A connected circular knitting machine can produce more data than a factory can use. Speed, stop reason, yarn break, alarm, production counter and energy values may exist in separate controllers, spreadsheets or operator notes. The hard part is not collecting another signal. It is preserving a trustworthy history that can be filtered by machine, shift, yarn, order and time without losing the meaning of the original event.
That is the job of a production data historian layer. It may be a small industrial computer with local storage, a plant server, a cloud service or a combination. The right choice depends on retention, network reliability, machine interfaces, cybersecurity and who needs the data. Cost comparisons become useful only after those boundaries are written down.
This article focuses on the historian and gateway layer for circular knitting factories. It does not replace the existing MES integration and OPC UA data-mapping checklist, which addresses production-system integration. A historian can feed an MES, but it can also be deployed first to preserve machine data while the wider integration remains under review.
What a historian must preserve
A value without context is difficult to audit. “RPM 28” is not enough unless the record says which machine produced it, when it was sampled, which recipe was active and whether the controller reported a valid value.
Define the minimum record before choosing hardware or software.
| Data field | Minimum context | Failure if missing |
|---|---|---|
| Machine ID | Stable asset ID and physical location | Trends from two machines can be mixed |
| Timestamp | Time zone, clock source and precision | Events cannot be aligned with orders or alarms |
| Signal name | Meaning, unit and source address | A value can be interpreted incorrectly |
| Quality flag | Valid, stale, substituted or unavailable | Bad data can look like production data |
| Recipe/order | Style, yarn and lot reference | A defect cannot be linked to the run |
| Event state | Start, stop, alarm, acknowledgement or reset | Downtime duration becomes ambiguous |
| Source version | Controller or mapping version | Changes become invisible after retrofit |
The historian should retain raw or minimally transformed values alongside calculated KPIs. If an OEE dashboard shows a number that cannot be traced back to the events used to calculate it, the dashboard is a presentation layer, not an evidence system.
Gateway architecture options
A gateway moves data between the machine network and the historian. It can be a protocol converter, an edge computer or a software service on an existing industrial PC. Compare architectures by what they do when a machine or network is unavailable.
Direct polling
A gateway polls a controller or data source at a defined interval. This is simple when the interface is stable, but it may miss short events if the polling interval is longer than the event itself. Ask the supplier whether the controller supports event buffers, change-of-value messages or a sequence number.
Message-based collection with MQTT
MQTT can publish values and events through a broker. It is useful when many machines send data to a common collection layer, but topic names, payload schema, retained messages, quality flags and authentication need to be designed. A topic tree that starts cleanly can become difficult to maintain when a plant adds a second gauge, line or site.
Specify the payload before accepting a quote. Include asset ID, signal, timestamp, value, unit, quality and source sequence. Do not rely on a dashboard screenshot as an interface specification.
File or CSV export
CSV export is often attractive for a pilot because it can work with existing controls and spreadsheet-based workflows. The limitation is not the comma. It is the loss of event meaning if the file has no stable headers, time zone, unit, machine ID or correction history.
If CSV is used, require a versioned header, one row definition, decimal convention, delimiter, encoding, file naming rule, export frequency and retry behaviour. State whether the export is a snapshot or an append-only event log. These details determine whether a file can be loaded into a historian without manual repair.
Define the signal list before asking for cost
A supplier cannot quote a meaningful gateway by machine count alone. Two factories with ten machines may have very different work if one needs four counters and the other needs alarms, recipes, energy and quality states.
Start with a signal register.
| Signal group | Example fields | Questions for supplier |
|---|---|---|
| Production | Revolutions, courses, roll length or counter | Is the value cumulative, resettable or calculated? |
| Machine state | Running, idle, stop, setup, fault | Is the state source-defined or inferred? |
| Yarn | Feeder status, break event, lot reference | Can the event be captured without polling loss? |
| Quality | Defect code, inspection result, roll reference | Who owns the code dictionary? |
| Energy | Meter value, interval consumption, power state | Is the meter integrated or a separate device? |
| Maintenance | Alarm, intervention, part change and acknowledgement | Can the event link to a work order or spare part? |
| Environment | Temperature, humidity or dust where relevant | Is the sensor calibrated and time-aligned? |
For a machine retrofit, include the physical interface, controller model, firmware, available ports, network zone, spare cabinet space and electrician/controls-engineer responsibility. A gateway quote that excludes panel work, network segmentation or commissioning can appear cheap while leaving the main integration work outside the purchase.
MQTT and CSV acceptance requirements
MQTT checklist
Ask the supplier to document the broker location, authentication method, topic naming, payload schema, retained-message policy, quality-of-service setting, offline buffer, replay behaviour and certificate renewal process. Require a test message and an example payload with realistic timestamps and quality flags.
The gateway should not publish a “zero” when the machine is offline unless the value is explicitly marked as substituted or unavailable. A zero speed and no speed reading are different events. The historian must preserve that difference.
CSV checklist
For CSV, require a sample file containing a start event, a stop event, an alarm, a missing value and a correction. Check whether timestamps use local time or UTC, whether daylight-saving changes are possible in the target plant and whether the file remains readable after a power interruption.
A file-based pilot is acceptable when the factory agrees on the limits. It should not be described as real-time control or complete event capture if the export is periodic.
Cost model: separate the layers
The phrase “gateway cost” hides several different purchases. Build a budget with separate lines so that a supplier can explain what is included.
| Cost layer | Typical scope | Buyer question |
|---|---|---|
| Hardware | Edge PC, protocol adapter, storage, UPS or panel parts | What temperature, vibration and service conditions are supported? |
| Connectivity | Ethernet, serial interface, OPC UA, MQTT or file export | Which machine protocols are included in the base quote? |
| Historian | Database, tags, retention, backup and user access | Is pricing by machine, tag, storage or user? |
| Engineering | Mapping, state model, dashboard and commissioning | How many signals and test scenarios are included? |
| Cybersecurity | Certificates, accounts, firewall rules and patching | Who owns renewal and incident response? |
| Data quality | Clock sync, validation, missing-data rules and audit log | How are bad or late values marked? |
| Support | Remote service, on-site visit, upgrades and training | What is the response time and what is excluded? |
| Recurring fees | License, cloud storage, broker, support or backup | What is the three-year cost, not just the first invoice? |
Use a worksheet rather than copying a market price. Hardware, machine protocol and signal count vary too much for a universal circular-knitting gateway price. The supplier should quote a pilot, a production rollout and the recurring operating cost separately.
Retention and ownership decisions
Retention is a production decision. Keep high-resolution data for a short diagnostic window if storage is limited, then retain lower-resolution aggregates for longer trend analysis. Do not discard raw events before deciding which quality and maintenance questions the factory needs to answer.
Define who owns the data, who can export it, how backups are tested and what happens when the supplier relationship ends. A proprietary dashboard is not the same as portable data. Require an export in an agreed format and a documented mapping so that a future system can reconstruct the machine and signal identity.
The circular knitting machine repairability documentation checklist is a useful adjacent reference because maintenance history and parts changes become more valuable when they can be joined to machine events. Keep the join key stable: machine ID, event time and work-order or part reference.
Commissioning and data-quality test
Do not accept a gateway because the dashboard shows moving numbers. Run a commissioning test that creates known events on the machine and checks whether the historian preserves them.
Test at least:
- Machine start, normal stop, emergency stop and restart.
- Yarn break or feeder alarm with the correct asset ID.
- Recipe or order change.
- Power interruption and gateway restart.
- Network interruption and recovery.
- A missing signal and a stale signal.
- Clock offset between controller, gateway and historian.
- Duplicate message or repeated CSV row.
- Manual correction and the audit record for that correction.
Record the expected event time, received time, stored time, quality flag and recovery result. A lost alarm during a network interruption is a design issue, not a training issue.
For machine performance context, the site’s circular knitting machine condition-monitoring checklist can help connect data signals to mechanical checks. The historian should expose the evidence; it should not claim that a signal alone proves cam wear, cylinder alignment or needle-track damage.
Vendor comparison questions
Ask each supplier to answer the same questions:
- Which machine controllers and protocols have you integrated in production?
- Can you show a real payload or export with timestamps, units and quality flags?
- What happens during a network or power outage?
- How are machine IDs, signal names and mapping changes governed?
- Can the factory export raw data without the supplier’s application license?
- What is included in commissioning and how many test cases are covered?
- Which cybersecurity controls are mandatory before connection?
- What are the one-time, recurring and decommissioning costs?
A useful quote includes exclusions. If the supplier does not cover electrical installation, network changes, controller programming, clock synchronization or historical-data migration, the buyer should see those items before approval.
FAQ
Is MQTT better than CSV for a circular knitting factory?
Neither is automatically better. MQTT suits event and message distribution when the broker, topics, payload and offline behaviour are controlled. CSV can be a practical pilot when the header, timestamp and correction rules are stable. Choose based on the factory’s signal, network and support requirements.
What does a production data historian store?
It stores time-stamped machine and process records with context such as asset ID, signal name, unit and quality status. It can also retain events, recipes, alarms and calculated values. The exact scope should be defined in the signal register and retention policy.
How much does a gateway cost?
There is no defensible universal price from machine count alone. Cost depends on protocols, signal count, hardware, panel work, software, storage, cybersecurity, commissioning and recurring support. Request separate pilot, rollout and three-year operating-cost lines.
Can a historian replace MES?
A historian preserves and serves time-series or event data. An MES usually adds production planning, dispatch, genealogy, quality workflows and other operational functions. A historian may feed an MES, but it does not automatically replace those workflows.
Conclusion
Specify the data model before buying the gateway. Give every value an asset, timestamp, unit and quality flag; separate MQTT, CSV and controller responsibilities; price hardware, engineering, cybersecurity and recurring service as different lines; then test outages, clock offsets and missing events before acceptance. The result is a production record that can support maintenance and quality decisions instead of another dashboard full of unexplained numbers.
References
This 2026 industry article describes the move from isolated textile machines toward connected production linking machines, sensors, materials, workers and digital platforms.
VDMA identifies digitalisation, Industry 4.0, regulations, sustainability and textile-machinery services as active sector workstreams.
The MQTT specification source provides the protocol context for topic, message and quality-of-service decisions in a gateway design.
The OPC Foundation overview provides the industrial interoperability context for secure machine-data exchange and information modelling.
The Commission page shows why traceability, product information and economic-operator data are becoming more important across textile value chains.
This 2026 report provides current examples of automation, digitalisation, recycling, resource efficiency and yarn-tension monitoring in textile technology.
The ISO reference supports the use of controlled records, change management, nonconformity handling and continual improvement for production data systems.
CISA provides the cybersecurity context for connecting industrial control environments and defining access, segmentation and incident responsibilities.