TB-7096 · REV B · Technical newsheet

Industrial IoT & MonitoringDevice profile

Dashboards Fail on Data Foundations, Not Charts, Nukon Warns

Nukon's Gareth Williams argues dashboard failures trace to data foundations — naming drift, undefined metrics and mismatched time bases between OT and ERP systems — not chart design.

By Nathan Brooks4 min read717 words

Features

  • One signal can carry multiple tags across a plant, e.g. PUMP3_RUN_ST, Pump 3 Running, P3_Status, breaking cross-site joins and comparisons
  • A PLC logging local time and an ERP posting in UTC can push a full shift's output into the wrong day
  • Two sites can log micro-stops and planned changeovers differently yet both label the result OEE
  • Williams prescribes one metric definition, one named owner, one calculation driving every report
  • Two or more uncertain answers on his six-question checklist indicate data foundations need repair before go-live
Your Dashboard Is Only As Reliable As The Data Behind It
Device photoYour Dashboard Is Only As Reliable As The Data Behind It — prashant hiremath / unsplash

A single uncertain number can derail a manufacturing analytics project months after go-live, and the failure mode is almost never the visual layer, according to Gareth Williams, Business Development Manager at Nukon, a Tetra Tech company within the SAGE Group. The failure mode is data governance.

Writing on dashboard reliability, Williams described the moment of collapse with precision: after months of requirements workshops, build reviews and formatting work, a dashboard goes live, and someone in the first production meeting scans the numbers and says, "I'm not sure that's right." From that point, supervisors, operators and finance teams begin checking the dashboard against spreadsheets, handwritten notes and gut feel. The tool built to create alignment becomes another thing to debate.

Where the conflicting numbers come from

The technical causes are concrete. An OEE figure can differ depending on who calculates it. Finance and operations may pull output figures from different systems. Missing, late or handwritten data does not become trustworthy by appearing in a polished chart. Williams argued these are foundation problems, cheaper to fix before go-live than after, and set out three checks.

What should be standardized before go-live?

Naming. In a typical plant, one signal carries multiple tags: PUMP3_RUN_ST, Pump 3 Running, P3_Status. Across lines or sites, those differences silently break joins, reports and comparisons. The fix is one convention across sites and vendors, a recognizable plant hierarchy, and consistent units and data types — three tags collapsing into one path such as Site1/Packaging/Line2/Pump3/Running. The business side mirrors the problem: if a "production order" in ERP and a "works order" in maintenance are treated as separate entities, the gap typically lands in a translation spreadsheet maintained by one person that nobody trusts.

Context. A temperature reading of 74 means nothing without asset, product, batch, shift and machine state. Williams noted that much of this information already exists in business systems and should be captured at collection time rather than rebuilt by hand. Metric definitions fail on inclusion boundaries, not formulas: whether a planned changeover counts as downtime or scheduled time, and whether micro-stops are logged at all, can be answered differently at two sites — with both sites calling the result OEE. His prescription: define each metric once, give it an owner, and let one calculation drive every report.

System integration. PLC, SCADA and historian data must meet ERP, MES, maintenance and quality data, and they rarely share timestamps, identifiers or update rates. A PLC logging in local time while an ERP posts in UTC is enough to push a shift's output into the wrong day. The targets Williams listed:

  • A single source of truth for each data element, with a named owner
  • Aligned time bases and reliable shared keys linking work order, machine and batch
  • Visible handling of gaps and poor-quality values

Ownership is a people decision before it is a technical one, he wrote: when systems disagree, someone needs the authority to decide which source wins.

What else does governed data buy?

Reliable foundations serve more than reporting, Williams argued. Consistent naming, shared definitions, clear ownership and connected systems become reusable building blocks for process improvement, quality reporting, predictive maintenance and AI initiatives. Treating data practices as core infrastructure means each project strengthens the next instead of adding another reporting layer to reconcile later.

He also supplied a five-minute project-review checklist: whether a documented naming standard exists and is followed; whether context is captured at source or reconstructed; whether each key metric is defined once with a named owner; whether the source of truth behind every dashboard figure is known; how missing or poor-quality data is handled; and whether identifiers match across OT and business systems or are translated somewhere in the middle. Two or more uncertain answers mean the foundations need work before anything goes live. His deployment advice: start with one line or area, prove the approach, then scale.

Nukon operates within the SAGE Group, a certified member of the Control System Integrators Association (CSIA), which works with manufacturers on integration, automation and ongoing support for data foundations.

The question the piece leaves with plant teams: before the next go-live, can anyone in the room name the owner and source of truth for every figure on the screen?

via nukon.com (Original)

Filed under

  • data-governance
  • manufacturing-analytics
  • oee
  • dashboard-reliability
  • system-integration
Share this article:

More from Nathan Brooks

Nathan Brooks

Show full bio

Staff writer covering people, launches and day-to-day practice in Instruments, test and measurement at Testbench Report.

105 articles

Application notes

« Previous articleNext article »