Real-Time OEE Dashboard with Grafana for a food packaging plant

The starting point: data locked inside PLCs, zero visibility in production

The client operates a food packaging plant with three production lines, each controlled by a Siemens S7-1500 PLC programmed in TIA Portal. The PLCs accurately logged piece counters, machine states, line speeds and active alarms — but that information remained locked inside TIA Portal and the local HMI panels.

The production management team made decisions based on manual shift reports: spreadsheets filled in by supervisors at the end of each shift and sent by email. OEE was calculated once a week, with a four-day delay. By the time a performance issue was identified, the root cause had already changed.

The objective was clear: real-time production visibility, without replacing the existing PLCs and without purchasing a full MES system. The client wanted to move fast, at a reasonable cost, with a tool their engineering team could maintain independently.

Why Grafana as the visualisation platform

Grafana is the de facto standard for time-series dashboards in both industrial and IT environments. Its advantages over other options were clear for this use case:

• Open source with enterprise support: no visualisation licence cost. The client can add users, panels and data sources at no additional cost.
• Native InfluxDB integration: the most widely used time-series database in IIoT environments. Grafana reads from InfluxDB using its Flux query language natively, without any adapters.
• Configurable alerts: Grafana allows alert rules on any metric — line speed below target, accumulated downtime exceeding a threshold — with notifications delivered via email, Telegram or Microsoft Teams.
• Web access without a client: any user with credentials accesses the dashboard from any browser, with no software to install. Production managers can check plant status from their phone.
• Viable internal maintenance: Grafana has a reasonable learning curve. The client's engineering team can create and modify panels without depending on Bluemation for every change.

Solution architecture

The architecture was designed in four clearly defined layers:

Layer 1 — Data acquisition: OPC-UA from the PLCs

Siemens S7-1500 PLCs have a built-in OPC-UA server that is activated directly from TIA Portal without modifying the control programme. We configured the OPC-UA servers on all three PLCs, exposing the relevant variables: good and rejected product counters, machine states (running, stopped, faulted), current line speed, sealing temperature and active alarms.

For data acquisition and brokering we used N3uron, an IIoT edge platform designed for industrial environments. N3uron acts as an OPC-UA client for all three PLCs, normalises the data (tag names, units, scaling) and publishes it to InfluxDB via MQTT. Sampling frequency was set to 1 second for critical process variables (speed, state) and 10 seconds for trend variables (temperature, cumulative counters).

Layer 2 — Storage: InfluxDB

InfluxDB is a database designed specifically for industrial time series: it can ingest millions of data points per second, automatically compress historical data, and respond to time-range queries with millisecond latency. It was installed on a Linux server on-premise on the client's OT network.

Three buckets were defined with different retention policies: real-time data with 7-day retention for high-resolution queries, data downsampled to 1 minute with 1-year retention for historical analysis, and data downsampled to 1 hour with indefinite retention for long-term trends. Downsampling tasks are automated in InfluxDB using Flux, without manual intervention.

Layer 3 — Visualisation: Grafana

Grafana connects to InfluxDB as a datasource and serves the dashboards via a web server accessible from the plant's internal network. Four main dashboards were built:

• Real-time production dashboard: main panel for the control room showing the current state of all three lines, shift counters, current vs target speed, and active alarms highlighted in red.
• OEE dashboard: automatic OEE calculation broken down into availability, performance and quality. Comparison by shift, by line and weekly trend. This dashboard completely replaced the manual shift reports.
• Downtime dashboard: automatic log of every stoppage with duration, machine state at the moment of the stop, and a histogram of the most frequent causes (Pareto analysis of breakdowns).
• Process trend dashboard: historical evolution of sealing temperature, packaging pressure and line speed. Allows process drifts to be identified before they affect product quality.

Layer 4 — Alerts and notifications

Automated alerts were configured in Grafana for the most critical events: shift OEE below 70% (notification to the shift manager), line stoppage exceeding 10 minutes (notification to the maintenance supervisor), sealing temperature out of range (immediate alarm to the line operator). Notifications were delivered via email and through an existing Microsoft Teams channel.

Implementation: from zero to production in 7 weeks

• Week 1 — Analysis and tag mapping: full inventory of variables available in all three PLCs, prioritisation with the production team, and data model design in InfluxDB.
• Weeks 2–3 — Infrastructure and integration: OPC-UA server configuration in TIA Portal, N3uron and InfluxDB installation, validation of communication and the full data flow to the database.
• Weeks 4–5 — Dashboard development: construction of the four Grafana dashboards with real production data. Daily iterations with the production team to adjust metrics, colours and panel layout.
• Week 6 — Alerts and testing: alert system configuration, load testing, and historical retention validation. OEE calculation verified against manual shift reports from the previous period.
• Week 7 — Training and go-live: training for operators, shift supervisors and the maintenance manager. Official go-live with supervised operation for 48 hours.

Results after three months

• +18% OEE improvement on the line with the highest history of micro-stoppages. Real-time visibility revealed that one specific packaging machine accounted for 40% of its line's total downtime — a figure that was impossible to detect with weekly manual reports.
• Complete elimination of manual shift reports. The time supervisors spent filling in spreadsheets (estimated at 20–30 minutes per shift) was reassigned to improvement tasks.
• ~35% reduction in average fault diagnosis time, thanks to the detailed downtime log with exact timestamps and the state of every variable at the moment of the fault.
• Early detection of a sealing temperature drift on line 2 that would have produced out-of-spec product. The Grafana alert arrived 8 minutes before the operator would have detected it visually.
• Full team adoption in under two weeks. Production managers check the OEE dashboard on their phones before entering the plant every morning.

Cost vs alternatives

Key lessons learned

• The data model is the most critical step. Defining tag names, units and sampling frequency correctly from the start avoids costly rework. A poorly named tag in InfluxDB is hard to correct once weeks of history have accumulated.
• Involve the production team from day one. The most useful dashboards are not designed by the integrator — they are designed by the shift supervisor who knows what they need to see. Our role is to translate that need into Grafana.
• Downsampling must be defined from the start. At 1-second frequency, three production lines generate millions of data points per day. Without a retention and downsampling policy from day one, storage grows out of control within weeks.
• Alerts are more valuable than dashboards. A dashboard nobody actively monitors is of little use. An alert that arrives on the supervisor's phone when something goes wrong generates immediate value, even if the user never opens Grafana.

If your plant has PLCs with data nobody is using and you want real-time production visibility without purchasing a full MES, tell us about your case. We provide a no-obligation assessment and can show you what a pilot dashboard would look like with your own data.