Can Niton analysis be this inaccurate?

[CollectForFun]

I have just come across this photo:

It seems to show a gold coin being analysed in what looks like a Niton machine and the display reads the coin contains 94.6% gold with 1.0% accuracy.

However, the coin is an Austrian 100 corona, which should be only .900 gold!

So the question is - can the machine be this inaccurate and overstating the gold content by that much? I know from @LawrenceChard videos that it must be periodically maintained and calibrated, but still... 

Or is there something "wrong" with the coin? 1915 year is a common modern restrike and undoubtedly fakes exist but with higher gold content..?

[LawrenceChard]

1 hour ago, CollectForFun said:

I have just come across this photo:

It seems to show a gold coin being analysed in what looks like a Niton machine and the display reads the coin contains 94.6% gold with 1.0% accuracy.

However, the coin is an Austrian 100 corona, which should be only .900 gold!

So the question is - can the machine be this inaccurate and overstating the gold content by that much? I know from @LawrenceChard videos that it must be periodically maintained and calibrated, but still... 

Or is there something "wrong" with the coin? 1915 year is a common modern restrike and undoubtedly fakes exist but with higher gold content..?

Look at the elapsed time 6.1 seconds.

I think Nitons are delivered with a default test time of 20 seconds, but this is from hazy memory.

We originally set our machine to 30 seconds, but then changed it to 60 seconds.

The longer the test time, the more reliable the statistical sampling, which should give more confidence that the result is accurate.

At the start of any test, the first reported figures are likely to be unreliable.

So, 6.1 seconds is not very long, and I would certainly have waited longer to see more trustworthy results.

Perhaps @ChardsCoinandBullionDealer should do a video demonstrating what I have just explained.

😎

[CollectForFun]

6 minutes ago, LawrenceChard said:

Look at the elapsed time 6.1 seconds.

I think Nitons are delivered with a default test time of 20 seconds, but this is from hazy memory.

We originally set our machine to 30 seconds, but then changed it to 60 seconds.

The longer the test time, the more reliable the statistical sampling, which should give more confidence that the result is accurate.

At the start of any test, the first reported figures are likely to be unreliable.

So, 6.1 seconds is not very long, and I would certainly have waited longer to see more trustworthy results.

Perhaps @ChardsCoinandBullionDealer should do a video demonstrating what I have just explained.

😎

Thanks, that's a great observation, I didn't notice the time elapsed indication. If it's normal that initial readings are off even by this much, that would solve this mystery.

Then the next mystery is why the seller of the coin thinks that providing photo of such a dubious analysis result helps his sale, but that's another story...

[LawrenceChard]

6 minutes ago, CollectForFun said:

Thanks, that's a great observation, I didn't notice the time elapsed indication. If it's normal that initial readings are off even by this much, that would solve this mystery.

Then the next mystery is why the seller of the coin thinks that providing photo of such a dubious analysis result helps his sale, but that's another story...

I simply don't take much notice of initial readings, for the above and other practical reasons. I am usually preparing a test result sheet, or checking other coins while I wait.

Sure, if after 10 seconds, you are seeing a very low reading, it is a good guess that the final result is going to be conclusive.

It does not surprise me that someone who owns or uses a piece of equipment turns out to be a dork, otherwise they would probably have developed their own judgement to a higher degree, and might not need the equipment in the first place.

I am sure that most of the scientists using the CERN Large Hadron Collider expect to see some random results when sampling rates or times are low, and it cost about $9 Billion!

But then, the CERN scientists are probably a lot brighter than whoever took the 100 Corona Niton photo.

😎

[dicker]

Dork.  An underused expression.  

[GoldDiggerDave]

YES!

[sovereignsteve]

Having been a scientist all my life with experience in all sorts of analyical techniques, I am intrigued and quite surprised by what @LawrenceChard says about the time aspect of XRF assay.

I have no experience of XRF but would be shocked if it took 10 seconds or more for the readings to settle down to their true levels.

12 hours ago, LawrenceChard said:

Perhaps @ChardsCoinandBullionDealer should do a video demonstrating what I have just explained.

I would be really interested to see some time plots as suggested here🤞

[LawrenceChard]

17 minutes ago, sovereignsteve said:

Having been a scientist all my life with experience in all sorts of analyical techniques, I am intrigued and quite surprised by what @LawrenceChard says about the time aspect of XRF assay.

I have no experience of XRF but would be shocked if it took 10 seconds or more for the readings to settle down to their true levels.

I would be really interested to see some time plots as suggested here🤞

Do bear in mind that a Niton XRF, or any  other test, may not produce a "true" level.

Consider the results shown here:

https://www.chards.co.uk/blog/krugerrand-gold-content/507

Reading	Gold	Maximum Gold Content	Copper
Obverse Centre	902 (+-3)	905	98 (+-1)
Obverse 3	905 (+-3)	908	95 (+-1)
Obverse 6	902 (+-3)	905	98 (+-1)
Obverse 9	902 (+-3)	905	98 (+-1)
Obverse 12	904 (+-4)	908	96 (+-1)
Reverse Centre	903 (+-3)	906	97 (+-1)
Reverse 3	904 (+-4)	908	96 (+-1)
Reverse 6	899 (+-3)	902	99 (+-1)
Reverse 9	904 (+-4)	908	94 (+-1)
Reverse 12	902 (+-3)	905	94 (+-1)
Total	9027	9060
Mean Average	902.7	906

Each of these tests was for 60 seconds, yet the mean average gold was 902.7 compared with a correct figure, as confirmed by cupelation (fire assay), of 916.6'; or an error of 

0.015236363636364 = 1.5236363636364%. This was over an effective test time of 600 seconds.

This was not a quirk of our machine, because Sheffield Assay Office got results very similar to ours using their own, different brand and model, of XRF machine.

While many or even most tests are indicating reasonably approximate figures after 10 seconds, I more or less disregard any readings before the first 10 seconds. I don't tend to stand and watch it, as I can be multi-tasking while I am waiting.

😎
 

[sovereignsteve]

19 minutes ago, LawrenceChard said:

Do bear in mind that a Niton XRF, or any  other test, may not produce a "true" level.

All analytical techniques have their own accuracy and precision profiles. Some of the former issues can be remedied by calibration but precision is more inherent in the scienctific principles and  technology employed.

The data you quote shows a clear bias but it is not clear if this was because of a calibration drift or some issue with the physical coin surface or other issue. The data is very precise.

It would be interesting to see a corresponding study employing various times of assay. Would the measured data vary around the final settled value? ie show a large CV, imprecision? or would you see a gradual rising or lowering of the reading until it settled?

[CollectForFun]

6 hours ago, LawrenceChard said:

Do bear in mind that a Niton XRF, or any  other test, may not produce a "true" level.

Consider the results shown here:

https://www.chards.co.uk/blog/krugerrand-gold-content/507

Thanks for the link, it must have been quite an exciting story as it developed. Perhaps just the final step was missing - contacting the manufacturer of your machine to find out what they think why their product failed with this particular coin...

[LawrenceChard]

2 hours ago, CollectForFun said:

Thanks for the link, it must have been quite an exciting story as it developed. Perhaps just the final step was missing - contacting the manufacturer of your machine to find out what they think why their product failed with this particular coin...

I did have a conversation with Niton UK, and will probably have more in the future.

As I was already aware that XRF testing is not as precise is some people might imagine, my principle concern at the time was whether our Krugerrand, and therefore others, were deficient, which would have been a big story. I was happy to conclude that our Krugerrand was OK, and only too happy to share our experience by publishing our experience.

It is not that their machine "failed", more a question that we had become accustomed to getting "closer" results, and for us the wider divergence, on that one coin, was surprising and unsettling.