Manufacturing data collection in an operating Thai factory rarely starts from a clean sheet. PLC brands, machine generations, networks, sensors, data loggers and business applications differ. A demo can display values quickly, but data that loses timestamps, misses outages, lacks context or cannot be restored is unsafe as a basis for production, maintenance or quality decisions.
This guide treats PLC data collection, equipment data collection, sensor data acquisition and factory data loggers as one governed operating capability. It shows how to convert an informational search into an actionable RFP, a 90-day proof of concept, FAT/SAT evidence, backup and recovery, and a comparable three-year TCO. Unless a primary source is cited, every target and price is explicitly an illustrative recommended design value, not a standard, law, market price or performance guarantee.
Define manufacturing data collection by the decision, not the tag count
Start with who must decide what, by when, and what happens if the decision is wrong. A ten-minute escalation during a shift, a next-day chronic-stop review and a customer traceability investigation need different granularity, timestamps, retention and approval. Write requirements as data → information → decision → action.
| Decision | Raw evidence | Context | Output and action |
|---|---|---|---|
| Which machine stops repeatedly? | RUN, STOP, FAULT, time | asset, product, shift, reason | Pareto, owner, due date |
| Did a condition deviation affect quality? | temperature, pressure, speed, setpoint | lot, process, recipe, calibration | scope, hold, investigation |
| Did energy intensity worsen? | power, flow, output quantity | product, state, time band | baseline variance, inspection |
| Is maintenance attention justified? | vibration, temperature, current, alarm | load, mode, history | candidate task and priority |
ISO 22400-1:2014 provides an industry-neutral framework for defining, composing, exchanging and using manufacturing operations management KPIs. ISO’s official page says the edition was reviewed and confirmed in 2025 and remains current. It does not define your plant’s numerator, denominator, exclusions, closing time or correction authority; those belong in the data contract.
Separate success into connectivity, information quality and business use. The following are recommended design examples only.
| Layer | Evidence | Illustrative target |
|---|---|---|
| Technical | reachability, cycle, buffering, replay, clock | at least 99.5% completeness in agreed production windows |
| Information | unit, quality, asset/product/lot links | owner, meaning and version for 100% of mandatory tags |
| Business | meeting, decision, action, follow-up | daily use with owner and due date for open actions |
Choose PLC, sensor, data logger and gateway roles deliberately
PLC data collection must protect control
PLCs hold states, alarms, counters, setpoints and interlocks, but unconstrained queries from an enterprise application can create an unacceptable control dependency. Specify read-only paths, allowed tags, connection limits, polling load, timeout, reconnection, failure behaviour and change approval. Verify the machine builder’s warranty and safety boundaries. Treat any future write-back as a separately approved risk and test scope.
Options include a native PLC protocol, an OPC UA server, an existing SCADA or historian, or an external edge gateway. Compare the exact PLC model and firmware, licence, timestamp source, quality codes, arrays, events, loading and support responsibility. The label “supported” is not an acceptance test.
Sensor data acquisition is a complete measurement chain
Retrofit vibration, temperature, current, pressure, flow or environmental sensors can make legacy assets observable. Accuracy still depends on mounting, range, sampling, conversion, wiring, noise, calibration, timestamp and operating mode. Link the signal to what the machine was doing on the same clock.
For high-frequency condition data, compare constant raw-waveform transfer with edge features plus event-triggered raw capture. Record what is discarded, whether the result can be recalculated, algorithm version, threshold changes and raw-data retention.
A factory data logger needs an integration exit
A data logger is useful for an isolated survey, calibration comparison, short pilot or machine that cannot yet join the plant network. Permanent manual USB or CSV collection introduces missed pickups, overwritten files, clock drift, version conflicts, malware exposure and transcription work.
Ask how the logger behaves when full or unpowered, how it synchronises time, exposes CSV or APIs, marks quality, stores configuration and calibration, and survives the environment. Define the route from the pilot logger into a governed gateway or data platform.
Select an IoT gateway for lifecycle, not only protocol conversion
The gateway becomes the boundary for acquisition, normalisation, buffering, replay, certificates, monitoring and remote support. Compare support life, OS updates, vulnerability handling, configuration export, bare-device restoration, spare replacement, logs, storage endurance, power and temperature. See our industrial IoT gateway selection guide for procurement and FAT/SAT details.
Build an equipment data architecture with clear ownership
ANSI/ISA-95.00.01-2025 updated Part 1 of the enterprise-control integration series, replacing the 2010 edition. ISA says it covers the scope of manufacturing operations and control, organisation of physical assets, functions at the enterprise/control interface and shared information. It is a vocabulary and boundary model, not a prescribed vendor topology.
Allocate responsibility across field assets, acquisition edge, OT network, OT DMZ, data platform and MES/maintenance/analytics. Whether the upper layer is on-premises, cloud or hybrid, the machine must not depend on an unavailable dashboard; the edge must buffer an agreed outage; replay must be idempotent; and upper layers must not gain unconstrained paths back into control.
| Layer | Responsibility | RFP evidence |
|---|---|---|
| PLC and sensors | source signal without harming control | source tag, cycle, load and approval |
| Edge/logger | normalisation, quality, time and buffer | configuration, capacity test, replay log |
| OT network/DMZ | segmentation, allowed flow, observation | data-flow map, rules and deny test |
| Data platform | retention, version, query, API and audit | schema, restore, retention and access |
| MES/analytics | context, KPI and decision workflow | calculation rule, approval and correction log |
If a wireless segment is involved, test production interference, real clients, authentication, buffering and recovery rather than headline speed. Our factory wireless LAN design guide explains that RFP and site acceptance boundary.
Do not treat OPC UA, MQTT and Sparkplug as synonyms
The OPC Foundation reference page lists OPC 10000-200 Industrial Automation release 1.01.4, published 23 May 2025, as current. OPC UA can carry typed industrial information, quality and security features, but “OPC UA supported” does not prove interoperability. State client/server or PubSub, profiles, authentication, certificates, information model, load, history, alarms and cross-vendor test cases.
OASIS MQTT 5.0 became an OASIS Standard on 7 March 2019. It is a lightweight client-server publish/subscribe messaging transport, payload-agnostic and offering three qualities of service. QoS does not define asset identity, unit, quality, birth/death state, business uniqueness or historical replay. Encryption, authentication and authorisation are implementation choices that must be specified.
Eclipse Sparkplug 3.0 is the first release managed under the Eclipse Foundation specification process. Eclipse says it formalised version 2.2, clarified ambiguity and added explicit normative statements without adding features that break existing devices and applications. It can standardise an MQTT topic namespace, payload and state model, but it is not mandatory for every MQTT project. Check compatible implementations, the boundary gateway and legacy constraints.
| Option | Mainly defines | Common misconception | Acceptance focus |
|---|---|---|---|
| OPC UA | industrial access and information model | a logo aligns semantics | profile, model, certificate and load |
| MQTT 5.0 | publish/subscribe transport | QoS guarantees business uniqueness | duplicate, order, reconnect and authorisation |
| Sparkplug 3.0 | topic, payload and state over MQTT | always required with MQTT | birth/death, metrics and compatibility |
| CSV/API | batch or application exchange | simple transfer means simple operations | version, encoding, replay and audit |
Make the tag dictionary and data contract the source of truth
Raw names such as D100, M201 or Temp1 do not scale. For every tag, record a stable ID, display name, asset hierarchy, source address, type, unit, scaling, quality, expected range, cycle, change trigger, timestamp source, retention, owner and change reason. Version its relationship to equipment, product, lot, process and shift.
A data contract defines mandatory fields, nulls, quality, timestamps, timezone, sequence, event ID, schema version, corrections, replay and notice of breaking changes. Test PLC shutdown, sensor break, gateway restart, clock correction, network loss and platform rejection—not only valid values.
Define which clock is authoritative, where source timestamps are created, how backward clock steps are handled and how loss of synchronisation is detected. A clock difference across a PLC event, barcode scan, quality measurement and MES transaction can change the conclusion of a traceability investigation.
Design for missing, duplicate and out-of-order data
Networks fail. The requirement is to detect the outage, retain an agreed amount at the edge, replay safely, handle duplicates at the consumer and expose remaining gaps. Use sequence numbers, event IDs, source and ingest times, quality and retry count according to the use case.
As an illustrative design scenario, take 40 machines and 1,200 tags: 200 critical tags at one second, 1,000 at ten seconds, and a 24-hour offline buffer. These are not standard values. Size with measured payloads, change rate, compression, metadata, storage endurance and replay duration; test at 80%, 100% and 120% of the agreed workload. Count daily records and burst/replay load, not only tags.
Expose completeness, latency, clock offset, range exceptions, frozen values, bad quality, duplicates, order inversion, unmapped tags and schema violations. Separate a process anomaly from an acquisition-system failure so that the data owner can decide which periods are fit for use.
Put OT security inside the RFP
NIST SP 800-82 Rev.3, final in September 2023, guides OT security while recognising its performance, reliability and safety needs. For manufacturing data collection, design asset and flow visibility, segmentation, least privilege, remote access, logging, change, backup and incident response with the connection—not after it.
Map destination, direction, port, protocol, certificate, DNS, time and update services. Avoid shared administrators, permanently enabled vendor VPNs and service accounts that can traverse the plant. Require individual identity, approval, time limits, session records and closure.
Test rejected unauthorised flows, expired certificates, clock faults, account disablement, log-forwarding failure, restore and spare replacement. Do not run an unplanned active scan against production equipment; agree the safe method with the machine vendor and IT/OT owners.
Run a 90-day proof, not a connection demonstration
Ninety days is an illustrative project design, not a standard duration. Adjust it to include representative products, shifts and maintenance windows.
Days 0–30: baseline and design conditions
Define the decision, equipment, tags, current records, outages, network, clocks, access and support model. Select representative assets across PLC generations and constraints. Approve the first tag dictionary, data flow, risk register, success criteria and exclusions before expanding.
Days 31–60: technical proof and controlled failure
Connect the selected PLCs, sensors, logger or gateway. Plan safe tests for cable loss, upper-platform outage, clock drift, power loss, storage pressure and certificate failure. Record buffering, replay, deduplication, alarms, monitoring and clean-device restoration. Every fault injection needs approval and rollback.
Days 61–90: controlled production use
Use the data in at least one shift, maintenance or quality workflow. Review quality and usage daily; correct bad reason codes, missing context, clock offsets and operator burden. Repeat candidate acceptance tests, then decide Continue, Modify or Stop. Success means an accountable owner can approve the scale conditions from evidence.
| Phase | Main deliverables | Gate |
|---|---|---|
| Baseline | use case, tag dictionary, data flow, risk | approve scope and pass/fail |
| Technical proof | buffer, security and restore records | approve controlled production |
| Controlled production | KPI, usage, defects and actions | scale, modify or stop |
Write an RFP that makes bids comparable
Give every bidder the same scope, assumptions, exclusions, deliverables and tests. “IoT platform package” hides licences, tag limits, integration, security, backup and support.
| RFP section | Required content |
|---|---|
| Purpose and scope | decision, assets, plants, outage windows, exclusions |
| Current state | PLC/sensor/network/application inventory and drawing confidence |
| Data | dictionary, time, quality, retention, schema and API |
| Non-functional | performance, completeness, buffer, recovery and scale |
| Security | flow, identity, certificates, patches and remote access |
| Migration | pilot, parallel run, rollback and record reconciliation |
| Testing | FAT, SAT, fault, load, restore and business acceptance |
| Handover | configuration, source, licences, backup, runbook and training |
| Cost | initial, annual, added tags, upgrades, travel and downtime work |
Require bidders to classify each answer as standard function, configuration, custom development, third-party dependency or roadmap. Compare Thai on-site response, support language, spare parts, end-of-support notice and export of your data.
Compare a three-year TCO, including operational change
The following is an illustrative recommended calculation, not a market quotation, BOI-eligible amount or promised outcome.
| Cost item | Initial example | Annual example | Check |
|---|---|---|---|
| gateways, loggers and sensors | THB 420,000 | THB 30,000 | spares, calibration, replacement |
| integration and dashboards | THB 650,000 | THB 90,000 | changes, added tags, APIs |
| network and security | THB 180,000 | THB 30,000 | certificates, monitoring, updates |
| training and FAT/SAT | THB 120,000 | THB 30,000 | retraining and recovery exercise |
| contingency | THB 130,000 | THB 0 | legacy differences and off-hours work |
| Example total | THB 1,500,000 | THB 180,000 | replace with project quotes |
The three-year example is 1,500,000 + 180,000 × 3 = THB 2,040,000. On the benefit side, separate cash realised, capacity redeployed and risk avoided. Do not double count labour time, downtime, scrap and throughput, and do not call a viewed dashboard a realised benefit without a changed action.
Thailand BOI’s current Smart and Sustainable Industry page describes, subject to eligibility, a minimum efficiency-enhancement investment of THB 1 million excluding land and working capital, machinery import-duty exemption, and a three-year corporate income tax exemption for existing projects capped at 50% of eligible improvement investment. It describes a 100% cap when machinery linked to the domestic automation industry represents at least 30% of the machinery, automation or robotics value. This example is not automatically eligible. Confirm the current activity, timing, cost categories, evidence and pre-approval purchasing treatment with BOI or an adviser before commitment.
Use FAT/SAT to create reproducible acceptance evidence
FAT uses a representative PLC or simulator, gateway, broker, database, dashboard, identity and monitoring in a controlled environment. Test tag mappings, types, units, quality, timestamps, load, disconnect, replay, duplicate handling, certificates and restore. Move actual factory RF, machine load and production conditions explicitly to SAT.
SAT uses real machines, networks, shifts and representative products. A screenshot of changing values is insufficient. Retain a test ID, prerequisites, input, expected and actual result, timestamps, logs, configuration version, operator, deviation, retest and approval.
| Example test ID | Scenario | Passing evidence |
|---|---|---|
| FAT-DQ-01 | type, unit and quality mapping | dictionary version reconciled to records |
| FAT-BUF-02 | upper-layer outage and near-full buffer | gaps, duplicates, replay time and alarm log |
| FAT-SEC-03 | unauthorised client and expired certificate | rejection and audit trail |
| FAT-RES-04 | restore to a clean spare gateway | approved config, hash, connection and monitoring |
| SAT-OPS-01 | peak-shift collection | completeness, latency, business view and exceptions |
| SAT-REC-02 | replacement using the maintenance runbook | duration, work record, issues and sign-off |
Illustrative candidates include 100% mapping of mandatory tags, at least 99.5% completeness in agreed windows, clock offset within one second for traceability events, replay from a 24-hour buffer and spare restoration within four hours. They are not standard limits; replace them where safety, quality, law or customer requirements are stricter.
Make operations and backup restorable by the plant
NIST SP 1339, the final OT Backup Quick Start Guide published in June 2026, says effective OT backup management integrates backups into change management, creates them regularly, tests them and reviews them during recovery exercises. A backup that has never restored is inventory, not recovery evidence.
Include gateway configuration, PLC communication settings, tag dictionaries, schemas, dashboards, certificate-related material, licences, installation media, network/firewall settings, runbooks and contacts. Store secrets in an approved secrets system, not clear text in a common document. Define owner, frequency, location, encryption, retention, a separated copy and restore responsibility.
Link a pre-change backup and restore point to every change ticket. A quarterly exercise can be a recommended operations example, but important changes, device replacement and certificate work may require additional proof. Have plant personnel start from a clean spare and restore connectivity, monitoring and replay using only the handed-over runbook.
Monitor gaps, time, disk, certificate expiry, connections, CPU, temperature, firmware, vulnerability status, backup outcome and unauthorised changes. Separate first response, vendor escalation, on-site dispatch and production decisions, including the Thai/English/Japanese communication path.
Scale with templates and controlled differences
Do not copy a successful pilot blindly to 100 machines. Build reusable connection kits, dictionary templates, security zones, FAT records, SAT checklists and restore runbooks. Record each machine’s differences and approved exceptions.
Before each rollout wave, recalculate data quality, tickets, storage, licences, gateway capacity, network and support load. Fix unknown tags, clock drift and operator burden from the previous wave before adding more assets. Prioritise business value and maintainability, not the machine count.
FAQ: PLC, equipment and sensor data collection
Where should manufacturing data collection begin?
Begin with one decision, a representative process and the minimum useful tags. Define users, deadline, current record and consequence of error before selecting technology.
Can PLC data collection affect control?
Yes, depending on method and load. Specify read-only access, allowed tags, connections, polling, timeout and reconnection, then test representative load. Separate write-back and confirm machine warranty and safety boundaries.
Is a one-second cycle enough for equipment data collection?
There is no universal cycle. Stops, energy, condition monitoring and raw vibration require different rates. Derive it from the phenomenon and decision, then test event bursts, storage and network effects.
Does sensor data collection create predictive maintenance automatically?
No. Mounting, calibration, operating context, maintenance history and a failure definition are required. Manage false alarms, missed detection, model version and response ownership.
Should a factory choose a data logger or IoT gateway?
Use a logger for isolated or short-term measurement; consider a gateway for continuous multi-asset acquisition, protocol conversion, buffering, certificates and monitoring. Compare the integration exit and lifecycle.
Does MQTT QoS 2 eliminate every duplicate?
Protocol delivery semantics and business-record uniqueness are different layers. Test client, bridge, store and consumer retries; use event IDs and idempotent processing where required.
Will two OPC UA products connect immediately?
Not necessarily. Verify profile, policy, certificates, namespace, model, data types, licences, tag load, update rate, alarms and history on the exact versions.
How should costs be compared?
Compare hardware, integration, licences, network/security, downtime work, training, backup, support, upgrades, added tags and migration on one three-year TCO basis. Separate cash, capacity and risk benefits.
What is the difference between FAT and SAT?
FAT proves configuration, mapping, faults, security and restore early in a controlled environment. SAT proves the system with actual machines, networks and production conditions. Both need reproducible records and sign-off.
Can a data collection system receive BOI incentives?
It may be possible, but not automatic. Confirm activity, investment category, minimum amount, eligible costs, domestic-automation linkage, application timing and evidence against current official requirements before ordering.
Conclusion: procure trustworthy use and recovery, not connections
Manufacturing data collection succeeds when the plant can trust, use and restore the data. Derive tags, context, quality, time, buffering, security and operations from a decision. Select PLC, sensor, logger and gateway roles deliberately; specify how OPC UA, MQTT or Sparkplug will be tested.
A 90-day pilot should break and restore the system, not merely keep a dashboard online. Comparable RFPs, a complete TCO and evidence-based FAT/SAT turn a demo into an operating capability.
You can start with use-case definition, site survey and an RFP even before every PLC and protocol is known. For a manufacturing data collection project in Thailand, contact TOMAS TECH at the English enquiry page from the concept stage.
Primary references
- ISA, ANSI/ISA-95.00.01-2025 update: https://www.isa.org/news-press-releases/2025/april/update-to-isa-95-standard-addresses-integration-of
- OPC Foundation, OPC 10000-200: https://reference.opcfoundation.org/specs/OPC-10000-200
- NIST SP 800-82 Rev.3: https://csrc.nist.gov/pubs/sp/800/82/r3/final
- NIST SP 1339: https://csrc.nist.gov/pubs/sp/1339/final
- OASIS MQTT 5.0: https://docs.oasis-open.org/mqtt/mqtt/v5.0/mqtt-v5.0.html
- Eclipse Sparkplug 3.0: https://sparkplug.eclipse.org/specification/version/3.0/
- ISO 22400-1:2014: https://www.iso.org/standard/56847.html
- Thailand BOI Smart and Sustainable Industry: https://www.boi.go.th/index.php?language=en&page=smart_sustainable
