A gearbox on a steel mill runout table started making a low rumble after 11 months in service. It was a 45 kW bevel-helical unit with 480 L of EP220 oil. The maintenance supervisor sent a sample to the lab. The lab report came back with a particle count of 19/17/14 (ISO 4406). The report flagged it as “elevated.” The supervisor looked at the number, decided it was marginal, and scheduled an oil change. Nothing else. Three weeks later the gearbox failed. The input pinion had pitting across 60% of the tooth face. The lab result had been a warning, not a suggestion.
The frustrating part: a particle count of 19/17/14 is not an alarm in most textbooks. Many published guidelines treat 19/17/14 as “monitor, change at next opportunity.” For a normal gearbox in a normal plant, that’s true. For a bevel-helical unit running at 1450 RPM with a 30-second duty cycle, it’s too late. The gearbox failed three weeks after that reading. The oil change didn’t fix anything because the pitting had already started. The particles in the oil were the evidence, and the evidence was read wrong.
What particle count measures, and what it misses
ISO 4406 particle count reports the number of particles larger than 4 μm, 6 μm, and 14 μm, expressed as range codes. A reading of 19/17/14 means roughly 2500 particles per mL larger than 4 μm, 630 per mL larger than 6 μm, and 80 per mL larger than 14 μm. Those are ballpark numbers from the code table, not exact counts. The test counts particles, but it cannot tell you what the particles are made of.
That last point is the one that gets plants into trouble. A particle count of 19/17/14 caused by ingressed dirt looks exactly the same on the report as the same count caused by gear wear. Same codes. Same “elevated” flag. But the response is different. Dirt calls for a filter change and a breather inspection. Gear wear calls for opening the inspection cover, looking at the teeth, and making a judgment about whether the unit survives until the next scheduled shutdown. The lab report alone cannot make that distinction. Someone has to look at the ferrous content, or look at the teeth.
The steel mill gearbox failure was textbook. The 14 μm count of 80/mL was borderline. The lab also reported iron at 220 ppm, which is high for a running-in unit. Nobody connected the two numbers. Iron at 220 ppm is not dirt. It is the gearbox eating itself. A dirt-ingressed system shows silicon and aluminum from the sand, not iron. The iron number was the clue. It got buried in a column of other elements.
The sampling ritual matters more than the lab
A lot of the scatter in particle count results comes from how the sample is taken, not from the oil. The steel mill sample was drawn from the drain port after the unit had been sitting idle for a weekend. The heavy particles had settled to the bottom of the sump. The sample was mostly bottom sludge. It read high, but the high reading was a sump snapshot, not the circulating oil state. The lab flagged it elevated; the truth was somewhere between “fine” and “already failing.” A sample drawn from a live return line, or from a dedicated sample port in the recirculating loop, gives a number that means something.
Rules that actually hold up:
- Sample while the unit runs, from a port in the return line or the recirculation loop, not from the drain.
- Flush the sample port for 10 seconds before filling the bottle. The port holds stagnant oil from the last run.
- Take the sample in the first hour of operation after a weekend, when the oil is warm and mixed, not cold and settled.
- Send the sample the same day. Particles settle in the bottle, and a week of sitting changes the count.
None of this is exotic. It is the difference between a number you can act on and a number you file.
Reading the trend instead of the single point
One sample is a snapshot. The same gearbox sampled quarterly gives a trend. The 19/17/14 that killed the steel mill unit might have been 16/14/11 six months earlier. The doubling of the 14 μm count between two samples is the alarm, not the absolute value. Oil analysis programs that work are built on trends. Plants that chase single-point numbers either over-maintain (change good oil because one reading looks high) or under-maintain (see a borderline number and wait).
The practical threshold for a bevel-helical unit under continuous duty:
| ISO 4406 code | What to do |
|---|---|
| 16/14/11 or lower | Normal for a well-sealed unit. Keep sampling. |
| 18/16/13 | Check breather and seals. Change filter if pressure drop rose. |
| 20/18/15 | Sampling interval down to monthly. Look for the ingress path. |
| 22/20/17 | Do not run to the next scheduled stop. Find the source now. |
If the iron or copper reading is rising at the same time as the particle code, treat the combined signal as a wear event, not a contamination event. That combination was exactly what the steel mill report showed and nobody read.
The oil change that should have been a teardown
The supervisor’s decision to change the oil was not unreasonable. Oil change is the cheapest intervention available. But the particle count did not drop after the change, because the particles were coming from the gear mesh, not from old oil. A post-change sample two weeks later would have been 18/16/14 again. That rising-again-after-change signature is the classic indicator of active wear. The unit should have been inspected at the next available window. It wasn’t. The pitting reached the point of tooth breakage risk, the unit started rumbling, and the runout table went down mid-shift.
Cost breakdown from that failure, as the plant actually recorded it:
- Replacement bevel-helical unit: $9,400, five-week lead time
- Removal and installation: two shifts, $2,100 labor
- Lost production on the runout table: 11 hours, valued by the plant at $6,300
- Total: about $17,800
The oil analysis program cost $180 per sample. Two samples a year on that unit: $360. The same failure would have been caught by a trend-based reading and a scheduled inspection at the next maintenance window. The inspection itself was free — it is a cover removal and a flashlight.
When particle count is not enough
Particle count misses the particles that do the most damage when they are small and the count is low. A 5 μm particle does not show up in a 14 μm count. In a precision gearbox with a tight mesh, 5 μm particles accelerate wear steadily even at a “clean” 17/15/12. Ferrography or direct-reading ferrous monitoring catches this. For critical units, the extra $60 per sample for a ferrous wear index is cheap insurance. For the steel mill unit, the iron number was already in the report. Nobody looked at it.
The other blind spot is water. Particle count says nothing about moisture. A unit that ingests water through a cracked breather shows a normal particle count while the bearings rust. Water should be checked quarterly with a crackle test or a Karl Fischer titration. On gearboxes in outdoor or washdown areas, it is the first thing to check, not the last.
Oil analysis is a monitoring tool, not a maintenance decision by itself. The steel mill failure was not an oil analysis failure. The program produced the right data. The plant read the wrong column and acted a month too late. That distinction matters, because the fix is not a better lab. It is a person who knows what the numbers mean together.
Particle count tells you how many, not what kind. Iron rising with the count means wear, not dirt. Sample from a live line, follow the trend, and treat a count that climbs back after an oil change as a teardown signal. The gearbox that died at 19/17/14 had been telling the plant for three months.
