How to Measure OEE Correctly: Align the Calculation Rules, Then Build the Loss Matrix

Until the rules are aligned, OEE numbers cannot be compared

The OEE formula looks simple: availability × performance × quality. The same formula does not mean the same rules. Whether planned stops sit in the denominator, whether minor stops sit in availability or performance, which standard cycle time you use — if any one of those differs, two lines cannot be ranked against each other. Write a one-page definition that the floor and headquarters can both sign before you build the board.

• One line, three departmental figures: maintenance counts breakdowns, production back-calculates from output, quality uses final inspection, and the books never meet
• Group rankings invert what the floor feels: Plant A drops planned maintenance from the denominator, Plant B keeps it in, so the ranking rewards the rule, not the line
• The figure looks worse after a countermeasure: standard cycle time or the minor-stop threshold was changed mid-stream, so the two periods cannot be compared
• The large screen and the monthly report are not the same set of numbers: the live path uses PLC cycle time, the monthly report uses MES output, and reconciliation becomes daily work
• Headquarters formula and shop-floor logs run as two systems: the software went in without encoding the rules into collection, so the system only automated the disagreement

All five share the same root: before go-live, nobody wrote an executable rule for what counts as running, stopped, or good. Align the calculation rules first, then put the system in — that order matters more than picking software. If the group already specifies a convention, follow it; if not, have the floor and management sign the page before you start collecting.

Availability, performance, quality: where each rate gets stuck

Each of the three rates has more than one calculation that can be defended. The risk is not the debate itself; it is ranking sites that still use two different methods. Japanese plants often split OEE into availability, performance and quality under a headquarters convention. That is not a rule every plant must copy. If your group specifies a convention — Japanese-style or otherwise — follow that convention.

• Availability: whether breaks, shift meetings and planned maintenance enter the denominator; whether changeover is treated as planned or unplanned; whether the stop clock starts at the alarm or at human confirmation
• Performance: whether ideal cycle time is theoretical, current process or a headquarters standard; whether minor stops sit in availability or performance; which SKU cycle you use on a mixed-model line
• Quality: whether reworked units return to the numerator; whether start-up scrap is a quality loss or a time loss; whether in-process or final inspection is the official count

How to build the loss matrix

Whole-line OEE only answers which of the three rates dragged the line today. To act, you also need a per-machine loss matrix: split the three rates into loss types the floor can point at, and assign a data source and an owner to each type. Finer is not always more useful. Stop at the grain where you can say who acts when the number moves.

• Expand the three rates into loss types and stop at a grain a supervisor can point to: breakdown, changeover, minor stop, speed loss, start-up, defect
• Name a data source for each type: PLC signal, Andon call, MES order, quality verdict, and short manual entry only when needed; write the primary source in one sentence
• Write ownership in one sentence: equipment, process, logistics or quality, so the loss is not reassigned in every meeting
• Fix the minor-stop threshold and auto-record rule together: how many seconds count as a stop, and whether PLC writes it automatically or Andon confirms it; link Andon so a minor stop is recorded at once and attributed at once
• Write the one-page definition before you configure the system: if the group or headquarters already specifies a convention, map to it; if not, collect only after the floor and management have signed

Loss type, data source, and where the rules usually split

The table lines up common losses, collection sources you can actually land, and the rule splits that most often make figures incomparable. It is a checklist for a working session, not a mandatory template. If a signal can be collected automatically, do not rely on memory after the shift.

From a single-line model to a plant-wide platform

We use an OEE model-line approach: pick one representative line, run the definition, the loss matrix and Andon linkage to completion, confirm that whole-line OEE and per-machine losses reconcile, then copy to other lines. Where a foreign-invested plant’s headquarters uses a Japanese-style convention, lock the formula, the denominator and the minor-stop threshold on the model line so you do not rewrite the rules after rollout. Discrete assembly and packaging lines with a clear cycle time take this path more readily.

The live chain is PLC → data hub → SignalR → Web board: the floor sees the line heatmap and failure intensity, and management sees the same numbers. Once the model line is stable, extend the same rules and matrix across the plant. Skipping the single-line pilot and covering the whole plant with boards usually turns unaligned rules into a plant-wide reconciliation load.