What OEE answers and what it does not

OEE answers this question: what share of the theoretically available equipment result was obtained within defined planned production time after stoppages, speed loss and unacceptable output are considered?

By itself it does not answer:

  • whether enough orders were available;
  • whether the right orders were produced;
  • whether the current portfolio is profitable;
  • whether the laser is the constraint of the overall flow;
  • who is responsible for a stoppage;
  • whether new equipment should be purchased;
  • what the specific technical cause of a failure is.

High OEE can coexist with excess work in progress. Low OEE can be expected in a cell handling experimental or one-off work. Read the figure together with the plan, product mix, queue and data quality.

A formula with clear boundaries

One practical form is:

Availability = actual operating time / planned production time.

Performance = standard time for actual output / actual operating time.

Quality = acceptable output / total output.

OEE = Availability × Performance × Quality.

This is not the only possible terminology. ISO 22400-2 defines KPIs through formulas and their elements, while NIST illustrates hierarchical relationships between metrics. Consistency of the definition matters more than the label of a numerator.

For a laser, output may be measured as parts, equivalent standard time or another agreed basis. Cut length often ignores pierce count, geometry and auxiliary movements; sheets ignore differences in nest complexity; part count distorts comparisons between large and small items. For a mixed product range, verified standard time for the actual jobs is usually a more suitable basis.

An example that is not a universal target

Assume a shift has 420 minutes of planned production time. Reliable operating time is 315 minutes. Standard time for the output actually produced is 270 minutes. Of 100 recorded units, 96 are confirmed acceptable.

  • availability: 315 / 420 = 0.75;
  • performance: 270 / 315 ≈ 0.857;
  • quality: 96 / 100 = 0.96;
  • OEE ≈ 0.75 × 0.857 × 0.96 ≈ 0.617, or 61.7%.

This figure shows the combined effect of three loss groups within the stated model. If 60 of the 105 non-operating minutes have no known reason, it is not valid to claim that material or the operator is the main problem. Add “unknown reason: 57% of stoppage time” to the report as a separate data-quality metric.

Why the same OEE can require different decisions

Imagine two shifts with OEE close to 60%. In the first, availability is low because of long, confirmed waits for material, but standard pace and quality are stable while the machine runs. In the second, the machine runs for almost all planned time, yet performance is weak and the reasons for variance from CAM time are not separated. An identical total does not make the situations identical.

For the first shift, analyze sheet readiness, the job queue, the logistics route and the time of delivery. Changing cutting parameters merely because of OEE is unjustified. For the second, review standard-cycle boundaries, product mix, postprocessor version, actual pauses and quantity reliability. Increasing availability will not remove the main uncertainty.

Add a short interpretation card to the report:

| Field | What to show | |---|---| | OEE | value and period | | Components | availability, performance, quality | | Denominator | planned-production-time definition and rule version | | Coverage | share of time, output and standards with valid data | | Largest confirmed loss | only what events prove | | Unknown | unclassified downtime and jobs without a valid link | | Next check | owner, deadline and evidence required |

This card does not turn OEE into a diagnosis. It separates a measured component from an assumption about its cause.

The method version is part of the result

A change to the denominator or standard time must create a new method version. If lack of orders used to be included in planned time and is now excluded, figures before and after the change do not form a continuous trend unless recalculated. The same applies to a new CAM standard, a different event for recording good quantity or a new mapping of machine states.

The minimum version record contains its effective date, changed definition, reason, author, list of recalculated periods and known limitations. Mark the version boundary on charts. If history cannot be recalculated, compare the old and new series separately.

Before approving a change, calculate several shifts under both the old and new rules. This reveals how much the number changed because of the method rather than production. It matters especially when OEE supports an investment decision or an automation assessment.

Planned production time is the most important decision

The denominator defines the meaning of OEE. Decide in writing before calculation:

  • whether breaks are included;
  • whether planned maintenance is included;
  • how lack of an order is treated;
  • whether first-sheet preparation is included;
  • whether changeover is included;
  • how to count time after cutting ends while the pallet remains blocked by unloading;
  • how to classify training, tests and demonstration cutting;
  • how emergency loss of electricity or gas is treated.

There is no need to prove that one option is universally correct, but rules must not change between shifts without a marker. When lack of orders is excluded, OEE measures execution during scheduled production rather than calendar-capacity use. Investment planning then needs an additional utilization factor.

How to define laser operating time

Power-on time is unsuitable. A machine may be on but in READY or WAIT. Beam-on time alone is also too narrow because a productive cycle includes necessary movements, pallet exchange and other actions. MTConnect distinguishes ACTIVE, READY, INTERRUPTED, WAIT, FEED_HOLD, STOPPED and PROGRAM_COMPLETED, as well as waiting reasons such as loading and unloading. It provides a vocabulary, but the local controller state must be validated on the actual machine.

A practical check is:

1. Select one typical shift. 2. Compare the machine log with actual events. 3. Decide for every state whether it belongs to operating time. 4. Check transitions: start, pause, completion and pallet waiting. 5. Document the mapping and its version.

Recheck the mapping after software or integration updates. Otherwise an OEE change may reflect reclassification rather than process improvement.

When downtime reasons are inaccurate

An inaccurate reason does not always invalidate availability. If the stoppage interval is reliable, availability can still be calculated, although a Pareto analysis of causes will be weak. Separate two fields:

  • machine state — an automatic fact such as WAIT or STOPPED;
  • operational reason — material, program, unloading, inspection, fault, planned action and so on.

Automation records time and state well. A person often knows the context better. Do not force an operator to identify the technical root cause before diagnosis. They can record an observed category, while a supervisor or service specialist clarifies it later without overwriting the original record.

A reason directory that works

The initial list can be short:

1. prepared material unavailable; 2. approved program or job unavailable; 3. setup or first part; 4. unloading not complete; 5. quality inspection or decision pending; 6. planned maintenance; 7. unplanned technical stoppage; 8. infrastructure: gas, electricity, network or extraction; 9. organizational pause; 10. unknown or requires clarification.

An “unknown” category is necessary. Banning it creates invented precise answers, but exceeding an agreed internal threshold should trigger review.

Inconsistent wording, abbreviations and excessive code lists make manufacturing records difficult to analyze; users select “other” when classification is inconvenient. Use a short top-level list, retain the raw description and normalize it later.

How to calculate performance when CAM time is inaccurate

Performance can become the largest error source. If standard cycle time comes from CAM, first compare plan and actual values for stable job groups. ART-142 addresses this validation separately.

Until the standard is stable:

  • show performance as provisional;
  • do not use it for bonuses;
  • do not silently cap values above 100%;
  • analyze variance by material, thickness, postprocessor version and geometry type;
  • retain the estimate version;
  • do not compare a trend across a method change without recalculation.

A result above 100% may represent genuine outperformance, but it can also reveal a wrong standard, different cycle boundaries or a quantity error. Investigate it instead of automatically declaring success.

How to calculate quality

Quality must be based on a confirmed result. Parts still held in the sheet skeleton and not identified should not automatically count as acceptable. Define the event when a unit enters total output and when it receives acceptable status.

Recutting must remain visible. If a defective part is not registered and is simply cut again, quality appears artificially perfect. Link the original item, nonconformity and replacement.

A defect may be discovered at a later operation. Recent values may then be refined. Show the data currency date on the dashboard instead of silently rewriting history.

Add a data-confidence score

This is a local trust control, not a standardized OEE component, and it should not be multiplied by OEE. Display several simple shares beside it:

  • machine-log coverage;
  • downtime with a classified reason;
  • output with confirmed acceptable and total quantities;
  • jobs with valid standard time;
  • events linked to the correct order ID.

For example: OEE 61.7%; reasons confirmed for 43% of downtime; quality confirmed for 100% of output; the standard covers 82% of jobs. This is much more honest than one green number.

Reconstructing history when no log exists

Do not invent exact codes retrospectively. Instead:

1. collect controller states and timestamps; 2. compare them with shift reports, service requests, CAM files and material movement; 3. mark confirmed, probable and unknown intervals; 4. calculate OEE only where the core elements are sufficient; 5. do not present a reconstructed Pareto chart as exact fact.

Reconstruction is useful for a baseline, but the future process should rely on events rather than permanent manual recovery.

A four-week data-improvement plan

Week 1: define the calendar and state mapping; manually verify two shifts.

Week 2: introduce short top-level reasons and measure the unknown share.

Week 3: link the job, program, sheet and acceptable or total quantity; check cycle boundaries.

Week 4: review the largest unknown intervals, refine the directory and record the method version.

The first month’s objective is a stable definition, not maximum OEE. Otherwise a better figure may only be the result of a new denominator.

How to read OEE in a laser cell

Always show all three components. The same 60% can mean:

  • low availability with good pace and quality;
  • stable operation with an incorrect standard;
  • extensive recutting;
  • a combination of small losses.

Then inspect the timeline and a Pareto chart of confirmed causes only. If sorting is the constraint, cutting automation will not solve it. If performance is weak for one geometry group only, perform a technological review instead of applying general pressure to the shift.

Common mistakes

  • calculating from 24 calendar hours when production is scheduled for one shift;
  • excluding inconvenient stoppages from the denominator after the fact;
  • equating powered with running;
  • using CAM time without validation;
  • counting cut parts as acceptable before inspection;
  • hiding “unknown”;
  • comparing different methods in one trend;
  • evaluating a person by OEE;
  • purchasing automation solely because the total is low;
  • failing to retain the formula and state-mapping version.

Checklist

  • [ ] Planned production time is defined before calculation.
  • [ ] All exclusions are documented.
  • [ ] Operating time is verified on a timeline.
  • [ ] Standard time has a version and an accuracy assessment.
  • [ ] Acceptable and total quantities have unambiguous events.
  • [ ] OEE is shown with its components.
  • [ ] Unknown reasons are not allocated by assumption.
  • [ ] Data-coverage indicators are present.
  • [ ] A method change creates a new version.
  • [ ] OEE does not replace analysis of flow, queues and economics.
  • [ ] The conclusion contains no unverified technical cause.
  • [ ] Actions target the largest confirmed loss.

Conclusion

When downtime reasons are inaccurate, OEE should neither be discarded nor embellished. Calculate it only from clearly defined, available elements and show its components and confidence level. Use the first period as a baseline and a data-improvement project. As unknown intervals shrink, OEE develops from a general signal into a useful prioritization tool.

Safe boundaries

  • The numerical example is illustrative, not a target.
  • The article does not establish a technical cause of downtime.
  • OEE is not used to assess personnel without context.
  • The article does not replace an audit of the controller, MES or ERP, or local rules.

Need help selecting equipment?

Describe your materials, parts and production task. An L-SEL specialist will help identify the next step without relying on a random specification.

Select equipment for the task