We recommend Internet Explorer set to 1024x768.

© 2003 Brian F. Schreurs
Even we have a disclaimer.

So just how good is the approximation? Most people tend to think of this as a margin of error expressed as a percentage. Well, sometimes that's true, but in the world of science it's seldom that easy. What we call margin of error, science geeks call precision, and they measure it in terms of repeatability and reproducibility:

Now, a complicating factor is that when we look at oil degradation trends, we're not actually using an identical sample as required by both measures of precision: each test point is a new sample (to continue the analogy, your buddy is using a car nominally identical to yours, rather than actually using your car). We have no basis at all to expect precision within the repeatability range over the course of this study, so we will instead hope to see precision within the reproducibility range.

To get an idea of the real-world reproducibility of oil analysis, we subjected Blackstone Laboratories to a blind test. When we drew the 6,000-mile sample, we drew a second sample immediately following the first. We submitted the sample by way of accomplice Les Carnes, who -- lucky dog -- "owns" a 2003 Corvette made of vaporware. Les sent in the sample for us and forwarded the results. Blackstone was therefore "blind" to the true origins of the sample, ruling out any potential bias at the lab. The results were quite favorable, well within reproducibility ranges for most tests.

It's worth noting that this is not a perfect reproducibility test. Though the oil samples were very similar, there is no way to ensure that they were identical. The second sample came out of the pan several minutes after the first, giving the oil more time to settle and cool. The second sample also had to travel to Texas before going to the lab; in all there is a six-day difference between tests, a point mostly of importance to TBN, which can sometimes age unpredictably. And then there is simply the random distribution of particulates, which cannot be assumed to be a uniform mixture.

It's worth keeping in mind that these reproducibility expectations are the industry standard. It is entirely possible that Blackstone Laboratories uses equipment capable of even better margins of error, and when that experience is known, we have noted it in the comments here.

Particle counts are determined through spectrometric analysis. In this process, a spectrometer examines the wavelength of particles in the oil and determines their concentration in milligrams per kilogram (mg/Kg, or parts per million). The repeatability of this test is amazing, down to one or two mg/Kg for many elements, but the reproducibility is two to four times less precise. This is one area, in particular, where advancing technology could well improve the test's precision to levels greater than the industry standard.

The viscosity test is essentially a measure of the amount of time it takes a fluid to flow through a fancy viscosity-testing tube under constant environmental conditions. This test is very precise, with less than 1% variation in reproducibility, and a third of that in repeatability. The industry standard is to report this viscosity in centistokes (cSt), though some labs (such as Blackstone) report in Saybolt Universal Seconds (SUS). Luckily, there is a conversion to go from cSt to SUS:

Thanks to Matthew Hunt who was able to reverse this equation for us, so now you can use the calculator in either direction. It's pretty accurate, within about 0.05 units or so, as long as you keep the viscosity within reasonable levels for engine oil. It seems to work best for viscosities in the 2.5 cSt to 40 cSt range. For the exact conversion number, you'd need to look it up on the tables in ASTM D2161. Those tables aren't based on mathematic equations, but rather observed testing. But this calculator should be close enough to compare results from different labs.

Flashpoint testing is pretty neat, if imprecise. In essence, the sample is gradually heated and an open flame is periodically passed over the sample. When the vapors ignite from the flame, that's the flashpoint. Unfortunately its reproducibility is only within 32 degrees Fahrenheit, and even its repeatability is not all that great at 15 degrees Fahrenheit.

Fuel is derived from the flashpoint testing, and antifreeze is derived from the sodium and phosporus tests. Water and insolubles are both found via D-96, a method wherein the test material is subjected to abuse via centrifuge. Remember when you were six and you swung the cat by his tail? Yeah, similar idea, except less teeth and claws. ASTM hasn't yet determined the repeatability of this test.

Total base number (TBN) testing is particularly tricky. It sounds easy enough, with the premise of the test being that acid is added to the sample until it is neutral, but it must not be so simple. Having a look at the examples in the standards and in the industry literature reinforces the idea that finding TBN is no picnic. The reproducibility of TBN testing is rather imprecise: only within 1.91 TBN units using D-5984 -- Blackstone reports that their own experience with D-5984 is that it delivers results within 1 TBN unit. But there are two other common test methods for TBN. One of them, D-2896, has a reproducibility of 16.1%; using chlorobenzene in the test instead of mixed xylenes brings the reproducibility to 7%, but we doubt anyone would do that (exposure to chlorobenzene causes liver, kidney, and brain damage, whereas xylene just makes you dizzy). And another TBN test method that some labs use, D-4739, has a reproducibility of 21.1%. Ouch!

Overall, oil analysis presents a pretty good picture of what's going on in the engine, but it would be erroneous to take it as gospel. Oil analysis works best as an indicator of trends, and to get the most out of it, it's important to sample at regular intervals.