

[madires]

Alm seems top have the best advice (call brltest), while Mikes seems a little cynical of eBay goods.  If an auction emphasizes a calibration is included, one expects the thing to be calibrated.  (Cosmetically the condition is like new, I don't know how old the unit is.)

Calibration != Adjustment

A lot of people mix up both or think both are the same :-) Calibration is the process of comparing what your device displays and what a calibration device (traceable to a standard) displays. It's about finding out how stable your device is over time. If you like to have a device which is spot on you'll adjust it to display a correct value. In case you want both (cal & adj) it becomes more complex :-) First you calibrate your device, then adjust it and calibrate again. The first calibration is to track the changes since the last one and the second calibration is the starting point for the new values. Otherwise the next calibration would show a drift of 0.07V in your case.

[alm]

The cheapest calibration is calibration without data. This is a certificate that states it was within specs at a certain time. If it was outside of the stated tolerance, then it should have been noted on the certificate. These calibration levels will often involve adjustment (eg. Fluke CalNet), since you can't track the stability without data anyhow.

[jucole]

If you like to have a device which is spot on you'll adjust it to display a correct value. In case you want both (cal & adj) it becomes more complex :-) First you calibrate your device, then adjust it and calibrate again.

If you wanted a device spot on; you'd ask for cal & adj, and if required, a "repair". no?   

Basically if the OP power supply is outside the manufacturers spec. but when adjusted  then drifts faster than the manufacturer stated 8hr stability drift,  it will need a repair.

[jancumps]

If you like to have a device which is spot on you'll adjust it to display a correct value. In case you want both (cal & adj) it becomes more complex :-) First you calibrate your device, then adjust it and calibrate again.

If you wanted a device spot on; you'd ask for cal & adj, and if required, a "repair". no?   

Basically if the OP power supply is outside the manufacturers spec. but when adjusted  then drifts faster than the manufacturer stated 8hr stability drift,  it will need a repair.

Doesn't that mean (if there was a log of old calibration data) that you loose calibration history as soon as you adjust?

[mikepa]

The certificate states:

As Found: In Tolerance
As left: In Tolerance
Adjustments: N/A

With adjustments stating N/A, not sure if it's not applicable because it wasn't needed or it wasn't part of the deal.

Regardless, a 0.2V deviation at 20V I believe I am correct as stating the device meter is not within tolerance, so I will contact Britest (is it Britest or Brltest?) and see what they say.  It was shipped only 100 miles and was very, very well packed so I doubt rough shipping caused it to drift.  I suspect the calibration was botched.

P.S. The current reading is equally out of tolerance.

[jucole]

If you like to have a device which is spot on you'll adjust it to display a correct value. In case you want both (cal & adj) it becomes more complex :-) First you calibrate your device, then adjust it and calibrate again.

If you wanted a device spot on; you'd ask for cal & adj, and if required, a "repair". no?   

Basically if the OP power supply is outside the manufacturers spec. but when adjusted  then drifts faster than the manufacturer stated 8hr stability drift,  it will need a repair.

Doesn't that mean (if there was a log of old calibration data) that you loose calibration history as soon as you adjust?

I'm not sure i'd be that interested in the historical calibration data for an Agilent E3615A with an accuracy of +- 0.5% + 2 counts.

[jancumps]

I'm not sure i'd be that interested in the historical calibration data for an Agilent E3615A with an accuracy of +- 0.5% + 2 counts.

You're probably right.

[Rufus]

Calibration != Adjustment

Calibration == adjustment (or recording of correction factors to be used in subsequent measurements).

Measuring without adjustment is called performance verification.

Performance verification provides a level of confidence in measurements already made. Calibration is for future measurements.

[jucole]

The certificate states:

As Found: In Tolerance
As left: In Tolerance
Adjustments: N/A

With adjustments stating N/A, not sure if it's not applicable because it wasn't needed or it wasn't part of the deal.

Regardless, a 0.2V deviation at 20V I believe I am correct as stating the device meter is not within tolerance, so I will contact Britest (is it Britest or Brltest?) and see what they say.  It was shipped only 100 miles and was very, very well packed so I doubt rough shipping caused it to drift.  I suspect the calibration was botched.

P.S. The current reading is equally out of tolerance.

I would definitely question the drift from spec. with them; also because it's a proper company and not a second-hand seller; it's their problem if the item shipped isn't within spec. when it reaches you.

[quantumvolt]

Or you can buy a 4 1/2 digit Voltmeter module on eBay for 10-20 USD (I got one, they claim informally around 0.3% accuracy, but mine is closer to 0.1% within the readings from an Agilent 34401A).

You mean 3 1/2 digit moduls, don't you?

http://www.ebay.co.uk/itm/5-Digit-0-36-red-LED-DC0-33-000V-3-wire-Digital-voltmeter-DC3-5-30V-powered-/350813582680?pt=LH_DefaultDomain_0&hash=item51ae1e7558

http://i01.i.aliimg.com/img/pb/870/750/558/558750870_396.jpg

[jancumps]

I would definitely question the drift from spec. with them; also because it's a proper company and not a second-hand seller; it's their problem if the item shipped isn't within spec. when it reaches you.

We are judging them by verifying their findings with measurements from uncalibrated equipment.  Even though multiple meters are used, their performance is not verified and 3 multimeters is too small a batch to give reliable probability that the calibration company is wrong.

[mikepa]

We are judging them by verifying their findings with measurements from uncalibrated equipment.  Even though multiple meters are used, their performance is not verified and 3 multimeters is too small a batch to give reliable probability that the calibration company is wrong.

Agreed the DMMs are uncalibrated and I was considering that at first.  But what are the chances that three high quality (well two are high quality, one is RS) agree to within .02 V that the supply is reading high by 0.2V?  I believe that probability is remote, even with only three samples.

[quantumvolt]

I am sorry to say so, but most of the posters here are talking about snow during the summer holidays.

This is the instrument in question or something very like it. All the OP complains about is that the Voltmeter MEASURES an UNKNOWN supposed to be 10.0 or 20.0 Volt voltage with a 'large' error.





His 'evidence'    is that he has 3 other meters showing at least 2 other values (within 20 mV of each other) for the same unknown voltage. So in fact the OP claims that based on an unknown value to be measured and at least 3 different measured values on 4 different voltmeters, the instrument under test is out of spec. The OP is probably right - but he does not seem to listen to the technical information given in this thread about linearity of digital meters.

Call Britest, tell them you think the meter is out of spec, and see what they say ...

[madires]

If you wanted a device spot on; you'd ask for cal & adj, and if required, a "repair". no?   

Basically if the OP power supply is outside the manufacturers spec. but when adjusted  then drifts faster than the manufacturer stated 8hr stability drift,  it will need a repair.

Doesn't that mean (if there was a log of old calibration data) that you loose calibration history as soon as you adjust?

It's not lost, but you need a second calibration just after the adjustment to be able to continue tracking any drifts.

[madires]

Calibration != Adjustment

Calibration == adjustment (or recording of correction factors to be used in subsequent measurements).

Measuring without adjustment is called performance verification.

Performance verification provides a level of confidence in measurements already made. Calibration is for future measurements.

Please see  http://www.npl.co.uk/reference/faqs/what-is-the-difference-between-calibration-and-adjustment-(faq-pressure)!

[quantumvolt]

At first I thought it must be my Fluke 179 that was out of calibration (it has never been calibrated), but as I have two other DMMs (Fluke 77 & one from RS) for a total of three that all read within .02 volts of each other, it makes me think the power supply is the one out out of calibration.

When you call them, please write down whether it is 0.02 or 0.002 Volts (as you state above). It might make a difference ...

[mikepa]

Thanks Quantumvolt for pointing out the error in one of my earlier posts, I meant to say the DMMs were within 0.02V of each other, not .002V.  I corrected that post.

While I concede that using un-calibrated DMMs does not provide conclusive evidence, I do believe that their very close agreement provides enough evidence to question the validity of the calibration.

I have contacted Britest and will see what they have to say.  If they won't entertain checking the calibration I will do it myself.

[Rufus]

Calibration != Adjustment

Calibration == adjustment (or recording of correction factors to be used in subsequent measurements).

Measuring without adjustment is called performance verification.

Performance verification provides a level of confidence in measurements already made. Calibration is for future measurements.

Please see  http://www.npl.co.uk/reference/faqs/what-is-the-difference-between-calibration-and-adjustment-(faq-pressure)!

One of many opinions on the ill defined difference between calibration, verification, and adjustment.

I would argue that calibration requires adjustment. For example when a 10g weight it calibrated and determined to be 9.999g that value can only be used to adjust measurements made using that weight as a reference or to at least adjust the claimed accuracy of measurements made using that weight. If that calibration value is not used to adjust something it has no value. Calibration without adjustment is a nonsense.

Many instruments provide the facility to store calibration results and automatically apply adjustments to future measurements using them. Storing calibration results in the instrument (usually without even seeing them) is still calibration and I consider adding or removing a bit of metal to/from a weight to be storing results and automatically applying them to future measurements.

Calibration and verification are used more or less interchangeably. I would argue that verification implies adjustment to measurements already made (you can't verify something that hasn't yet happened).

So I argue that calibration used to verify measurements already made does not involve adjustment to future measurements and should be called verification. Calibration to adjust future measurements should be called calibration and if the instrument has the facility to store calibration results and automatically make measurement adjustments that is a required part of the calibration.

Verification may show that no calibration is required.

[Bored@Work]

One of many opinions on the ill defined difference between calibration, verification, and adjustment.

The terms are well defined. E.g.  http://www.bipm.org/utils/common/documents/jcgm/JCGM_200_2012.pdf It is just that the definitions aren't properly taught and that "practitioners" pretty much hate the definitions and ignore them to cover up their sloppy behavior and practice.

Calibration without adjustment is a nonsense.

Well, if you say so.

[grumpydoc]

Calibration without adjustment is a nonsense.

No, not at all - in fact adjusting precision instruments can be a nightmare.

If it's a reasonably modern piece of equipment where the adjustment is done electronically it might not be too be, however it's best avoided if the instrument is in spec.

If it's a mechanical adjustment then you can get all sorts of mechanical settling/whiplash type effects which can make it extremely difficult to improve the accuracy, even it it starts out of spec.

I do tend to adjust stuff that I sell on ebay, if it's out of spec, but then I find myself running things for days afterwards to make sure they're not going to drift right back out of spec on me by the time the customer receives the item.

Even then, I can only reliably say that it was in spec when it left my hands as I'm sure a careless courier could bounce something which had trim pots or variable capacitors out of adjustment in all sorts of interesting ways once I have handed it over to their tender mercies.

[madires]

I would argue that calibration requires adjustment. For example when a 10g weight it calibrated and determined to be 9.999g that value can only be used to adjust measurements made using that weight as a reference or to at least adjust the claimed accuracy of measurements made using that weight. If that calibration value is not used to adjust something it has no value. Calibration without adjustment is a nonsense.

The calibration would tell you how the weight drifts over time and if it's within a defined range/spec. If the weight keeps beeing 9.999g over some years you know that you got a stable weigth. In case it drifts and exceeds the allowed range then you would adjust it or buy a new one. If it drifts a lot I would buy a new one anyway because it's a bad weigth, I can't trust it. A good weigth with no or very little drift is usabe much longer. Would you adjust a bad weight every year (paying for adjustment and a second calibration) or try to get a good one?

Many instruments provide the facility to store calibration results and automatically apply adjustments to future measurements using them. Storing calibration results in the instrument (usually without even seeing them) is still calibration and I consider adding or removing a bit of metal to/from a weight to be storing results and automatically applying them to future measurements.

I don't know if applying calibration results is already some kind of adjustment or not. If not, the instrument would require a special mode in which it doesn't apply the calibration results to be able to be re-calibrated, I guess.

[Rufus]

Calibration without adjustment is a nonsense.

No, not at all - in fact adjusting precision instruments can be a nightmare.

Adjusting measurements made with an instrument (for example dividing by 1.01 because calibration showed it read 1% high) is still adjustment. Adjusting the claimed accuracy or tolerance of a measurement made with the instrument because calibration showed it was 'out' is still an adjustment. If you don't use the results of a calibration to adjust something it doesn't matter what the results are and there was no point calibrating in the first place.

[Christe4nM]

One very important question: where exactly did you measure with your DMM's? If right at the binding posts, that's OK. But any test lead /wire connected will introduce a voltage drop. Your DMM leads' drop might be very very low while measuring voltages though. So if you measure at your load you will find a lower value than the PSU displays.

[off topic: it works the other way around too. So don't trouble yourself to set your PSU to 3.30 V precisely if you need 3.3 V for example. Measure at the load and adjust accordingly if needed, while keeping in mind that for general use the supply voltages for IC's can be +-5% or even +-10%.]

Also, I'd first do a worst case calculation to the values measured by my DMM's. Same for the PSU display. Then see what's left of the error.

I'm not saying there is nothing wrong with your instrument. Just make sure that you don't introduce errors yourself that might lead to false conclusions.

[mikepa]

Thanks Christe4nM, the leads to the DMMs were connected right at the binding posts using Pomona 4mm plug patch leads.

I have certainly been thinking, "why do I really care".  No circuit that I know of requires high precision voltages.  5.1V is as good as 5.0V.

[grumpydoc]

Adjusting measurements made with an instrument (for example dividing by 1.01 because calibration showed it read 1% high) is still adjustment. Adjusting the claimed accuracy or tolerance of a measurement made with the instrument because calibration showed it was 'out' is still an adjustment. If you don't use the results of a calibration to adjust something it doesn't matter what the results are and there was no point calibrating in the first place.

Surely that's compensation rather than adjustment.....

[SLJ]

I worked for a defense contractor for several years.  If the contract called for it (which it often did), test equipment would be calibrated to the spec called for in the contract or by the design engineers, used in production test, and then a performance evaluation would be run on the test equipment.  If the test equipment failed the performance evaluation after use, every product it tested would be pulled and the process would be started all over again.

[mikepa]

Good news.  BSLTesst agreed to calibrate the power supply and have returned it to me (they paid shipping both ways).  I'm pleased to report the unit is now "bang-on". 18.00V indicated shows 18.00 on my Fluke.  The current reading is equally accurate.  Thanks BSLTest.

[Stonent]

Good news.  BSLTesst agreed to calibrate the power supply and have returned it to me (they paid shipping both ways).  I'm pleased to report the unit is now "bang-on". 18.00V indicated shows 18.00 on my Fluke.  The current reading is equally accurate.  Thanks BSLTest.

 

That's pretty good service, do they list a website? Because I could not find one.  I'm curious as to what the cost for this calibration was.

[jucole]

Great result! ;-)

[mikepa]

That's pretty good service, do they list a website? Because I could not find one.  I'm curious as to what the cost for this calibration was.

http://esitest.com/

This was a calibration (adjustment) on a power supply I purchased from them on eBay, so there was no charge.