Back to all posts
Plant Data7 min read·October 26, 2026

How Often Should You Sample Machine Data?

TB
ThynkBlox Automation Team
Automation Engineering

Choose an industrial data sampling interval by identifying the shortest event the record must reveal, then testing whether your collection method captures it reliably. Give slow process trends and brief machine states separate requirements. Specify where values are sampled, how events are recorded, and how much controller and network load the design may add.

What must the data help you decide?

Start with a question someone needs to answer. “Collect machine data” is too broad to guide an interval.

Useful questions include:

  • Did the process drift outside its operating range?
  • Which machine state began the stoppage?
  • How many parts were completed during the shift?
  • What happened immediately before the fault?

For each question, record the shortest relevant event, the timing precision needed and the consequence of missing it. Ask the people who will use the record to review those choices.

For production reporting, agree what counts as running, stopped and producing before collecting those states. Our guide to what an OEE number hides explains why those definitions deserve attention.

Which interval are you actually choosing?

Separate four settings in the specification:

SettingQuestion to answer
Source updateWhen does the instrument or controller update the value?
AcquisitionWhen does the collector read or receive it?
StorageWhich readings or events become retained records?
DisplayHow often should the screen refresh?

Reading faster than a source updates does not create new measurements. A frequently refreshed dashboard also does not prove that the stored history has equivalent detail.

Ask the supplier to document each stage. For every timestamp, specify whether it represents measurement, detection, receipt or storage.

For trends, request units, timestamps and data-quality indicators. For state changes, request the previous state, new state and event time where supported.

How should you handle slow process trends?

For a slow trend, choose an initial interval around the decision it supports. A shift review and an investigation into a short process excursion should have separate acceptance criteria.

Illustrative example: a tank-temperature review needs to distinguish a sustained change developing over ten minutes. Start a trial with readings every thirty seconds. This is a proposed test setting, not a general recommendation for temperature measurements.

Compare the trial record with a temporary reference capture at finer resolution. Check whether the proposed interval preserves the start, peak and recovery of the changes that matter.

Review storage settings too. A deadband can suppress small changes, while aggregation can replace individual readings with summaries. Either can change what an investigator sees later.

Specify how the record should distinguish a steady value from missing communication.

If a slow trend shows a control loop cycling, our guide to why a PID control loop keeps oscillating explains which trends to capture.

How do you capture brief machine states?

Define the shortest state you must retain before choosing a polling interval. Include its start time, end time and required duration accuracy.

Illustrative example: a machine enters a fault state for two hundred milliseconds. A collector reads the current state once each second. If the fault starts and clears between those reads, neither read contains the fault state. This illustrates a limitation of snapshot polling, not measured plant performance.

Faster polling may improve visibility, but reliability still depends on source updates, scheduling delays and communication behaviour.

Where every transition matters, consider recording transitions at the source and transferring those records. For counts, consider a source counter rather than reconstructing production from individual pulses.

Keep state definitions unambiguous. If an operator needs to distinguish a fault from waiting for material, agree how both conditions are represented.

When should you use event-driven capture?

Consider event-driven capture when the requirement is to retain a transition, completed cycle, alarm occurrence or recipe change.

Ask where the event is detected and retained. Some subscriptions report changes found by a server's sampling process; they do not necessarily preserve every physical transition. Queue limits and communication outages can also affect delivery.

Put these questions into the design review:

  • What creates the event?
  • Where is its timestamp assigned?
  • How are event order and duplicates handled?
  • What happens when the connection fails?
  • How is a full buffer or missing sequence reported?

For alarm records, agree the meaning of occurrence, acknowledgement and clearance. Our guide to why operators ignore alarms provides related context for reviewing alarm behaviour.

What does faster polling cost?

Shorter polling intervals increase the number of scheduled read attempts for an unchanged tag set. The resulting controller and network load depends on request grouping, protocol behaviour, response size, retries and the equipment involved.

Do not approve an interval from tag count alone. Ask how tags are grouped into requests and how collectors behave after failed reads.

Illustrative example: changing a group's interval from one second to one hundred milliseconds schedules ten read attempts in the time previously allocated to one. That arithmetic does not establish a tenfold increase in controller utilisation or network traffic.

Test the proposed configuration while normal communication clients are active. Agree acceptable limits with the controls team and equipment supplier before the trial. Include timeout and reconnect behaviour in the review.

Can different tags use different intervals?

Use separate collection groups where their purposes differ. The following is an illustrative planning framework, not a set of universal settings.

Data purposeProposed approachAcceptance question
Slow operating trendTrial periodic samplingDoes it preserve relevant excursions?
Brief machine stateRetained transitions where supportedIs every required transition present?
Production totalSource counter with periodic readsCan totals be reconciled after an outage?
Fault investigationTemporary detailed captureDoes it retain the required sequence?
Operator overviewDisplay refresh chosen for the taskIs the information clear and timely?

Put these decisions in the functional design specification. Include the collection method, timing requirement, retention rule, outage behaviour and acceptance test for each group.

How do you verify an interval before rollout?

Test the requirement under conditions that could expose a missed event.

  1. Create a reference. Use an agreed record with enough detail to assess the proposed collection.
  2. Challenge the timing. Test short events at different points between expected polls.
  3. Check stored records. Inspect timestamps, state order, quality flags and retained values.
  4. Interrupt communication. Check recovery, missing-data indications and duplicate handling.
  5. Observe load. Review controller and network behaviour with normal clients connected.

Define passing conditions beforehand. For example, require every generated test transition to appear in order, with any timing error within an agreed tolerance.

Keep testing within an authorised procedure. Any safety-related design or change should be reviewed with a qualified assessor.

What common mistakes should you avoid?

Avoid choosing one fast interval for the whole plant before defining what each record must show.

Do not treat an apparently smooth chart as proof of complete capture. A chart alone cannot establish whether events were missed between readings.

Do not choose an interval solely from a mathematical rule for waveform reconstruction. That addresses a different requirement from retaining every discrete state transition.

Finally, keep diagnostic acquisition separate from protection requirements. Do not assume an ordinary data collector can provide a protective function; have safety-related requirements assessed explicitly.

What are the frequently asked questions?

Is there one recommended sampling interval?

Use a requirement for each data group. Choose a candidate setting, then verify event visibility, timing accuracy and equipment load.

Should event-driven capture replace all polling?

Choose by purpose. Specify retained events for required transitions, and periodic readings where they meet the trend or reporting requirement.

What should we ask an integrator to deliver?

Request a collection schedule, timestamp definitions, outage behaviour, load observations and test evidence. Require the stored records to demonstrate the questions the system must answer.


Planning plant data collection? Explore our automation services and frequently asked questions, or discuss your requirements.

Next step

Want to apply this to your plant?

Every project starts with a fixed-fee assessment, and you can stop after the report.