How Do You Add Product and Shift Context to Machine Data?
To add production context to machine data, keep the original machine records and associate them with time-bounded product, order and shift assignments. Match each record using the machine identity and the time the event occurred. Define how changeovers, late entries, overlapping assignments and corrections affect that match before using the results in production comparisons.
What context should each machine record carry?
Start with the questions the plant needs to answer: which product was running, which order it belonged to, which shift was active, and whether the machine was producing or changing over.
Keep these as separate assignments. A shift can change while the same order continues. An order can change while the product stays the same. Avoid making one combined label that must be replaced whenever any detail changes.
For each assignment, record:
- The machine, line or production position it covers.
- The product, order or shift identity.
- The effective start and end.
- Whether it is planned, confirmed or awaiting review.
- Where the assignment came from and who confirmed it.
- When it was entered or corrected.
Use stable identities alongside readable names. If a product description changes, earlier records should still point to the intended product.
Also define scope. If different positions on a line can run different products, a single line-wide assignment needs an explicit rule for those differences.
How should time boundaries work?
Adopt a consistent boundary rule: include the start instant and exclude the end instant. A record exactly at a handover then belongs to the assignment starting at that instant.
Use the event time for matching, rather than the time the record reached the reporting system. Keep both times so delayed delivery remains visible. Document the time zone used for shift schedules and how machine timestamps are interpreted.
An assignment without an end can mean “still active”, but it should remain visible for review. Do not let a missed changeover confirmation silently carry yesterday's product into the next run.
The following assignment history is illustrative. Every time and identifier is an example.
| Assignment | Effective interval | Context |
|---|---|---|
| Product run | 06:00 to 09:20 | Product A, Order A |
| Changeover | 09:20 to 09:45 | From Product A to Product B |
| Product run | 09:45 to 14:00 | Product B, Order B |
| Morning shift | 06:00 to 14:00 | Morning shift |
| Afternoon shift | 14:00 onward | Afternoon shift |
Under these rules, a record at 09:20 falls in the changeover. A record at 09:45 belongs to Product B. At 14:00, the shift changes; that alone does not establish the next product assignment.
How should changeovers be represented?
Give changeover its own operating state. Record the outgoing product, intended incoming product and actual changeover boundaries separately.
Agree what marks the end: setup completion, the first accepted piece, or another observable event. Choose a definition that answers the plant's reporting question and apply it consistently.
Decide how to handle trial pieces, setup scrap and cleaning. A trial piece may need association with the incoming product while remaining classified as setup output. Product identity and operating state should both remain available.
Do not divide a changeover between products merely to fill a reporting gap. If responsibility for that time matters, define a separate allocation rule and show it in the report.
For related decisions about planned time, losses and comparisons, see OEE explained: what the number hides.
What happens when an assignment is entered late?
Distinguish the effective time from the entry time.
In an illustrative example, an operator records a product change at 10:15 but confirms that production actually changed at 09:45. Store both times. Recalculate the affected context from 09:45 after confirmation; do not treat the entry time as the production boundary.
Until confirmation, mark the affected records as provisional or unassigned. Show that status wherever the figures are used.
Define who may backdate an assignment, what evidence is required, and which reports need refreshing. Evidence might include an order scan, a setup confirmation or an operator note.
If a previously issued report changes, retain its earlier version or provide a visible revision record. Someone reviewing a saved report should be able to understand why today's figure differs.
How do you resolve overlapping assignments?
Check overlaps within each context type and scope. A product assignment and a shift assignment can cover the same interval by design. Two confirmed product assignments covering the same exclusive production position require resolution.
Do not resolve a conflict by choosing whichever entry arrived last. That hides the disagreement.
Use an explicit sequence:
- Identify the conflicting assignments and affected interval.
- Check whether concurrent production is legitimate for that equipment.
- If it is, refine the scope by position, lane or another agreed unit.
- Otherwise, ask the authorised owner to correct the boundary or assignment.
- Recalculate the affected records and retain the decision history.
Keep conflicting records visible as unresolved. Exclude them from comparisons that require confirmed product context, and show the excluded time or quantity alongside the result.
Planned assignments should remain distinguishable from confirmed actual assignments. Decide whether a plan may fill a gap provisionally, and label that use clearly.
How should machine records spanning a boundary be handled?
A point event can be matched at its timestamp. A record covering an interval needs a separate rule when it crosses a product or shift boundary.
For an illustrative example, suppose a machine sends a production total covering 09:40 to 09:50, while Product B starts at 09:45. The total alone does not identify how many pieces belong on either side.
Choose an approach supported by the available evidence:
- Use individual piece events or counter readings at the boundary.
- Request smaller reporting intervals where practical.
- Hold the total as unresolved until supporting records are available.
- Use an agreed allocation estimate and label it as estimated.
Do not present a time-based split as an observed product count. Also define how counter resets, missing readings and uncertain timestamps are flagged before calculating production differences.
How should corrections affect historical reports?
Preserve the original machine record. Correct the assignment history and the association derived from it.
For each correction, retain the previous value, replacement value, effective interval, reason, person making the change and entry time. Restrict correction rights to agreed roles.
Recalculate only the affected machines and intervals, including dependent totals and comparisons. Check that previously associated records are reassigned rather than duplicated.
Decide whether reports show the latest corrected history or the history approved when the report was issued. Where both views are needed, name them clearly and show the revision status.
Write these behaviours into the functional design specification, including ownership, conflict handling and report revision rules.
Which common mistakes should you avoid?
Applying the current product to historical records. Match against the assignment effective when the event occurred.
Treating the shift schedule as proof of production. A scheduled shift establishes a reporting period; require separate evidence for the product and order actually running.
Filling every gap automatically. Keep unknown context visible until it can be confirmed.
Comparing unlike runs. Agree whether comparisons include setup, trial output, rework and unresolved periods. Show those choices beside the figures.
Correcting a dashboard alone. Apply corrections to the shared assignment history so other reports receive the same context.
Using product identity to select a target without versioning it. Where comparisons depend on a target rate or recipe version, retain the version applicable to the run.
What should you test before trusting the comparisons?
Use a small, labelled test history with expected results:
- An event exactly at a product boundary.
- A shift change during an uninterrupted order.
- A changeover containing trial output.
- A late assignment followed by confirmation.
- Conflicting assignments and a genuine context gap.
- A production total spanning a boundary.
- A correction that changes an already issued report.
For each case, check the assigned context, uncertainty status, totals and revision history. Confirm that the report explains exclusions and estimates.
Frequently asked questions
Do we need a full production management system?
Start by defining the assignment records, ownership and reporting rules. Assess whether existing systems can support them before choosing additional software.
Can operators enter context manually?
Yes, if the agreed process includes effective times, confirmation, overlap checks and correction history. Make unresolved entries visible to the person responsible for closing them.
Should a shift change close the production order?
Only if that reflects the plant's process. Keep shift and order boundaries independent so an order can continue across shifts without creating an artificial production break.
What if the correct product cannot be established?
Leave the interval unassigned and explain its effect on the report. Do not turn an unsupported guess into confirmed production history.