OEE without illusions

Many plants report an OEE of 80-odd percent, and the line still does not deliver the planned quantity. The calculation is usually correct. The definitions are chosen for comfort. OEE can be inflated by 15 to 20 points with nothing changing on the line, just through what you call a “planned stop” and which cycle time you take as the reference.

The formula in brief

OEE multiplies three factors. Availability is the time the line ran divided by the planned production time. Performance compares what the line produced while running with what it would have produced at the ideal cycle time. Quality is the share of parts that are good first time, with no rework and no scrap.

The version that is hardest to fool is also the simplest: good parts multiplied by the ideal cycle time, divided by planned production time. If the two calculations do not give the same result, there is an error in your data.

How OEE gets inflated

The first method is to move stops into the “planned” category. Lunch breaks and the shift meeting can sit there, if the line really stands still. Changeovers, cleaning, customer trials and missing material cannot. These are losses you can reduce, and if you take them out of the calculation nobody sees them any more.

The second is an ideal cycle time chosen to match how the line runs today. The correct reference is the machine’s nominal speed or the best sustained rate it has demonstrated, not the average of the last few months. When someone slows the line down “to avoid scrap” and then updates the standard cycle time, the speed loss disappears from the report and stays in production.

The third is ignoring micro-stops. A 40-second jam at the feeder never reaches a log sheet, but repeated 60 times per shift it adds up to 40 minutes. If you do not measure them separately, they show up only as low performance and nobody knows where it comes from.

The fourth is averaging across several lines or the whole plant. One line at 85% and another at 45% give an average that describes neither. For decisions, what counts is the line that limits the flow.

The fifth is the 85% “world class” target. The figure has circulated for decades and takes no account of process type, number of part numbers or frequency of changeovers. An assembly line with 30 variants a day and an extrusion line that runs the same profile for weeks do not have the same ceiling. The useful target is your own level today, measured honestly, and the pace at which you improve it.

An example

The example below uses invented, round figures to show the mechanism.

Say a shift is 480 minutes, of which 30 are a lunch break, so 450 minutes of planned time. The line stood for 60 minutes: 30 for a changeover and 30 for defects and waiting for material. It ran for 390 minutes. The machine has a nominal speed of 60 parts an hour, one minute per part. It produced 300 parts, of which 285 were good.

The honest calculation gives availability of 390/450 = 86.7%, performance of 300/390 = 76.9% and quality of 285/300 = 95%. OEE comes out at 63.3%, the same result as 285 good parts times one minute, divided by 450.

Now the same shift with two “adjustments”: the changeover moves to planned, and the reference cycle time becomes 1.2 minutes, “because that is how it actually runs”. Planned time drops to 420 minutes, availability rises to 92.9%, performance to 92.3%, and OEE reaches 81.4%. The same 285 good parts, 18 points higher on paper.

Define it once, in writing

Before any dashboard, write the rules on one page. What counts as planned time (usually only breaks where the line stands still and periods with no order). What the ideal cycle time is for each part number and where it comes from. What a good part is (whether rework counts or not). From how many seconds a stop is logged with a reason, and below which threshold it counts as a micro-stop.

Production, quality and maintenance sign this page, and after that it does not change depending on how the month looks.

Start with one line

Pick the bottleneck line, the one that limits deliveries or where stock piles up in front of it. A minute gained there is a minute gained for the whole plant. A minute gained on a line that is waiting for material anyway changes nothing.

Measure for a few weeks on that line before you extend. You will find collection problems (operators filling in the sheet at the end of the shift, reasons picked at random, counters that also count setup parts), and it is easier to fix them on one line than on ten.

Stop reasons you can use

The list of reasons decides whether the data is good for anything. A few practical rules:

• Between 10 and 20 reasons at the first level, written in the operators’ own words. A list of 80 codes guarantees that “other” gets picked every time.
• Reasons describe what happened, not who is to blame: “material missing from warehouse”, not “logistics”.
• “Other” cannot exceed 5-10% of stop time. If it does, the list needs revising.
• Separate changeovers, equipment failures, waiting (material, operator, quality approval) and adjustments after start-up.
• If the line has a PLC, take stop times automatically and leave the operator only the choice of reason. Times written by hand at the end of a shift are rounded and incomplete.

Read the losses, not the percentage

The OEE percentage only tells you whether you have losses. What you do about them comes from the breakdown. Put the minutes lost by reason into a Pareto chart for the last week, separately for availability, speed and quality.

Look at minutes first, not at the number of events. Ten short jams can weigh less than a single two-hour wait after a quality adjustment. Then look at speed loss: if performance stays below 85-90% with no recorded stops, the cause is micro-stops or a line running below nominal speed, and you only see that by standing next to the line for an hour with a stopwatch.

In the daily meeting

Yesterday’s OEE on the bottleneck line gets two minutes of discussion, not twenty. The question is not “why did we get 64%” but “what were the top two losses in minutes and who is dealing with each”.

Each big loss gets an action with an owner and a deadline, and the next day the open actions are checked first. The four-week trend is displayed next to the line, with the same definitions from one month to the next. When someone proposes changing the definition to make the numbers look better, the answer is the page signed at the start.

If you have been through this, the discussion continues in the Production & Operations group.