OEE Dashboards from IoT Data: What to Compute, What to Ignore

Overall Equipment Effectiveness is the most-used and most-mis-implemented metric in manufacturing. Most OEE dashboards either compute it badly or compute it correctly and present it uselessly. Both lead to the same outcome: operators stop trusting the number.

Done well, OEE drives genuine improvement conversations. Done badly, it’s a vanity number on a wall TV.

The formula

OEE = Availability × Performance × Quality

Each factor is between 0 and 1; world-class is 0.85, typical manufacturing is 0.6.

• Availability = Run Time / Planned Production Time
• Performance = (Ideal Cycle Time × Total Count) / Run Time
• Quality = Good Count / Total Count

The formula is simple. The data behind each factor is where the work is.

Where the data comes from

Availability — the easy one

Run Time = the equipment was actively producing. Planned Production Time = the equipment was scheduled to produce.

Sources:

• Equipment-state telemetry from the PLC (Running, Idle, Faulted, Setup)
• Production schedule from MES or scheduling system
• Logged downtime reasons (manual or auto-classified)

The trap: not every “stopped” state is unplanned downtime. Setup, scheduled maintenance, and breaks are different from breakdowns. If your dashboard treats them the same, the conversation about how to improve goes nowhere.

Performance — the medium-hard one

Ideal Cycle Time = the engineering or design-target time per unit. Total Count = number of units produced.

Sources:

• Cycle counter from the PLC (units made)
• Engineering specs for ideal cycle time

The trap: ideal cycle time is rarely fixed across product mix. A line running product A at 1 unit/sec and product B at 0.7 units/sec needs a weighted target, not a single constant. Use a per-product cycle time and weight by mix.

Quality — the hardest one

Good Count = units that met spec. Total Count = all units produced.

Sources:

• Quality station data — manual reject codes, automatic vision inspection, inline metrology
• First-pass yield (FPY) from end-of-line testing

The trap: quality data lags. A unit produced at 10:00 AM may not be inspected until 11:30 AM. The OEE for the 10:00 AM hour can’t be computed until quality data lands. Either present OEE with a known delay or compute it provisionally and update.

Granularity choices

OEE can be computed at multiple time grains:

• Per shift (8 hours) — coarse enough to be stable, fine enough for daily improvement conversations
• Per hour — finer; useful for diagnosing what happened during a shift but noisier
• Per minute — almost always too noisy to be useful; reserve for engineering investigation

The right answer for most operations: shift-level OEE as the primary number, hour-level as a drill-down for diagnostic conversations, per-event loss tracking (downtime reasons, slow cycles, quality rejects) underneath.

A dashboard showing a single per-minute OEE number is usually a mistake. It moves up and down constantly and tells operators nothing actionable.

What to put on the dashboard

Three layers, ordered by audience:

Operator floor display

• Current shift OEE with target line
• Big A / P / Q numbers for the current shift
• Active losses ranked by impact
• Time-to-shift-end

The operator dashboard is the only place a real-time-ish number works. Operators react to what’s happening now. Even here, “real-time” is best at minute-level updates, not sub-second.

Production manager dashboard

• Trend of OEE by shift across the last 30 days
• Pareto chart of losses (sorted by hours lost)
• Comparison across lines or cells
• Drill-down into specific shifts that under-performed

This dashboard drives the daily and weekly improvement conversations.

Plant manager / executive view

• OEE trend by month
• Comparison across plants
• Improvement-project tracking with before/after OEE

This dashboard is for strategic decisions, not operational ones. Update weekly or monthly.

The Pareto chart that drives improvement

The single most useful artefact: a Pareto chart of losses for the most recent week.

X-axis: loss reasons (downtime causes, performance losses, quality losses). Y-axis: hours lost (or units lost).

Sorted descending. Top 3-5 reasons get the focus. Everything else is noise until the top reasons are addressed.

This is what continuous-improvement programs (TPM, Lean, Six Sigma) actually use. The OEE number on the wall is the headline; the Pareto is the work.

Common dashboard sins

Three things that kill operator trust:

1. OEE > 1.0 — happens when ideal cycle time is unrealistic, or when a counter is wrong. Always display ideal cycle time alongside; expose the assumption.
2. OEE that doesn’t match what the operator saw — the operator was running well; the dashboard says 60%. Usually a definition mismatch (planned vs actual schedule, what counts as downtime). Reconcile every dispute; document the resolution.
3. OEE without the loss breakdown — a number with no story. The number alone changes nothing; the breakdown drives action.

What we typically build

For an OEE dashboard engagement:

• Data pipeline: equipment-state and production-count telemetry → time-series store → aggregation service → dashboard
• OEE calculation service — clear, documented logic. No spaghetti SQL.
• Loss classification — automatic where possible, manual codes for the long tail
• Multi-tier dashboards — operator floor, manager, executive
• Pareto chart engine — sortable, filterable, drillable
• Alignment workshop with operations — sit with the team, define what counts as availability/performance/quality, document the rules

The dashboards are the visible deliverable; the definitions and the operational alignment are the durable one.

What we hand over

For an OEE engagement we ship:

• The OEE definition document signed off by operations and engineering
• Per-line Pareto charts of losses for the previous 30 days
• A monthly review template that operations runs without us
• The pipeline as code (Terraform / Helm) so it’s maintainable
• An expansion plan for adding the next line / cell / plant

If you are building OEE dashboards or running ones operators don’t trust, we have shipped this combination across multiple plants.