“Which machine settings applied to this returned part?” Production equipment traceability fails when a factory stores PLC tags but cannot connect a machine event to the right output lot or serial. A temperature chart may be accurate and still be useless for investigating a specific product. This guide gives manufacturing, quality, and engineering teams in Thailand a practical machine-to-part event contract, procurement checklist, and FAT/SAT acceptance tests. It covers asset ID, job, recipe revision, time window, actual process values, tool and calibration state, alarms, material lot, and output identity.
Buy a reliable machine-to-product link, not a collection of tags
Equipment monitoring answers whether a machine is running normally now. Product genealogy answers what happened to a named part when it passed that machine. The latter requires a verifiable connection between product identity and the equipment cycle or process window. If the link is inferred solely from nearby timestamps, clock drift, overlapping cycles, buffered workpieces, changeovers, and delayed data delivery can attach the wrong values to a part.
The wider choice of implementation model is covered in our traceability implementation guide. Approval and governance of machine, material, method, and people changes belong to a broader 4M change management system. This article takes one equipment operation and asks a narrower question: what evidence proves which input lot became which output lot or serial on which machine under which actual conditions? It does not duplicate a general traceability program, inspection data management, or the connection specifications for SEMI EDA and SECS/GEM. Existing interfaces can supply data; the contract here defines what that data means for a product.
Two search directions reveal whether the link works. Enter a product ID and retrieve its equipment asset, job, recipe revision, processing interval, actual parameters, tool state, alarms, and inputs. Then enter an equipment alarm and retrieve the potentially affected output identities. A dashboard that answers only the first question, or only shows raw machine history, leaves an important gap.
Set the traceability unit and process boundary before specifying software
Not every process can identify each unit at the machine. A single-piece assembly station with entrance and exit scans can often bind a serial to a cycle. A multi-cavity moulding machine may need shot and cavity IDs. A continuous oven may need material transitions, residence time, output lot, and a time window. A batch vessel may need a batch ID and multiple input and output lots. Define the finest unit that can be evidenced under real production conditions. Promising serial-level precision when only a batch is observed creates a misleading record.
| Process | Practical first linking key | Boundary evidence | Exception to test |
|---|---|---|---|
| One-piece flow | Serial plus cycle ID | Entry scan, completion, exit scan | Rework, empty cycle, early eject |
| Multi-cavity process | Shot ID and cavity or sublot | Tool ID and shot complete | Missing cavity, mixed output |
| Continuous process | Time window plus input/output lots | Material and product change points | Residence time, mixed material |
| Batch process | Batch ID and multiple material lots | Charge, start, discharge | Mid-batch addition or split |
Higher granularity is useful only when the data capture is reliable. An operator forced to make repeated manual scans may leave gaps. If a process can prove only lot-by-time-window membership, show that as the confidence level and conservatively widen the affected set. Do not present a calculated link as an observed scan.
Define what “processed” means
A cycle-start signal, cycle-complete signal, and good-part-ejected signal represent different facts. A cancelled or interrupted cycle cannot silently become a completed product event. Distinguish input confirmed, start, completion, good or reject discharge, rework, and scrap. If the machine does not expose one of these signals, specify the additional sensor or operator confirmation and record its source. For example, an entry barcode scan alone should not count as manufacturing output. A normal completion could require both machine-complete and exit confirmation; without the latter, keep the item in an unmatched state.
The machine-to-part event contract
“Save all PLC tags” is not an acceptance specification. The contract must define the identity of an event, how it is joined to a product, what happened, when, and how uncertainty is handled. The following fields are a project design baseline, not a claim that a particular standard or law requires every field in every factory.
| Field group | Candidate event fields | Acceptance question |
|---|---|---|
| Event identity | Event ID, process, asset ID, type, data source | Does retransmission remain one event? |
| Work instruction | Order/job ID, item, operation revision | Can a changeover attach a part to an old job? |
| Program | Recipe or NC program ID and revision actually loaded | Is the historical revision retained after master updates? |
| Time | Source and received timestamps, start/end, zone, clock quality | Are drift and late arrivals visible? |
| Actual result | Disposition, measured values, units, sampling point, aggregation | Are setpoints distinguished from measured values? |
| Resource state | Tool, die, fixture, gauge, remaining life, calibration validity | Can state at processing time be reconstructed? |
| Exceptions | Alarm and recovery times, stop, restart, manual amendment | Is affected output identified and correction auditable? |
| Genealogy | Input material lot(s), output lot/serial(s), quantity, split/merge | Do trace-back and trace-forward agree? |
An asset ID should be stable even if operators rename the machine or move it to another line. Keep the display name and line assignment as time-bounded attributes. Tools and gauges need their own identities because they can move between machines. Process parameters need a unit and a definition of the value: setpoint, instant reading, peak, average, or end-of-cycle reading. “Temperature 160” without those facts cannot support an investigation. ISA-95 is a useful reference for an enterprise and equipment hierarchy, but it does not assign a factory’s asset keys automatically.
Preserve input-to-output relationships
One input lot can feed many output lots; a process can mix several inputs; a rejected part can return for rework. One “input lot” column and one “output lot” column lose this structure. Record multiple parent and child relationships, quantities, and the reason for split, merge, or re-entry. GS1 EPCIS offers a visibility-event model for what, when, where, why, and how regarding products and assets. It is a useful reference if information may later be shared across organisations. It does not mean every internal machine event must be exported as EPCIS. Establish identifiers and event times early so future mapping remains possible.
Five situations where timestamp-only joining breaks
First, machine, gateway, and MES clocks may disagree. Keep source time and receipt time separately, monitor synchronisation, and flag periods in which the time-based link is uncertain. Set a tolerance appropriate to each process: a ten-second assembly cycle and a two-hour furnace process have different needs.
Second, cycles may overlap while workpieces wait in a buffer. Prefer a cycle ID or entry/exit scans over the nearest timestamp. Third, a network outage can deliver stored events long after processing. Keep a stable event ID, make replay idempotent, and distinguish event time from arrival time. Fourth, recipe changeover creates a boundary: the loaded revision and the first affected workpiece must be explicit. Fifth, rework, manual ejection, scan failure, and cycle cancellation need exception states, not silent assignment to normal good output. An amendment should preserve the earlier record, editor, reason, and before/after value.
Where OPC UA, ISA-95, and GS1 fit
OPC UA for Machinery defines common information structures for machines; its Part 3 covers job management. The OPC UA for Machine Tools specification describes machine monitoring and job-management information. These can guide how a machine exposes its identity, state, and job information. They do not determine which serial in your factory passed a specific cycle. The local signal-to-event rule and identification checkpoint still need engineering.
ISA-95 supplies a common framework for linking enterprise and control domains. It helps keep ERP orders, MES work instructions, and machine jobs distinct. GS1 EPCIS addresses visibility events that can be captured and shared across an organisation or supply chain. They serve different layers of the design, rather than being alternative answers to one procurement question.
| Layer | Useful reference | Factory decision still required |
|---|---|---|
| Equipment connection | OPC UA Machinery and Machine Tools | Meaning and timing of actual signals |
| Business context | ISA-95 | IDs, revisions, effective periods |
| Visibility sharing | GS1 EPCIS | Sharing scope and vocabulary mapping |
| Machine-to-part event | Project event contract | Granularity, exceptions, proof |
Put event examples and acceptance scenarios in the RFP
“Connect the PLC and provide traceability” produces incomparable quotations. Give bidders the process map, available signal list, product identification points, source of jobs and recipe versions, required queries, exception decisions, and representative test data. If machine signal documentation is missing, request a priced discovery task rather than burying unknown work in a fixed estimate. Separate connection acceptance from genealogy acceptance.
For one representative machine and product, annotate where each ID becomes authoritative. Specify each event field as mandatory, conditional, or outside scope; identify its source system, owner, missing-data behaviour, master record, retention basis, and test case. Retention duration depends on the product, customer contract, applicable rules, and quality system. Request a distinct estimate for archives, backup, restore, query performance, and export on supplier change.
Ask each bidder for a signal-to-event mapping with units and update conditions; an ID relationship diagram; normal, alarm, rework, outage, and amendment state transitions; sample event exports; serial-to-history and alarm-to-outputs queries; FAT/SAT protocols; and operator procedures for unresolved links. “Unlimited tags” says little about whether a returned part can be traced.
Separate responsibilities and cost layers
The machine maker can explain a PLC signal, while the factory owns the decision about product identity and its processing boundary. An ERP team knows the order ID source, but may not know whether a program was actually loaded on the machine. Assign who supplies, converts, approves, and troubleshoots each field. Especially for recipe revisions and calibration status, a query to today’s master is inadequate: retain the revision or valid historical reference at the time of processing.
Break the quote into identification hardware, signal work, gateway, event transformation, MES/ERP interface, search, storage, FAT/SAT, training, and support. An equipment survey and line downtime may need separate allowances. When extending a proven event contract to another machine, check whether “cycle complete” means process end or output discharged. A shared PLC family does not guarantee equivalent events.
FAT and SAT must prove the product link
FAT checks the contract and software with controlled signals, scans, and expected results at the supplier site. SAT repeats representative scenarios on the actual line, with its wiring, network, clocks, operators, and workpiece movement. A successful screen load is not enough. Retain the tested product and event IDs, inputs, expected outcome, actual export, and discrepancy decision.
| Test | FAT expected result | SAT addition |
|---|---|---|
| Normal cycle | Serial, job, recipe revision, actual values on one event | Real entry/exit scans match completion |
| Changeover | Products before/after use correct revision | Check residual work and first-piece boundary |
| Alarm | Event window returns affected output set | Include stopped and buffered pieces |
| Outage | Buffer, replay, deduplication, uncertainty visible | Disconnect and restore the actual network |
| Rework | New event linked to original; old record retained | Follow real re-entry procedure |
| Unknown identity | Missing scan never auto-becomes a normal part | Operator can hold and resolve it |
For an alarm, decide in advance how to treat the workpiece inside the machine at alarm onset, pieces in the buffer, and the first output after restart. An acceptance criterion that merely returns every lot may avoid omissions but provides no useful containment. A criterion that returns a narrow list without evidence risks missing affected parts. Require both defensible inclusion and a documented boundary.
Numerical criteria need a denominator and test method. “Zero missing completion events” needs a defined machine, period, scheduled stops, empty cycles, and outage rule. A mis-link rate needs an independent truth set, such as controlled test workpieces and scan logs. Query response time needs a defined data volume. Document pass, conditional pass, and fail rules before testing. If delayed replay works but products cannot be queried during the outage, the plant needs an interim procedure. Treat a signal-meaning discrepancy discovered in SAT as a change to the event contract.
Start with one machine and two investigation questions
Choose one representative machine and product family. Test whether a defective serial can reveal its machine conditions, and whether a machine alarm can return its affected output. Mark on the process map when material lot, work instruction, loaded recipe revision, tool change, and output identity become known. Do not mark an unobserved fact as “normal” on the screen. Freeze the event contract and mapping only after production, quality, maintenance, and IT accept these boundaries.
Use FAT to prove the rules, then SAT with normal running, changeover, stop, rework, and outage. Extend the accepted contract to similar machines, rechecking each signal’s physical meaning. In routine operation, monitor unmatched machine events, products with no machine history, clock exceptions, amendments, late replays, and unresolved alarm impact sets. Investigate by machine, product, and shift; a zero count is trustworthy only when exception detection itself is working.
Frequently asked questions
Is saving every PLC tag enough for production equipment traceability?
No. Raw time series can support detailed analysis, but product history requires event and product IDs, process boundaries, actual conditions, and exception rules. Keep both the evidence and the verified join.
What if an existing machine cannot read serial numbers?
Check whether scanners or sensors can be added before or after processing and whether conveyor order can be guaranteed. If neither is possible, use a defensible lot and time-window link, disclose its precision, and set a hold procedure for mismatches.
What should a supplier demonstrate in the equipment traceability RFP?
Ask for serial-to-history and alarm-to-output queries using supplied data. Include a changeover, outage with duplicate replay, rework, and failed scan, as well as a normal cycle. Request the event export and expected results as evidence.
Can SAT be skipped after a successful FAT?
FAT tests controlled simulations. SAT exposes actual signal semantics, wiring, clock differences, buffering, and operator behaviour. These may change the event contract, so the site test is essential to accepting a usable history.
Conclusion
The purchasing unit for production equipment traceability is a verified machine-to-part event, not a tag count. Tie stable asset ID, job, applied recipe revision, process window, actual values, tool and calibration state, alarms, input material, and output identity at the finest reliably observable unit. Test normal and exceptional flows in FAT and SAT and retain the evidence behind each answer.
If your Thailand plant is evaluating how to bridge existing equipment with MES or ERP, a process map for one machine and its scan points is enough to start. Contact TOMAS TECH to discuss the event contract and acceptance questions while the RFP is still being drafted.
Sources and scope
- OPC Foundation, OPC UA for Machinery Part 1
- OPC Foundation, OPC UA for Machinery Part 3: Job Management
- OPC Foundation, OPC UA for Machine Tools Part 1
- International Society of Automation, ISA-95
- GS1, EPCIS and CBV
- GS1, EPCIS 2.0.1 specification
The proposed event fields and tests are implementation guidance. Confirm mandatory fields and retention against the customer contract, product rules, and plant quality system.
