An OEE Week on One Machine: 40 Hours In, 65.5% Out

Overall equipment effectiveness is one number built from three, and the only way to use it is to read the three. This walkthrough runs a single CNC machine through one ordinary week in EDGEBIC by User Solutions: 40 hours of shift capacity, one day lost, 110 pieces off the machine, and a final figure of 65.5%. Every number below comes from the same report you can open on your own plant, and the point of the exercise is not the 65.5%. It is knowing which of the three factors is worth a conversation on Monday.

If you want the definition first, read what overall equipment effectiveness is. If you want the buttons, running the OEE report covers them. This post is the arithmetic, done slowly, on one machine, and it sits alongside the rest of the EDGEBIC worked examples.

The Machine and the Week

CNC-1 is a single machine: one instance, one day shift running 08:00 to 16:00, Monday through Friday. It spent the week on one product, and the routing step it runs carries a cycle time the scheduler already uses to plan with.

Piece counts for the week, taken from the shop floor: 110 pieces came off the machine, of which 105 passed first time, 3 were scrapped, and 2 went back for rework.

That is the entire input. Three factors come out of it.

Availability: 32 Divided by 40

Availability asks how much of the time the machine was theoretically open did it actually run.

Available hours = 8 h x 5 days x 1 instance = 40 h
Actual hours    = 32 h
Availability    = 32 / 40 x 100 = 80%

Two things matter about that top line. First, the available hours come from the machine's own shift calendar, not from a flat assumption: a work center with two shifts, three instances, or a plant holiday inside the window gets a different denominator, computed from its real calendar. Second, availability is capped at 100%, so a machine that logs more hours than its calendar says it had reads as 100% rather than 120%. If you see a capped machine, the interesting question is why the calendar disagrees with the floor, and that belongs in the capacity resolution conversation rather than in OEE.

Eighty percent is the week's single largest loss, and it is one whole day.

Performance: 110 Pieces Against 128

Performance asks how fast the machine ran while it was running. The comparison is against the routing's own cycle time, which is the same number the scheduler used to plan the job in the first place.

The step carries 2 pieces per unit and 0.5 hours per unit, and the work center's efficiency factor is 1.0. That gives a theoretical rate of 4 pieces per hour:

Theoretical rate   = (2 pieces / 0.5 h) x 1.0 = 4 pieces per hour
Theoretical pieces = 32 h x 4 = 128 pieces
Actual pieces      = 110
Performance        = 110 / 128 x 100 = 85.9%

Note the denominator. Theoretical pieces are computed from the hours the machine actually ran, not from the 40 hours it might have run. That is what keeps the factors independent: the lost Wednesday is already charged to availability, and charging it again here would double count it. Performance is a clean question about speed during the hours worked.

Performance is also capped at 100%. If your routing cycle time is generous, every machine will read 100% and the factor stops telling you anything, which is a quiet argument for keeping routing hours honest.

When a schedule has no routing cycle time to work from, performance falls back to actual hours over planned hours. That fallback is worth knowing about, because it silently changes the question from "how fast did the machine run" to "how close was the plan", and the two answers can point in opposite directions.

Quality: 105 Good Out of 110

Quality asks how much of what came off the machine was right the first time.

Good   = 105
Scrap  = 3
Rework = 2
Quality = 105 / (105 + 3 + 2) x 100 = 95.5%

This factor is the one that depends entirely on the shop floor. Good, scrap, and rework counts come from operator punches, and at the kiosk a scrap piece can never be recorded without a reason: the operator picks a category (machine, material, quality, waiting) and then the specific reason. That requirement is what turns three scrapped pieces into a Pareto you can act on rather than a number you can only regret. The mechanics are in standing up the kiosk and logging a first shift.

With no punches at all, quality reads n/a. It does not read 100%.

The Product: 65.5%

OEE = 80 x 85.9 x 95.5 / 10,000 = 65.5%

Here is the part that surprises people the first time. Nothing in this week was a disaster. The machine ran four days out of five, held 86% of its planned speed, and produced 95.5% good parts. Any one of those numbers read alone sounds like a decent week. Multiplied, they land at 65.5%, against a world class benchmark of 85%.

That is not the report being harsh. It is the arithmetic of three sequential filters, and it is exactly why OEE is worth computing: the compounding is invisible until you do the multiplication.

Which Factor to Act On

Read the gaps, not the product:

The lost Wednesday is worth more than four times the scrap. If you spent the week chasing three bad parts, you spent the week on the smallest of your three problems. The conversation this report earns is about why the machine was down a full day and whether that day was planned maintenance the schedule already knew about, a breakdown nobody logged, or a day with no work queued.

Each of those has a different answer. Planned maintenance belongs on the calendar, where it reduces available hours honestly and stops depressing the availability figure for something you chose. A breakdown belongs in the record so the pattern is visible next quarter. An empty day is a scheduling question, and the work center utilization report is the surface that shows whether the machine was starved while others were overloaded.

OEE Is Not Utilization, and the Difference Matters

Two reports in the plant look like they are measuring the same thing, and they are not.

Work center utilization answers a capacity question: how much of this machine's real calendar hours are booked, computed from actual shift calendars, holidays, and instances. It is the number to argue about capital equipment with, and it is the authority when two reports disagree about how loaded a machine is.

OEE answers an effectiveness question: of the time the machine did run, how much of it produced good parts at the intended rate. It says nothing about whether the machine was booked or starved.

A machine can sit at 100% utilization and 55% OEE, which means it is fully committed and running badly, and buying a second one will fully commit a second machine to running badly. The reverse also happens: 60% utilization with 90% OEE means the machine works well when it works, and the problem is upstream in the schedule. Those are opposite decisions, and only reading both numbers tells you which one you are in.

Running It on Your Own Plant

The report is per work center and covers whatever date range you set when you open it. Rows with a computable OEE sort first, descending, and the n/a rows collect at the bottom, so a plant part way through adopting piece counting still gets a usable ranking of the machines it does count.

Two habits make it worth opening:

Keep the range short and repeated. One week, read the same way each Monday, tells you more than one quarter read once. The lost day in this example is invisible in a 90 day average.

Use the Column Details button when a number is disputed. Every report window carries it, and it gives each column's plain description, its formula, its unit, and example values with one-line interpretations. It is faster than an argument and it leaves both people reading the number the same way next month.

What Makes the Number Unusable

Three practical failures account for most disappointing OEE reports.

No piece counts. Quality reads n/a, so OEE reads n/a, and you are left with two factors out of three. Availability still works from logged hours alone, so this is not a reason to skip the report; it is a reason to start counting on your most suspect machine.

Routing hours nobody has revisited. Performance is measured against the routing cycle time. If that cycle time was set optimistically at implementation and never corrected, performance reads low forever and tells you nothing new each week. If it was set generously, performance pins at 100% and tells you nothing at all.

Hours logged in batches. Availability compares logged hours to calendar hours day by day. A week logged in one entry on Friday still totals correctly, but you lose the ability to see which day was lost, which is the whole diagnostic value of the exercise.

What to Do With It on Monday

Run the report on a range you can remember, one week rather than one quarter, on a handful of machines rather than all of them. Read the three factors before the product. Pick the largest gap and ask one question about it. Then use the Column Details button in the report window if any column's definition is in dispute: it carries the formula, the unit, and example values for each column, so two people reading the same number read it the same way.

OEE is not a scorecard to publish. It is a way of deciding, in about a minute, whether this machine's problem this week was time, speed, or scrap. On CNC-1 it was time, and the answer was one day.