How to Read a Fluid Cleanliness Code and Act On It

Why this matters

A cleanliness report comes back with three numbers on it and almost nobody does anything with them. The report gets filed, the elements get changed on the calendar, and the machine keeps eating pumps. The three numbers are not a grade. They are a readout of a balance between dirt going in and dirt coming out, and a single reading of a balance cannot tell you which side moved. This article turns a report you are holding into one of three moves, and it does that by putting the number next to a target and next to the previous sample before it lets you decide anything.

The sibling card on why contamination is the dominant failure cause owns the mechanism, the beta-ratio arithmetic and the sampling method. Read it once. This one starts after the bottle has gone to the lab.

Before you touch the number: was the sample valid

A sample drawn from a still, cold reservoir through the fill port describes the tank, not the fluid the pump is drinking. There is no correction that rescues it. A valid sample comes from a dedicated sampling valve on a live line, at operating temperature, after the machine has run long enough to turn the whole charge over, with several times the sample line's own volume flushed and discarded first.

While that warm-up run happens, the machine is cycling under its own controls with guards in place and nobody in the motion path, and the sampling valve is operated with nitrile gloves and eye protection because the fluid coming out is hot and the line is pressurized. Never crack a fitting or back off a plug to get fluid out of a pressurized circuit: a stream escaping a small opening injects through skin, which the card on fluid injection injury covers and which no glove prevents.

If the sample was not taken that way, the report is discarded rather than adjusted, and the arithmetic below never runs on it.

Step 1: Find the target the machine is actually held to

The three numbers are an ISO 4406 code counting particles per millilitre larger than 4, 6 and 14 micrometres. A code alone means nothing without a target, and the target belongs to the machine builder's specification for the pump and valve package, or to your service contract. ISO 4406 is a consensus standard, so it binds you through that specification or that contract, never on its own, and the edition matters because the 1999 revision changed both the size channels and the particle-counter calibration reference. Ask which edition the target was written under before you compare a modern report to an old one.

If nobody can produce a target, you do not have a pass or fail, you have a baseline. Say that out loud rather than inventing a threshold.

Step 2: Convert the gap to a factor, not a feeling

Each code number is a doubling of particle count, so the gap between your reading and the target is a factor of two raised to the difference. Three codes over is eight times the specified particle count, not three percent and not "a bit dirty". Do this per channel, because the three channels are three separate measurements and they can disagree.

Step 3: Decide whether the number is signal or noise

Write the basis and the character of the error before it enters any arithmetic. The uncertainty on a cleanliness code is a spread in code units, not a percent of reading and not a percent of span, and it is an independent random spread rather than a fixed offset, because it comes from where in the flow your bottle caught its millilitre and from counting statistics on a small volume. Treat about one code number as the working spread of a single sample unless your lab publishes tighter, and ask them.

Two consequences follow from that character, and they are different:

• A single sample's channel is read as roughly its stated code plus or minus about one code.
• A difference between two independent samples carries both spreads, and independent spreads combine in quadrature, so two equal terms multiply by the square root of two. About 1.4 code numbers of noise sit on any sample-to-sample comparison. A one-code move between visits is not a finding.

This is the step people skip, and skipping it is how a shop declares victory on a filter change that did nothing.

Step 4: Read the three channels against each other

The 4 and 6 micrometre channels are dominated by silt, the fine population that abrades clearances slowly. The 14 micrometre channel is the coarse population that jams a spool or blocks an orifice. Compare the gap on each:

• Fine channels well over target, coarse channel near it. Something is admitting fine particles and the filtration is handling the coarse end. Look at breathers, rod wipers and fill practice.
• All three over by about the same amount. A general ingression path or a charge that was already dirty when it went in.
• Coarse channel over, fine channels near target. Usually a component making its own debris, or a recent piece of service work that let something in through an open port.

Step 5: Pick the move the balance calls for

One rule, stated per channel, on a valid live-line sample, for one machine:

• Gap of 1 code or less on every channel. Inside the noise band from step 3. No move beyond holding the sampling cadence.
• Gap of 2 or 3 codes on any channel. Cut ingression first. A factor of 4 to 8 is an open door far more often than it is an undersized filter, and closing a door costs less than changing the filtration architecture.
• Gap of 4 codes or more on any channel, OR a gap that has not closed after the ingression paths are shut. Filtration capacity as well, and consider removing the particles already in the charge, because a filter only removes what passes through it and the existing population is already inside.

The Boolean on the last one is OR, not AND, deliberately: a machine that stays four codes over after you have shut every door you can find has a removal problem whether or not the doors were the original cause. Gain: one change at a time, then re-sample before the next. Changing a breather, an element and the fill practice in one visit tells you the number moved and never tells you which move did it, and next year on the next machine you will guess.

Worked example: three samples on one machine, then a fourth

A machine that has replaced two piston pumps in three years. The builder's specification calls for 18/16/13 for the pump and valve package, written to the current edition of ISO 4406, confirmed with the builder rather than copied off a sibling machine.

Sample A, discarded. Drawn cold with a dipper through the fill port: 22/20/17. Not corrected, not adjusted, not used. The correction line here is that there is no correction; a tank-bottom dip and a live-line sample measure different fluid.

Sample B, valid. Live sampling valve, machine hot after a full cycling run with guards in place, sample line flushed and the flush discarded: 21/19/15.

Gap per channel, and the factor from step 2:

The three-channel read from step 4. The fine channels are eight times over and the coarse channel is four times over, so the fine end is the one being outrun. That points at a door, not at a component shedding chunks.

The rule from step 5, applied as written. The largest gap on any channel is 3 codes, which lands in the middle band: ingression first, and only one change. Two doors were found on the walk-around, so this is where the gain rule bites. The breather cap was finger-tight over a housing with no element in it, and the reservoir was being topped up from an open pail. The breather went in first because it is the door that is open every time a cylinder retracts, and the fill practice was moved to a filtered transfer cart on the following visit.

Sample C, three weeks after the breather only. Same port, same method, same lab: 20/19/15.

The 4 micrometre channel moved 1.0 code. Against the noise band from step 3, a difference between two independent samples carries about 1.4 codes of spread from method alone. 1.0 observed against 1.4 of noise: no change demonstrated. Not "a small improvement", and not "trending the right way". Nothing was shown, and the honest report to the customer says so.

That is not a reason to undo the breather. It is a reason to keep going, because the second door was still open the whole three weeks.

Sample D, six weeks after the filtered fill practice went in. 17/16/13.

Change on the 4 micrometre channel from Sample B: 21 down to 17, which is 4 codes, a factor of 2^4 = 16 reduction in particle count. Against the same 1.4-code noise band on a difference, 4 codes is real by a wide margin. The 6 micrometre channel closed 3 codes and the 14 micrometre channel closed 2, and all three now sit at or below the specification's 18/16/13.

Why no element change was ever made. The rule said ingression first at a 3-code gap, the two doors closed the gap, and the machine never entered the band where filtration capacity is the move. A shop that had started with a filter upgrade would have spent the visit, seen the number improve some amount, and still had a missing breather element admitting room air on every retraction stroke.

The failure mode of doing this without the noise step. Sample C is the trap. A shop reading 21 down to 20 as a win closes the job, tells the customer the breather fixed it, and finds out at the next pump failure. The one-code move looked like progress and was indistinguishable from the same fluid measured twice.

How to verify you acted on the right thing

• Re-sample from the same port, the same way, at the same operating temperature. Whatever the method contributes it contributes to both samples, so a fixed component of it falls out of the difference. Change the port or the method and you have lost the comparison, not improved it.
• State the move you made and the change you expect before the bottle goes out. A prediction you wrote down is the only thing that separates a fix from a coincidence.
• Compare the change against 1.4 codes rather than against zero. Anything smaller is not evidence.
• Check that your comparator got the same treatment. If you corrected the sampling method between visits, the earlier number was taken the old way and cannot be benchmarked against the new one without saying so in the same breath.
• Close on the pump, not the report. The point of the code is component life. A machine at target that still fails pumps has a second problem, and the report has done its job by ruling this one out.

References

• ISO 4406 for fluid cleanliness coding, in the edition named by your machine builder's specification or your service contract, which is the instrument that binds you rather than any regulation
• Machine and component manufacturer documentation for the required cleanliness target, the working clearances and the element specification for each filter housing
• Safety data sheet for the specific hydraulic fluid, which governs skin and eye handling, spill response and disposal of the used sample
• See related: Why Contamination Is the Dominant Failure Cause in Hydraulics; What a Hydraulic Filter Catches and Why Its Location Matters; What Fluid Injection Injury Is and Why It Is a Surgical Emergency