Repeatability of Oil Analysis Testing - Same Lab and Two Different Techs

[JohnBrowning]

If you are measureing wear in "PPM" then 1%-10% is not a significant difference in repeatability. This is especialy true when you look at the fact that any wear metal amounts below 150 PPM are consider fine by most large domestic auto makers. SO what is an error rate of +/- .35% to 10%? It is not enough to affect the over all trend!

If you think this is bad far more serious decisions are made about peoples health with either less information then this, more variation in out come or a total lack of trending based on so called established norms!

What other protocal will detect glycol even when itis not visable int the oil? WHat other tests will reveal an extreme parts failure before it completly fails?

 

[XS650]

quote:

---

Originally posted by JohnBrowning:
If you are measureing wear in "PPM" then 1%-10% is not a significant difference in repeatability. This is especialy true when you look at the fact that any wear metal amounts below 150 PPM are consider fine by most large domestic auto makers. SO what is an error rate of +/- .35% to 10%? It is not enough to affect the over all trend!

---

The example I gave, 10 ppm vs 20 ppm was a 100% error. You can't establish a trend with that type of garbage data.

 

[Ray H]

I agree with XS650 that Russ's posts are thought provoking. While Russ does go out of his way to tweak the established party line, there's nevertheless a knee jerk over-reaction to his posts. Russ isn't a threat unless the vested interests insist on making him one.

"We have met the enemy and he is us" -Pogo (in collaboration with Walt Kelley)

[ February 17, 2004, 11:02 AM: Message edited by: Ray H ]

 

[Colorado]

The only values that are of any accurate significance are the % of dilution(water&fuel) and the TBN. All of the wear metals are in the noise floor,as Russ said. Also,the high ppm values such as calcium , boron, zinc etc, are out of the noise floor and can be accurate enough for a trend. So with that info in mind, the only value a UOA is to a consumer is to show if there are any contaminants in the oil(water etc.), How well the oil is holding up(TBN), and quality of the detergent and anti-wear package(calcium,zinc). It has no value for showing wear, unless the wear metal values are high(over 100 or 200 ppm)

 

[MNgopher]

So as I see it, the elemental values can be +/- 10% and we have folks feeling like that is in the "noise".

Say for 10 ppm (true value), then low would be 9ppm, and high would be 11ppm.

So if one analysis showed 9ppm, and another showed 11 ppm, should I draw big conclusions from the difference?

No, because it is in the "noise".

If one test shows 10 ppm and the other shows 20 ppm, then there is at minimum a 7 ppm difference (11ppm high and 18 ppm low), which is enough to start drawing some conclusions, as it IS out of the "noise" if you will.

Of course, this is all predicated on each sample being identical, which we can't say based on our sampling techniques.

Does it mean the numbers are useless? Hardly. If they are consistent or trending one way or another, then conclusions can be drawn. A one time snapshot offers other variability.

This also explains why the "big" numbres like P and Ca can vary so much. Say true is 1000 ppm, then low is 900 and high is 1100. Big range!

 

[jason07-1zzfe]

quote:

---

Originally posted by XS650:
The example I gave, 10 ppm vs 20 ppm was a 100% error. You can't establish a trend with that type of garbage data.

---

But you won't get that kind of variation.

I'd have to agree w/ John.

 

[paco]

I would imagine that the testing methods are validated. If so, almost surely some of the validation criteria would include intermediate precision and robust. The acceptance criteria would be no more than 5% and most likely NMT 1%.

Therefore, the variablity between running the tests on different days, after calibration of the equipment, between technicians etc. should be next to zero. However, between labs and that's a story altogether different as each lab probably has it's own test method and the equipment is almost surely not the same make and model.

Unless you are able to get your hands on the actual test methods or are able to assess how good of methods it is (reproducability, linearity, specificity etc) - there will alway's be an unknown element as to how accurate your results are comparable they are to other lab's.

Simply having a +/- 2% isn't sufficient.

Paco

 

[headfirst]

If there is a variance between different labs for the same sample one would need to question the accuracy of both labs. As has been stated before, accuracy and precision are not mutually exclusive. Results in general should be fairly repeatable, if not one would question the test descipline.

 

[Schmoe]

One thing we do for EPA labs is send out blank, dupicates and spiked samples. If some folks here are worried about laboratory precision, send out duplicates and spikes.

 

[edhackett]

Here's an example of what a good lab and chemist can do. In this case it's me.

This is the Mo data from an EPA sample that I ran to establish the repeatability of our new ICP/MS. The sample was run 5 times over a 33 day period.

Data is in parts per billion. The true value as given by the EPA is 437 ppb. My results were: 432, 437, 436, 436, 436.

Before you ask, no I don't do oil anyalysis. I'm not set up for organics, and I wouldn't let something as yucky as used motor oil into my lab. Most of my work is in the part per trillion range, occasionally chasing U to the high part per quadrillion range.

If I were to do an oil analysis with a guaranteed 2% precision and accuracy, it would run you about $800 just for the wear metals.

Ed

P.S. the acceptance limits for EPA drinking water certification is between 375 and 500 ppb. Pretty loose by my standards.

[ February 18, 2004, 03:45 PM: Message edited by: edhackett ]

 

[mrhonda]

As a rule of thumb, your 'gauge' (whatever it may be) should have 10X the resolution of what you are trying to measure. So, in this case values are being reported in PPM (parts per million). The equipment being used to gather those values 'should' have resolution to PPB (parts per billion).

 

[satterfi]

mrhonda

I would rephrase that to this:

"As a rule of thumb, you should not report more significant digits than you are certain of."

The $25 UOA is the worst offender. I especially like Wear Check numbers. 0.1 resolution on wear metals with a +-3ppm uncertainty?

No offense to the Wear Check customers, but you'd have to be an idiot to think you are getting that kind of accuracy for $25.

 

[XS650]

quote:

---

Originally posted by satterfi:
mrhonda

The $25 UOA is the worst offender. I especially like Wear Check numbers. 0.1 resolution on wear metals with a +-3ppm uncertainty?

---

Looks like their results are reported to two insignificant figures.

 

[Doug Hillary]

Hi,
if you care to look under Diesel UOAs "Detroit Diesel........-8" thread you will see two different Lab test results and the variables

There was distance and time between the too UOA's

I would appreciate your comments

Regards

 

[jsharp]

quote:

---

Originally posted by XS650:

quote:

---

Originally posted by buster:
Question I have is how many millions of parts does it take to wear out an engine?

---

If your engine has millions of parts, then it is either a German engine or a Jaguar engine

---

Well, it might be a Japaneses engine. But with it, you'd at least be able to identify all the parts, determine their function, and every one would have "Made in Japan" cast into it...