OEE Pilot Acceptance: A Test Plan Before Factory Rollout

Start with a signed measurement definition

Choose one line or machine and name the production boundary. Define the counted unit, planned production time, source of total and good counts, ideal cycle reference and treatment of stops. Identify who owns the schedule, product master data, quality decisions and final acceptance. This makes the test about agreed evidence rather than whether the dashboard resembles a supplier's demonstration.

Download the editable OEE pilot acceptance worksheet. Each row has fields for the observed event, manual or independent reference, system result, difference, agreed tolerance, evidence and sign-off. The template starts with every test marked not tested. Add site-specific test windows and acceptance criteria before commissioning; its example categories are not universal performance thresholds.

1. Test counts against an independent reference

Select an appropriate reference with the production team. It may be a witnessed tally over a bounded run, an independently checked batch quantity or another method whose uncertainty is understood. Record the start and finish conditions and any work already between the measurement boundaries. Comparing a shift-end dispatch total with a machine counter without accounting for work in progress can create a false discrepancy.

Include representative difficult events. A packaging line may need closely spaced products, diverted rejects and reverse conveyor movement. A moulding machine may need a changed active-cavity configuration. A machining operation may produce several parts per cycle or only one step in a longer routing. Keep the reference in the same unit as the system result, and retain the conversion where a cycle becomes a part, carton or set.

2. Witness the time events that affect availability

Agree the start and end of the planned production period, including the treatment of breaks, changeovers and periods with no intention to produce. Then record representative stopped and running intervals with a witness. Use a common time reference or measure the offset between clocks. If the operator's record and the system record use different time zones, a correct event can appear to belong to the wrong shift.

Define the boundary between separately recorded stop time and small stops represented within performance loss. A short interruption should not be deducted twice. Check events that cross a shift boundary and events whose reason is entered later. The observed timing and the reason for the event should remain distinguishable so a corrected explanation does not silently change the measured duration.

3. Verify the product reference and quality decision

For each sampled job, check that the count unit and ideal cycle reference match. Record the source and approved version of that reference. An average production rate that already includes routine losses can make performance look stronger by building those losses into the benchmark. Conversely, a reference for a different product or configuration can create an apparent problem that belongs to the master data.

Follow good count back to its actual quality source. Include rejected output, output awaiting a decision and rework passing through the process again. A missing quality result must remain visible. Agree how a later disposition revises the reporting period and how the previous version is retained. The pilot should demonstrate this workflow with the quality owner, rather than assuming every counted unit was accepted first time.

4. Recalculate one result outside the dashboard

Use an agreed, simple data set to test the arithmetic. The following is a hypothetical example, not a factory result or a target. A 480-minute shift includes 30 minutes with no planned production, leaving 450 planned minutes. Recorded stop time is 90 minutes, leaving 360 run minutes. Ideal cycle time is one second per unit, total count is 18,000 and first-pass good count is 17,100.

Availability is 360 divided by 450, or 80%. Performance is 18,000 seconds divided by 21,600 run seconds, or approximately 83.33%. Quality is 17,100 divided by 18,000, or 95%. Their product is approximately 63.33% OEE. The independent check is 17,100 good units multiplied by one second, divided by 27,000 planned seconds, which produces the same result. Use unrounded inputs for the calculation and round only the displayed result.

For mixed products, keep the appropriate ideal-time reference for each interval. A simple average of the percentages displayed for individual jobs is not generally the aggregate OEE. Ask the report owner to reconstruct the combined result from the agreed production-time and good-output records. If the inputs are incomplete or incompatible, show that condition and investigate it instead of forcing a plausible percentage.

5. Exercise missing data and recovery safely

The test plan should distinguish no production from no information. Use an approved simulator, isolated connection or site-authorised test to interrupt the data path. Verify when the report identifies stale or missing data, what the local equipment retains and how the history is handled on reconnection. Record the configuration and actual interruption duration so the outcome is reproducible.

Check duplicate delivery, delayed records and counter restart where these are relevant to the design. A recovered event should return to the correct production interval. Importing the same data twice should not increase output. If part of the history cannot be recovered, the affected calculation should indicate the gap. Plan tests with the site's engineering owner; the method must suit the operational and safety conditions of the installation.

6. Test whether the team can use the record

Ask the supervisor to trace a count difference, review an unresolved stop reason and explain a changed good quantity. Ask the maintenance or technical contact to identify a stale source and follow the support route. These exercises show whether the handover provides more than a correctly calculated screen. They also reveal responsibilities that were missing from the original scope.

The pilot's reporting period should include the conditions it claims to represent. Stable production alone cannot demonstrate a changeover workflow. A single product cannot test how references change between products. List the conditions that occurred, those simulated and those still untested. This allows an informed acceptance decision and a clear plan for completing outstanding observations without overstating the evidence.

7. Make the acceptance record explicit

A test that was not performed remains not tested. A failed event needs an owner and retest condition. If a restricted handover is accepted with open defects, record which use is permitted and what remains outstanding. Preserve the previous results so a later correction can be explained. Approval belongs to the roles named in the project agreement; a supplier's completion notice alone does not resolve a disputed production definition.

Use the pilot to prepare the next installation

The accepted point list, unit definitions, reference versions and operating procedure become a starting package for expansion. Check what changes on the next line: sensors, product units, interfaces, quality data and local responsibilities may differ. Reusing a reporting layout saves work only when the underlying definitions still apply. Keep machine-specific acceptance evidence alongside the common project method.

For the relevant commercial scope, see packaging line monitoring, injection moulding monitoring or CNC machine monitoring. For choosing a supplier before the pilot, use the industrial IoT provider selection guide. This worksheet focuses on checking the delivered record once the scope is defined.

Use the OEE management service to define the wider delivery scope and the manufacturing overview to identify adjacent processes and their owners. Keep those responsibilities explicit when extending the accepted pilot.

Frequently asked questions

Method source

Vorne: calculating OEE provides the factor equations and first-pass good-count interpretation. Checked 5 November 2026. The worksheet, test sequence and hypothetical arithmetic example are addanode's engineering planning materials; site-specific requirements and signatories are set in the project agreement.