One number for how well a machine is used
Overall equipment effectiveness, usually shortened to OEE, is a widely used way to describe how much of a machine’s possible productive time is actually spent making good products at full speed. It combines three questions into one percentage:
- Was the machine running when it was supposed to be? (availability)
- When it ran, did it run at the speed it should? (performance)
- Of what it made, how much was good? (quality)
Multiplying the three gives OEE. The idea is that losses show up in each factor, and the product reveals how much potential output was lost overall. The calculation is simple, and IoT data makes it practical to compute automatically rather than from paper logs.
The three factors
Availability is the share of planned production time that the machine actually ran.
Availability = run time divided by planned production time.
Planned production time excludes breaks and periods when nothing was scheduled. Time lost to breakdowns, changeovers and waiting for materials counts against availability.
Performance compares actual output to what the machine could have made in the time it ran.
Performance = (ideal cycle time multiplied by total parts made) divided by run time.
The ideal cycle time is the fastest steady time per part the machine is designed or proven to achieve. Slow running, small stops and brief jams reduce performance.
Quality is the share of parts that were good the first time.
Quality = good parts divided by total parts.
Scrap and parts that need rework reduce quality.
A worked example
The numbers below are hypothetical, chosen to keep the arithmetic easy to follow.
Suppose a packaging machine has an 8-hour shift, or 480 minutes, of planned production time.
- Downtime: 60 minutes of stops (a jam, a changeover and a wait for material). Run time is 480 minus 60, which is 420 minutes.
- Availability: 420 divided by 480 equals 0.875, or 87.5 percent.
- Ideal cycle time: 1 minute per part.
- Total parts made: 340 in the 420 minutes of running.
- Performance: (1 minute times 340 parts) divided by 420 minutes equals about 0.810, or roughly 81 percent.
- Rejected parts: 17, so good parts are 340 minus 17, which is 323.
- Quality: 323 divided by 340 equals 0.95, or 95 percent.
Multiplying: 0.875 times 0.810 times 0.95 comes to about 0.673, or roughly 67 percent.
A handy cross-check: the same result comes from dividing the time needed to make the good parts by the planned time. 323 good parts at 1 minute each is 323 minutes, and 323 divided by 480 is about 0.673. The two routes agree, which is a useful test that the factors were calculated consistently.
In this example, a score near 67 percent says that about a third of the planned time did not turn into good output at ideal speed. It also shows where: the largest single loss is downtime, followed by slow running, then rejects.
Where IoT data comes in
Each factor depends on information that machines can supply:
- Run and stop states. A signal from a controller, a current sensor on the motor or a simple light-stack reader can record whether the machine is running. Many plants take these from programmable logic controllers, since the controller already knows the machine’s state.
- Stop reasons. Automatic detection tells you a stop happened, but not always why. Many sites ask operators to tag reasons on a screen, because knowing “changeover” versus “jam” guides what to fix.
- Part counts. Counters on the line, sensors that detect each part or counts from the controller supply total output.
- Rejects. Counts from inspection stations or manual entries provide quality figures. Automated checks, such as camera-based inspection, can feed this directly.
- Time stamps. Readings are usually stored as time series data so periods can be assigned to shifts and products.
Collecting this data automatically avoids the delays and rounding of handwritten sheets. Dashboards can show current OEE during a shift, though sensible sites are careful about how they use the number.
Pitfalls to watch
- Definitions vary. Companies differ on what counts as planned time and what the ideal cycle time is. Compare OEE only between machines or sites that use the same rules.
- A poor ideal cycle time distorts everything. If it is set too slow, performance can exceed 100 percent, a sign the reference is wrong.
- Data gaps. Lost connections or missed counts produce false results. Good data quality practice and handling offline operation, such as local buffering, help.
- Averages hide detail. A single figure for a whole plant can conceal one failing machine. Looking at each factor separately is more informative.
- Gaming. If a number becomes a target, people may find ways to improve it without improving the process, such as excluding downtime categories. It works best as a tool for finding problems, not for ranking people.
- It is not everything. OEE says nothing about energy use, safety or whether you are making the right products.
Used with clear definitions and reliable inputs, OEE gives teams a shared, simple way to see where lost time goes and to check whether changes make a difference.