Oil Analysis Limits Belong to the Compartment, Not the Lab

Why do labs analyse machines rather than samples?#

Used oil analysis labs do not analyse samples, they analyse machines. A single sample from a haul truck’s engine tells you almost nothing, because the amount of iron in used oil depends on the engine, its age, its duty, how long the oil has been in it and what the last sample showed. The same reading is unremarkable in one context and alarming in another. The information is in the run of samples from the same compartment over months and years, read as a trend, ending in a judgement about what is wrong and what to do.

That shape does not fit a system built around the sample. A sample-centred system shows one record at a time, compares each result to one limit, and has no idea that the record beside it came from the same compartment three hundred hours earlier. It holds the data well and judges none of it. The diagnostician does the joining in their head, or in a spreadsheet, and the value the lab sells is produced outside the system that stores its data.

So the unit of record in an oil lab is the compartment: an engine, a transmission, a hydraulic system, a final drive, a transformer’s main tank. Each has its own oil, its own sampling interval, its own history and its own limits. Everything below follows from taking that seriously.

Why should limits belong to the compartment?#

Because the compartments on one machine are different machines. A diesel engine, a hydraulic system and a final drive on the same haul truck carry different oils at different viscosities, run at different temperatures, wear different metals and fail in different ways. Iron in a final drive comes from gear teeth and is expected in quantities that would condemn an engine. Copper in a hydraulic system may be a cooler core dissolving; in an engine it may be a bearing. A lab-wide table with one caution and one critical limit per element gets one of those compartments wrong every time.

The extreme case is the transformer. A load tap changer switches current under oil and arcs by design, so its oil carries gases that would take a main tank out of service. The same acetylene reading is normal in one compartment and critical in the other, on the same transformer. There is no limit that is right for both, and a lab that applies one produces either false alarms on every tap changer or silence on a failing main tank.

Limits should therefore be set per compartment type at minimum, and refined per compartment as its history accumulates. Published guidance on statistically deriving alarm limits from a population of similar compartments is the right starting point, and the limit set should be a record the lab can edit, so that lowering the caution limit for iron on the hydraulic set takes effect on the next reading without a software change. The limits are the lab’s knowledge about its customers’ machines, and they should live where the lab can maintain them.

What does "inside the limit and moving" mean?#

A limit table only ever tells you about a compartment that has already crossed a line. By the time iron breaches the critical limit, the bearing has been shedding metal for two or three samples, and each of those samples was inside every limit and looked normal on a table. The failure was predictable and the table did not predict it.

So a second rule is needed beside the limits: a reading that is close to its caution limit and has risen sharply against the comparable earlier sample is marginal, even though it has breached nothing. Marginal means inside the limit and moving. It is the rate-of-change rule, and it is the one that catches a failure early enough to schedule a repair rather than recover from a breakdown. Where a low reading is the fault, as with an additive being depleted, the same rule runs the other way: near the floor and falling.

A flag raised by this rule should say so in words: which reading, how close to which limit, and which earlier sample it was measured against. A flag that cannot be explained is a flag the diagnostician learns to ignore, and once the flags are ignored the screen is decoration.

Why must the trend restart at an oil change?#

Wear metals accumulate in the oil. They climb through an oil’s life and drop to near nothing when the oil is drained and replaced. So a compartment’s raw trend is a sawtooth, not a ramp, and a naive comparison reads every drain as an improvement and every fresh fill as the machine deteriorating again. The machine did not get better because the oil was changed.

The history therefore has to be cut into oil fills, and the comparable earlier sample means one of two things. If there is a previous sample in the same fill, compare against that: same oil, further into its life. If this is the first sample of a fresh fill, compare against the sample at the nearest oil age in the previous fill, so that a reading at two hundred hours on new oil is judged against the reading at two hundred hours on the last oil, not against the gassy, metal-laden sample taken just before the drain. A new fill starts where the oil was renewed or where its recorded age has gone down; either is proof that the fluid is not the fluid that was in it last time.

Charts should follow the same rule. The trend line should break wherever the oil age falls, so a drain is never drawn as a downward slope. Where a compartment is sampled at every oil change, every sample sits at the same oil age, the readings are directly comparable and the line runs straight through, and the chart should say which of the two it is showing. For transformers, where oil is degassed rather than drained, the same applies: a trend read straight through a degassing job says the fault cured itself, the most expensive misreading of a transformer.

Should the report name the fault or list the numbers?#

Name the fault. A report that says iron, chromium and the ferrous debris index all breached is a restatement of the table. A report that says a bearing or gear is breaking up, and that the filter should be cut, is what a diagnostician is paid for. Faults show up as combinations of readings moving together, and the diagnosis is recognising the combination.

Where two patterns match one sample, the report should name one primary fault and give the other a sentence. Where the second pattern is a known consequence of the first, it should be worded as corroboration rather than as a second fault. Coolant attacks bearing overlay and eats the additive package; saying “two faults” there would send somebody looking for a second cause that does not exist. And where nothing matches, the report should say plainly what breached and by how much, and claim nothing more.

Who gets the last word?#

The analyst. Everything above is arithmetic and pattern-matching, and it should be presented as a proposal: a severity the software suggests, a diagnosis it drafts, drawn on the screen as proposed rather than as decided. The diagnostician reads it against what they know about the customer, the site and the machine’s history, and releases the sample as whatever severity they judge right. Nothing should be released without a person choosing to release it.

When the analyst overrules the proposal, the record should say so. A released sample whose severity differs from what the software proposed should carry a sentence noting the difference, so that nobody reading it later has to compare two fields to find out. Over time that record is the most useful thing the lab has: it shows where the limits are wrong, where the patterns miss, and where the analyst’s judgement is consistently better than the arithmetic, which is exactly what should drive the next revision of the limit sets.

The same respect applies to the smaller fields. If the software fills in a resample interval from the severity and the analyst types their own, the software should stop overwriting it. A screen that overwrites a number somebody has typed is a screen nobody should trust. A reissued diagnosis should keep the earlier one readable, with who changed it and why. The customer’s own view of the report should show the released diagnosis and nothing of the proposal, since the proposal was never the lab’s opinion.

What an oil analysis lab carries in this work#

LabLynx has served used oil analysis and analytical laboratories for long enough to have seen a lab-wide limit table produce both kinds of error on the same customer’s fleet. The review workspace described below is the shape we build toward, and the boundary is stated plainly.