IMD SolutionComputing LibraryShop main store

Selection guide

Manual lab data entry into a historian: choosing the write path

Edge computing and IIoT Published 8 min read

Short answer

Put every measured variable into its own timestamped PI point and write it from inside PI Vision with a manual-entry symbol; keep specifications, qualifiers, blanks and notes as separate AF fields, not in the value. The load is only 1,800 values per day, so the decision rides on the write path, authentication and per-operator write rights, never on avoiding the 300 points.

On this page
  1. What the entry path must satisfy
  2. The write paths compared
  3. Why the Asset Framework table is not the history store
  4. Storing values, specifications, qualifiers and blanks
  5. The 100% total check, with numbers
  6. The load arithmetic decides the point count
  7. Write rights, identities and the firewall
  8. Pilot it on one variable first
  9. The decision, case by case
  10. Questions asked

Commit to the archive: every measured variable gets its own timestamped PI point, written from inside PI Vision through a manual-entry symbol, with specifications, qualifiers, blanks and notes held in separate AF fields instead of inside the value. At 300 variables sampled every four hours that is 1,800 values per day, so avoiding PI points buys nothing and costs the trend. The selection turns on three things the entry interface must satisfy: a write path that crosses your firewalls on an already-approved service, an authentication chain that delivers the operator's identity, and per-operator write rights enforced at the archive. A hand-edited Asset Framework table fails the first test: it holds the current row, not history.

What the entry path must satisfy #

The workflow is fixed, so the interface has to clear all of it:

  • One timestamped historical series per measured variable, roughly 300 of them, with the list changing as substances are added or retired.
  • Four-hourly samples, so the sample time and the time the shift actually keys the value are two different times.
  • Specifications such as <10 ppm carried beside the results: a comparison operator, a number and a unit together.
  • Blanks: some samples are genuinely not taken, and they must never look like zero.
  • Free-text notes and corrections attached to a sample.
  • Writes restricted to authorized staff; everyone else reads only.
  • The operator sees the mass total at 100% before submitting, in one system, not several.

And the constraint that decides it: the office network is separated from the server network by two firewalls, and the only approved services are DataLink, PI Vision, and PI System Explorer including Asset Analytics.

The write paths compared #

Write path What it writes Firewall fit Write control Decision
Manual-entry symbol in PI Vision (community symbol or supported suite) Point values straight into the PI Data Archive from the display Rides the approved PI Vision channel Point-level write security for the identity PI Vision authenticates First choice when one screen is required; prove the identity chain with one test point first
Excel workbook plus DataLink writes Archive values written from workbook cells Rides the approved DataLink channel Rights of the DataLink user plus workbook access control Fallback when the symbol cannot authenticate or is not funded; the operator then works in two systems
Hand-edited AF table in PI System Explorer The current row of a lookup table Rides the approved PSE channel AF database write access in PSE Reject for measurements; keep it for specifications and lookups
Purpose-built entry application with its own server-side writer Timestamped values from its own service Needs a new rule through both firewalls Its own service and user accounts Reject under the stated constraints; revisit only if the first two fail

The symbol route is real and documented: a PI Vision manual-data-entry symbol is a custom symbol that lets a user enter point values directly into the PI Data Archive from inside the display, and supported third-party versions exist on the vendor's marketplace. The reported catches for the supported suite are a recurring per-feature fee and a prerequisite for Kerberos identification.

The DataLink route needs no new network permission at all: the channel is already approved office-to-server, and DataLink can write values into the archive from the workbook.

Why the Asset Framework table is not the history store #

AVEVA's documentation on table lookup data references settles it in one line: the reference "is not optimized to treat data in an external table as a time series and query for it." A lookup answers what the current row for a key is, which is reference data, not history. A display bound to it ends up exactly where a display does when the archive stores nothing: latest state, no trend — the failure behind why PI shows live values but no trend history. Put specifications, target ranges and substance lists in AF, where they belong; put the measurements in points.

Storing values, specifications, qualifiers and blanks #

Information Where it lives Effect on the trend
Numeric result PI point carrying value, sample timestamp and engineering unit Trends, alarms and calculations work like any other tag
Specification such as <10 ppm AF attribute set: operator, threshold, unit, effective from and to dates The limit displays beside the value without overwriting it
Result reported as below a limit Numeric point value plus a qualifier state or string tag The below-limit report keeps its meaning
Blank or not sampled An explicit digital state, or a logged non-entry A gap never reads as zero
Free-text note A comment stream per sample set, or per variable where needed Context sits beside the value, not inside it
Sample time versus entry time Two timestamps in ISO 8601 form with explicit UTC offset A late entry cannot masquerade as on-time sampling

ISO 8601 requires the full date, the time and the zone designator together, which is what keeps a 06:00 sample keyed after a daylight-saving change unambiguous. The sample timestamp you submit, not arrival time, is what positions the value on the trend — the same discipline as writing timestamped history back into a SCADA server. Some of these limits are permit or exposure limits, and effective-dated specifications are how you prove later which limit was in force when a sample was taken.

The 100% total check, with numbers #

For one-decimal mass percentages, each entry carries up to 0.05 percentage point of rounding error, so the acceptance band is entries × 0.05. Five components give 5 × 0.05 = 0.25 percentage point, and the check accepts anything from 99.75 to 100.25.

A passing shift: 41.2 + 28.5 + 16.1 + 9.7 + 4.5 = 100.0 mass % — submit. The same shift with 8.7 keyed instead of 9.7 totals 41.2 + 28.5 + 16.1 + 8.7 + 4.5 = 99.0 mass %, which is 1.0 percentage point low and outside the 0.25 band — block the submission and highlight the row. Compute the sum in the entry surface where the operator sees it before submitting; the archive has no idea what a total is.

The load arithmetic decides the point count #

At one sample per variable every four hours, each variable produces 24 / 4 = 6 values per day. Across 300 variables that is 300 × 6 = 1,800 values per day, and 1,800 × 365 = 657,000 values across a full year. The point budget is 300 measurement points, qualifier tags only for variables reported as below a limit, and one comment stream per sample set; specifications are AF attributes and consume no archive points.

This is why avoiding PI points is the wrong objective: the volume is trivial, and each point then trends, alarms and feeds calculations like any other tag. The changing substance list is maintenance either way — a table you hand-edit is not less work than a template you copy. Archived series are the whole reason to take the historian route over a live tag server, the trade laid out in do you need a historian or just an OPC tag server.

Write rights, identities and the firewall #

  • The write must ride an approved service: the symbol on the PI Vision channel, the workbook on the DataLink channel. Anything else is a new rule through two firewalls and, under the stated constraints, out.
  • A Vision-based write leaves as the identity the authentication chain delivers. The Kerberos prerequisite reported for the supported suite is one instance of that rule. If users authenticate only to the web server and the archive sees one service account, per-operator write rights collapse into that single identity — either fix the chain or scope that identity's rights deliberately.
  • Enforce authorization at the archive: a read mapping for everyone, a write mapping for the entry group, restricted to the lab points. Hiding the entry control is not access control.
  • Prove it once: one limited operator account writes one test point, then read back the value, its timestamp and who the archive records as the writer.

Pilot it on one variable first #

  1. Inventory the variables: unit, numeric or qualified result, blank allowed, specification source, in or out of the total.
  2. Build one element template — value point, qualifier, spec attributes with effective dates — and instantiate one variable.
  3. Put the entry symbol on one display and submit a value as a limited operator; confirm value, timestamp and write attribution.
  4. Force the failure cases: a blank, a below-limit result, an out-of-tolerance total. Each must be visibly rejected or recorded as a state.
  5. Change one threshold with a new effective date; the display shows the new limit, the old sample keeps the old one.
  6. Roll out with a fixed naming and metadata scheme, so a new substance becomes a template copy.

The decision, case by case #

  • One screen, no new spending, no new firewall rules: the manual-entry symbol in PI Vision writing the 300 archive points, specifications and qualifiers in AF, blanks and the 100% total enforced in the symbol — after the one-point authentication test passes.
  • The authentication chain cannot deliver per-user identity and will not be changed: write from the Excel workbook through DataLink over the approved channel, with the total check and validation in the workbook; the operator accepts two systems.
  • Budget and support requirements arrive: evaluate the supported marketplace suite, or the platform vendor's dedicated manual-entry client, against the same firewall and identity tests before committing.
  • Never: measurement history as hand-edited Asset Framework table rows. The table keeps specifications and lookups; the archive keeps the measurements.

Questions asked

Do 300 manual variables really need 300 PI points?

Yes, and it is the cheap part: 300 points at six four-hourly samples per day is 1,800 values per day and about 657,000 across a full year, volume an archive handles without noticing. Build them from a single element template so adding a substance is a copy, not a redesign. Resisting the points is what pushes you toward a table that cannot trend.

Can the manual-entry symbol work without Kerberos?

Test it before believing either yes or no. The symbol writes as the identity your PI Vision authentication chain delivers, so if users are only authenticated to the web server and the archive sees a service account, per-operator write rights collapse into one identity. Submit one value from a limited account to one test point; if the write fails or arrives as the wrong identity, fall back to the DataLink workbook over the already-open channel.

Why is an Asset Framework table the wrong store for results?

A table row is current state. The vendor's documentation on table lookup data references states they are not optimized to treat external table data as a time series you query over time, so PI Vision gets the latest row, not a trend. Keep specifications and lookup lists there; keep measurements in archive points.

How should a '<10 ppm' specification be stored?

Split it into four fields: comparison operator, threshold of 10 ppm, unit, and effective dates. The measured result stays a number in its own point, with a qualifier state when it is reported as below a limit. Effective dating is what lets you prove, years later, which limit applied when a sample was taken.

How tight should the 100 percent total check be?

For one-decimal mass percentages, each entry carries up to 0.05 percentage point of rounding error, so the tolerance is entries times 0.05: five components give 0.25, and anything outside 99.75 to 100.25 blocks submission. Compute it in the entry symbol or workbook where the operator sees it, not in the archive after the fact.

Keep reading

  • Comparison

    Move DeltaV Continuous Historian data into a PI system

    When PI points exist, collect DeltaV live values with OPC DA and recover history with OPC HDA; when points and AF structure are missing, let the DeltaV Smart Connector create them from EntHist.xml. Add PI-to-PI only for a second office-LAN server. Prove buffering, service identity and archive coverage by test before the bulk load.

    Edge computing and IIoT

  • How-to

    Trace a PLC tag from controller to historian display

    Trace one existing, noncritical measurement end to end: controller source, collector mapping, historian point, optional calculation, then display binding. Check the value and the timestamp at each boundary before you change any configuration. A value that matches and a timestamp that agrees at every step means the path is sound; the first mismatch names the layer to fix.

    Edge computing and IIoT

  • Selection guide

    Choosing a PI System to Oracle integration route with validation

    With read-only PI access and cleansing owned by the receiving side, pull AF assets and status with the PI OLEDB Enterprise Provider through SSIS into a staging SQL Server, qualify transitions there — reject momentary interruptions and anything older than 2 hours — then write Oracle. Event Frames and the RDBMS interface need PI-side configuration or tag-based mapping that loses AF context.

    Edge computing and IIoT