Tec Nikan
فارسی
Talk to us
All posts

Why Your OEE Number Is Wrong

The arithmetic is trivial and the definitions are where it breaks: the denominator, the ideal cycle time, the stop threshold and where quality is counted. Four choices that move the number thirty points.

OEEmanufacturing metricsoperationsdata qualitycontinuous improvement

Overall equipment effectiveness is availability multiplied by performance multiplied by quality. The arithmetic is trivial, which is why the number is so widely trusted and so frequently wrong. Everything that matters lives in the definitions underneath, and most plants have never written theirs down.

Start with the denominator, which is where the largest errors hide. Availability is run time divided by planned production time, and "planned" is doing enormous work. Excluding planned maintenance, breaks, changeovers and unscheduled shifts produces one number. Counting all calendar time produces another, twenty or thirty points lower. Neither is wrong, but only one of them is being compared against last month, and a plant that quietly moves a changeover from unplanned to planned improves its OEE without a single additional part leaving the line.

Then the ideal cycle time, which sets performance. Three sources are common and they disagree. The nameplate rate from the equipment supplier is optimistic and assumes a product mix you do not run. The best hour ever achieved is real but usually unrepeatable. The current standard is often whatever was negotiated to be achievable, which makes performance a measure of how conservative the standard is. Same line, same shift, three different OEEs — and the one that gets reported is the one that was easiest to defend.

Micro-stops are the classic hidden loss. A jam cleared in twenty seconds, forty times a shift, is twenty minutes of lost production that most reporting systems never see, because the stop is shorter than the threshold that triggers a downtime event. The signature is a line whose availability looks excellent and whose output does not match the maths. If your data comes from an operator writing on a sheet at the end of the shift, micro-stops are invisible by construction.

Speed loss hides in the opposite direction. A machine running at eighty per cent of rate all shift never triggers a downtime event at all; it just quietly produces less. That is exactly what the performance factor exists to catch, which it does only if the ideal cycle time is honest — an inflated standard converts a speed loss into apparent perfection.

Quality has its own ambiguity: measured at the machine or at final inspection, before rework or after. Counting a part as good when it leaves the station and scrapping it two operations later means the OEE of the first machine is flattering and the loss lands on somebody else's number. First-pass yield is almost always the more useful measure, and almost always the less popular one.

Reason codes deserve a mention because they are where the improvement value actually sits, and where data quality collapses fastest. A stop log in which sixty per cent of events are "other" or "miscellaneous" is not a dataset, it is a formality. Reason codes have to be short, mutually exclusive, chosen by someone who was there, and few enough to pick from a screen in five seconds.

The most damaging use of OEE is comparison between assets. A packaging line and a CNC cell have different loss structures, different standards and different denominators, and ranking them against each other produces league tables that reward definition-shopping rather than improvement. OEE is a trend instrument for one asset against its own history. The moment it becomes a target, the definitions start to move, which is Goodhart's law arriving on schedule.

The eighty-five per cent world-class figure deserves the same scepticism. It circulates widely, has no traceable derivation, and functions mainly as an argument for adjusting the denominator until the plant clears the bar.

What to do instead is dull and effective. Write the definitions down once — the denominator, the source of the ideal cycle time, the threshold below which a stop is not counted, and where quality is measured — and treat changing them as an event that resets the baseline. Take the data from the machine rather than from a person, because the losses that matter are the ones too short to write down. Report the loss breakdown, not the headline: the number that changes behaviour is "we lost fourteen minutes to infeed jams on Tuesday", not "we were at sixty-one per cent". And if you want to see the capacity you are not using at all, look at total effective equipment performance, which uses calendar time and therefore includes the shifts you do not run — a much less comfortable number, and often the one that answers the actual question.

A defensible fifty-five per cent that everyone understands is worth more than an eighty-five per cent nobody can reconstruct.

Want to work with us?

Tell us what you're building and we'll help you scope the first deployment.