Most maintenance departments have started a bearing failure log at some point. Most of those logs died within six months, somewhere between the third week of production pressure and the realisation that nobody had looked at the entries since they started making them. The design that survives is not the comprehensive one; it is the one that takes two minutes to fill in and produces something visibly useful once a month. Everything else about it follows from those two constraints.
Why the thorough version always fails
A form with twenty-five fields is a form that gets skipped when the line is down and the shift is short. The technician standing at the machine has a job to finish, and the log is competing with that job for the same fifteen minutes. Meanwhile the entries that do get made pile up in a spreadsheet nobody opens, which means the person filling it in never sees anything come back from the effort. Both failure modes have the same root: the log was designed around what would be nice to know rather than around what someone will actually do under pressure.
Eight fields, and no more
Date. Equipment identifier. Position on the equipment. Bearing designation. Running hours or an estimate since installation. Observed failure mode, chosen from a short list. What was done. What went back in, brand and specification — which for most plants means something from the deep groove or housed unit ranges. That is enough to support every analysis worth doing, and it fits on a phone screen. Adding a ninth field is easy and each one costs you completion rate; the discipline is in refusing them, because a log that captures eight fields on ninety percent of failures is worth far more than one that captures twenty on a third.
Give them a list, not a text box
Free-text failure descriptions produce data that cannot be aggregated — one technician writes “worn out”, another writes “raceway damage”, and no analysis can tell whether those were the same thing. A fixed list of eight options solves it: fatigue spalling, contamination, corrosion, lubrication failure, misalignment wear, overload or cage failure, electrical arcing, unknown. Keeping “unknown” on the list matters more than it seems, because without it people guess, and a log full of confident guesses is worse than one that admits uncertainty. Training the distinctions is a separate job, and our visual guide to failure modes is the reference to put in front of technicians who are learning to tell them apart.
Two minutes on a phone, with a photo
Entry has to happen at the machine, not back at a desk where it will not happen at all. A simple mobile form, voice input for anything that needs describing, and a photo attached — the photo alone frequently answers questions the fields did not anticipate, and it makes later failure analysis possible without anyone having kept the component. If the entry takes longer than the walk back to the workshop, it will lose to the walk back to the workshop.
The monthly review is what keeps it alive
Once a month, someone looks at the accumulated entries and answers four questions: which failure modes recur, which positions fail most, which specifications underperform, and is anything trending that was not before. Then they tell the team what was found. That last step is the one that gets skipped and the one that determines whether the log survives, because it is the only moment when the people entering data see why they bothered. Half an hour a month, shared in five minutes at a shift meeting, is the entire maintenance cost of the system.
Findings have to turn into assigned work
A pattern identified and not acted on trains everyone to ignore the next one. Each review should produce specific actions with a name and a date attached — revise a lubrication interval, check a housing bore, change a specification on a position, retrain on an installation procedure. Then the following month, check whether the failure frequency on that position actually moved. The closed loop is what converts the log from a record into a tool, and it is also what produces the evidence you need for a supplier conversation when a specification genuinely is the problem. Recurring failures in the same place are rarely a bearing quality issue, as the cases in our piece on rolling mill reliability illustrate.
Where the accumulated data pays off
After a year the log answers questions that were unanswerable before. Whether the premium specification on a given position actually delivered the service life it promised. Whether a lubrication change helped. Whether one machine is genuinely worse than its identical neighbour, and if so what is different about it. It also gives you something concrete to put in front of application engineering at SKF, FAG or TIMKEN when a pattern needs manufacturer input — and a supplier presented with twelve months of structured failure data responds very differently from one presented with a complaint.
Setting up failure tracking that your team will actually use? Our team supports European plants on reliability programmes and failure analysis. Book a free consultation.