Implementing automatic production actuals collection starts with a definition, not a device purchase: what counts as an actual, which work order and lot own it, and when is it final? A PLC counter can report movement, but it cannot by itself explain changeovers, rework, rejects, network outages, or the value posted to ERP. This practical guide helps plant management, production engineering, operators, and IT in Thailand specify and buy a trustworthy capture system, from the first site survey through the RFP, FAT, and SAT.
The key decision: collect attributable events, not just counts
A part passing a sensor is useful evidence, but it is not necessarily completed good production. A usable actual includes a work order, operation, equipment, time, quantity and disposition. Where needed it also includes input and output lots and the person who confirmed an exception. Sensors should record what machines know reliably. Operators should record the reject reason, rework decision, or material choice that the machine cannot know. The aim is a record whose source and business meaning can be reconstructed.
Define five things together: the event grain; attribution to orders and lots; local persistence and replay after a network outage; reconciliation and approval in MES or the production system; and transmission of final values to ERP. Judge proposals by whether an end-to-end production scenario works, rather than by the number of features on a slide. ISA-95 provides a shared vocabulary for the boundary between control, manufacturing operations, and enterprise systems; it does not choose the product for your plant.
What data should be captured for each use case?
“Visibility” is too vague to price. Order progress needs good and reject quantities by operation and timestamp. Work in process needs both input and completion. Costing may need materials, labour time, rework and scrap assignment. Quality evidence may need inspection results, instrument IDs, approval and corrections. These are different data models from a machine runtime monitor, even if both read the same PLC.
For each user, write down the screen or decision the data supports and the acceptable delay. A supervisor may need a shortage before shift handover; production control may need it before tomorrow’s schedule is released; finance may accept only approved totals at period close. Separate provisional events, reconciled actuals, and quantities posted to inventory or costing.
| Purpose | Minimum record | Failure if omitted | Consumer |
|---|---|---|---|
| Order progress | Order, operation, equipment, good count, time | Counts assigned to the wrong order | Supervisor and planner |
| Lot trace | Order, input and output lots, operation, time | History cannot be reconstructed | Quality and warehouse |
| Inventory | Approved quantity, unit, location, correction | Duplicate posting or negative stock | ERP and warehouse |
| Cost | Quantity, time, material, rework, scrap reason | Variance cannot be assigned | Finance and management |
This table is a scoping example, not a universal mandatory field list. For the wider ERP/MES boundary, see our production management system guide.
Survey the changeover process before specifying PLC tags
Record equipment model, PLC, available communication ports, operation, product changeover frequency, tooling, physical count point, rework loops and terminal location. Observe one physical item while watching the tag change. Ask when counters reset, what happens after a power loss, whether trial runs increment the value, and whether a part can cross the detector more than once. A tag name alone cannot establish its business meaning.
The work order changeover is often the critical boundary. Does an operator swap a paper card, scan an order, or load a PLC recipe? If the switch time is uncertain, counts near that boundary may belong to either order. Agree whether they stay with the previous order, move to an unassigned queue, or require supervisor approval. Observe night shifts, breaks, stoppages, mixed orders and rework as well as steady running. For legacy equipment, compare an added sensor with a simple terminal entry. Include shutdown time, machine warranty and maintenance responsibility if the control program must change.
When can a PLC count become a production quantity?
Reading a cumulative counter can be robust, but subtracting successive readings is not enough. Specify how to recognize rollover, manual reset, restart, missed reads and repeated reads. A sensor at operation entry is not evidence of completed good units. Agree the physical point used for each count, and verify it on the machine. Keep equipment ID, tag ID, event time and gateway receipt time, raw value, quality flag, collection method and connectivity state. If the equipment clock is unreliable, preserve both times and define which one drives shift reporting.
OPC UA provides an information and communication model spanning sensors, controllers, MES and ERP. “OPC UA compatible” alone does not guarantee that the required tags are exposed or that their meanings, permissions, timestamps and reconnection behaviour suit your application. Test those points with an actual machine or a representative simulator. See the OPC Foundation overview and its security model.
Divide automatic and operator input by responsibility
Machines can detect motion and counts, but they do not automatically know why a visual inspection rejected a part, who approved rework, or which substitute material was loaded. Conversely, asking operators to key every count invites late entry. Combine reliable machine signals with a short, contextual terminal workflow. Show the current order, product, operation, latest automatic count and unconfirmed events. Use language and reason codes the local team actually understands.
Operator identity supports permission and correction history, not just blame. Design shared terminal sessions, shift logout and approval separation so they can work during real production. Show whether an event is stored locally, awaiting transmission, accepted, or rejected. Hiding an offline state erodes trust.
Bind events to work orders and lots at collection time
Joining counts to an order later using only timestamps is fragile across changeovers and shifts. Capture order, operation, equipment and product IDs when the event is created, plus input and output lot IDs when needed. If the PLC has no order context, the terminal can mark an order’s active period. A count during an overlap or gap should be queued as unassigned, not silently guessed.
Use stable IDs rather than translated display names. Store the relevant revision of the route or master data so later changes do not rewrite history. Validate a scanned barcode against the active order, operation sequence and permitted material lot. Lot granularity should follow recall scope, customer requirements and quality risk; individual serialization is not always justified. For scope and cost, see our MES implementation and ERP boundary guide.
Offline buffering, replay, and deduplication are core requirements
Factory networks fail during outages, maintenance, wireless congestion or server restarts. Specify what the gateway and terminals store locally, for how long, what happens at capacity, and the replay sequence. “Syncs when the connection returns” cannot be accepted without a test. Each event needs a persistent ID; the MES must count the same ID only once even if the gateway sends it twice. Corrections should reference the original event instead of silently overwriting it.
If MQTT is selected, Sparkplug can standardize topic namespaces, payload and session state. It does not assign your plant’s order or lot IDs and does not replace local persistence rules. See the Eclipse Foundation specification. In a test, create events offline, reboot the device, reconnect, resend the same event and compare raw records with MES and ERP. Include late arrivals, reversed order and an event received after its work order has closed.
Reconcile through MES before posting final values to ERP
A successful network response means a message arrived; it does not prove ERP inventory is right. MES or the production system should validate order status, unit, sequence, duplicates and tolerance. A named owner resolves exceptions and approves the final actual before ERP posting. Document which system is the source of truth for each concept and which copies are read only.
Reconcile at three boundaries: physical counter to collected events; collected events to order-level MES actuals; approved MES actuals to ERP inventory and cost records. Equal totals may hide a transfer between lots. Keep a path from every difference back to its event. Define the daily owner, resolution deadline, correction rights and replay method. ISA-95 helps discuss the Level 3/Level 4 boundary, while the exact ownership must match your deployed applications. For custom edge cases, consult our custom production system development guide.
Preserve evidence quality, including corrections
Keep raw signals separate from processed values. The raw record needs source, timing, quality and connection state. The transformation needs its rule version, attribution, exclusion, correction and approval. The final record needs approval time, ERP transaction ID, response and retry history. Set retention rules based on your business and applicable obligations; this guide does not assert that one retention period satisfies every industry.
Equipment, gateway and MES clocks may differ enough to push one item into the wrong shift. Preserve occurrence and receipt timestamps, state the time basis for reporting, and model the plant shift as an identified calendar interval. Show operators and supervisors the old value, new value, reason and approver for each correction. Distinguish training and trial events from live production.
Put acceptance scenarios, not just connection points, in the RFP
An equipment count and screen count allow bidders to price different scopes. Give each vendor the same equipment list, tag meanings, changeover flow, lot rules, operator input, outage rules, MES/ERP interfaces and maintenance assumptions. Supply controlled examples of network topology, APIs and master data. Protect sensitive control-network details.
Write scenarios for normal output, stop and restart, order change, reject, rework, lost network, terminal reboot, ERP rejection and later correction. For each, state input conditions, expected event, display, approval rule and ERP result. Ask the bidder to mark whether each is standard, configured, developed, or requires equipment modification, with price and exclusions. A verbal “supported” claim is not an acceptance criterion.
| RFP item | Question | Evidence to request |
|---|---|---|
| Machine connection | Can the specified PLC expose each required tag? | Tag map and machine test |
| Attribution | How are counts at a changeover handled? | State model and unassigned example |
| Offline | What survives outage and reboot? | Buffer limit, replay and deduplication design |
| Integration | When is the value final for ERP? | API schema and failure handling |
| Audit | Can a correction be reconstructed? | Sample audit record and rights matrix |
| Support | Who changes mappings after a new machine is added? | Procedure, scope and recovery target |
Compare license, signal extraction, gateway, terminal, wiring, segmentation, API work, migration, training, site acceptance and support separately. The Thailand BOI Smart and Sustainable Industry page describes current promotion conditions; eligibility must be checked for the actual project before putting it in an investment case.
FAT proves logic; SAT proves operation on the shop floor
In FAT, use simulated PLC signals or recorded sequences to test count, reset, rollover, late delivery, duplicate event, wrong order and API failure. Record expected event count, unresolved queue, correction trail and ERP transmission—not merely whether a dashboard appears. Record machine-specific conditions FAT cannot reproduce and carry them into SAT.
In SAT, use real equipment, shift changes and the actual plant network. Have operators switch orders, register rejects, complete a run and reconcile it with a supervisor. Confirm who resolves exceptions and how night shifts obtain support. Do not accept a result that depends on a vendor manually fixing hidden data after the demonstration. Capture input conditions, timestamps, screen and log evidence, pass or fail, open issues and retest date.
Any numerical acceptance target needs an explicit denominator. “99% collection” means little unless it specifies eligible events, correct order and lot attribution, and a time window. Agree whether operations without any signal are excluded. A target in the contract is a project-specific commitment, not an industry benchmark.
Roll out from a representative line, then test different equipment
A pilot on only the easiest or hardest machine gives a distorted expansion estimate. Choose a line with both PLC signals and operator decisions, order changes and rejects. Complete the path from event definition to ERP posting. Then add a different controller or manual operation and measure how many mappings and exception rules can be reused. Verify that the plant’s own master-data and maintenance owners can make routine changes.
Run paper and digital records in parallel only for a defined comparison period. Set exit criteria for each old form so duplicate entry does not become permanent. Train staff to recognize an anomalous count and route unassigned events, not just click buttons. Align reason-code translations and work instructions for Thai and Japanese teams. After go-live, watch unassigned events, duplicates, late arrivals, daily reconciliation time, manual corrections and offline duration before claiming productivity benefits.
Protect OT and make the installation maintainable
Give a gateway only the machine rights it needs; separate read paths from administration. Define device updates, certificate expiry, account removal, backups and restore testing. Plant safety and uptime constrain changes to control equipment. NIST SP 800-82 Rev.3 discusses OT-specific performance, reliability and safety needs when applying security controls.
For OPC UA, certificate renewal and machine replacement are operational tasks, not one-time setup. Ask for the configuration inventory, connection diagram, access matrix, backup and test records as handover deliverables. If remote vendor support is allowed, define its approval, time window, logs and revocation after work. “Encrypted” communication alone does not mean trust lists will be maintained correctly.
Make a data contract before comparing quotes
For each transmitted field, define business meaning, unit, mandatory conditions, ID, time basis, version, absence rule, correction, retention, frequency and response. “Quantity” might mean input, completion, passed inspection or packed units. Document who issues and retires plant, line, equipment, order, operation and lot IDs. Do not create a fresh event ID on every retry; that would turn the same production into multiple actuals.
Separate recoverable transport failures from business rejection such as unknown order, missing lot, invalid sequence, closed period or unit mismatch. An endless retry cannot repair master data. Specify an exception queue, owner, notification and audited reprocess action. Test coexistence of old and new API versions during cutover.
Choose automation level by three tests
First, is the signal physically dependable? A detector crossed repeatedly by the same reworked item is not a good completion counter. Second, does the decision require human judgment? Visual rejects and material substitutions should not be invented by software. Third, is the device change economical and maintainable? A rarely used legacy machine may justify simple terminal entry, while a high-volume line may justify full automatic capture. Mix fully automatic, automatic plus confirmation, and manual approaches, provided all create the same core event structure.
Diagnose a mismatch from the physical process outward
When figures disagree, fix the order, operation, machine, lot, shift and time interval under investigation. Compare physical items and paper record with PLC, gateway, MES and ERP counts. Find the first boundary where the difference appears; totals alone can hide overstatement in one lot and understatement in another. Trace the original event, buffer, replay ID, MES acceptance or unassigned queue, and ERP response. Correct with a reason and approver, then fix the root cause: detector placement, terminal defaults, delayed master data or network capacity. Rehearse this procedure during SAT so the night shift has an evidence-based route to recovery.
Frequently asked questions about production actuals collection
Is reading a PLC enough for automatic production actuals collection?
No. It supplies a machine observation. The work order, lot, good/reject decision, correction, approval and ERP posting still need defined rules. Confirm what the counter physically measures and how it resets.
Can older machines without a PLC port be included?
Often yes, through an approved external contact, added sensor or terminal entry. Check safety, warranty and maintenance ownership before modification. A well-designed manual step may be better than an unreliable automatic signal.
What changes the implementation cost?
Equipment variety, accessible signals, terminals, attribution rules, buffering, MES/ERP interface work, tests, training and support matter more than equipment count alone. Ask bidders to price the same acceptance scenarios with separate cost lines; this article does not present a universal price.
Should offline counts automatically post to ERP after reconnection?
First validate order and lot status and period close. Some late events should await approval. Reconnection and business approval are separate states. Verify that receiving a duplicate event cannot increment quantity twice.
How should FAT and SAT be divided?
FAT stresses logic with simulated signals and failures. SAT checks real equipment, people, shifts and network conditions. Record what FAT could not cover and make it a named SAT test.
Summary
A production actuals collection project is also a decision about ownership of orders, lots, quality and inventory. Define events and use cases first, allocate machine and human input, put changeover and offline scenarios into the RFP, then prove logic in FAT and plant operation in SAT. Only after MES reconciliation and ERP posting can the numbers support lasting improvement.
If you are mapping equipment signals and the MES/ERP boundary for a plant in Thailand, contact TOMAS TECH. We can discuss a small, representative validation scope even before a product shortlist exists.
