A furnace plant can collect millions of measurements and still be short of decision-grade information. The quiet infrastructure between the PLC and the boardroom is the historian: it must preserve trustworthy measurements, connect them to production events, and make useful performance indicators available in time to act.
This article focuses on that measurement and data-infrastructure layer, not process models or automatic setpoint optimisation. A sound design gives plant teams operating thermal processing equipment a dependable basis for energy accounting, reliability reviews, and future analytics.
What a historian actually is, and why plants misconfigure theirs
A historian stores time-series data from control and instrumentation systems, then makes it available for operations and analysis. In ISA-95 terms, it generally sits at Level 3, bridging real-time control at Level 2 and business systems at Level 4. Its asset model should follow the production hierarchy: enterprise → site → area → work centre → work unit.[3]
A historian is not simply a large database or a trend screen. Its key engineering decisions include:
- Quality: each value should carry a quality status, such as an OPC UA StatusCode. A zero flow can mean a closed valve, or a disconnected transmitter; quality distinguishes the two.[5]
- Time: timestamps are the primary key. Unsynchronised clocks and unclear event boundaries undermine energy calculations.
- Interpolation: stepped values suit discrete states; interpolated values may suit analog trends. Neither method can restore samples lost through filtering or poor acquisition.
- Context: the asset model connects tags to equipment, operating modes, and production records. Treat this model as part of the data product, not an afterthought.
A familiar failure is to retain readings without preserving what was happening when they were captured. The result is a large archive that cannot reliably answer “energy per tonne for the last shift?”
The three-layer architecture that keeps analytics off the control layer
Keep the control system responsible for control. Let the historian and analytics layers consume data without becoming part of the real-time control path.
- Raw instrument layer: capture source readings at approximately 1 second to 1 minute via OPC UA or another approved interface. Keep this layer for evidence and troubleshooting; do not query it directly for routine dashboards.
- Calculated and aggregated layer: create energy totals by heat, shift, and hour; apply operating-mode filters; and standardise engineering units.
- Analytics and benchmark layer: publish rolling 30- and 90-day baselines, KPI deviation tags, and dashboard-ready summaries.
A common retention policy keeps raw 1-second data for about 2 years, 1-minute aggregates for 10 years, and hourly aggregates in cold storage indefinitely. Define retention against compliance, investigation, and storage requirements rather than copying a generic policy.[1][2]
Compression: where energy-accounting accuracy is won or lost
Swinging-door trending (SDT), with parameters such as CompDev and CompMax, has been used in PI/AVEVA environments since the late 1980s. It can greatly reduce data volume, but compression settings suitable for a visual trend may be unsuitable for integrating energy. A slow analog loop with CompDev at 0.1% of span may retain only 5–10% of raw samples. A 0.5% deadband on a 40 MW power tag can introduce 2–4% error into a heat-energy calculation.[1]
The following are published reference practices, not universal requirements. Validate settings against meter accuracy, sampling, historian behaviour, and an independent totaliser.
| Tag category | Reference deadband | Maximum interval | Why it matters |
|---|---|---|---|
| EAF power meter | ≤0.05% of span | 5 s | Protects integrated heat-energy totals |
| Reheating-furnace gas flow | ≤0.1% of span | 10 s | Captures slow consumption drift |
| Oxygen / nitrogen flow | ≤0.2% of span | 10 s | Preserves meaningful flow changes |
| Steam flow | ≤0.2% of span | 30 s | Supports utility and recovery accounting |
| Furnace zone temperature | ±1°C | 60 s | Suitable reference for process trending |
| Utility pressure | ±0.5% of span | 60 s | Generally supports service-level trending |
Set CompDev = 0 for tags used in compliance reporting where required by the reporting method. For other tags, validate compression before relying on stored values for accounting. Lossless delta-of-delta timestamp and XOR value compression, as described in the 2015 Gorilla paper, is a different technique: it reduces storage without approximating values, but its strongest compression assumes regular sampling.[4]
Tag naming and context: make furnace data machine-readable
Use a consistent hierarchical convention such as:
SITE.AREA.UNIT.ETYPE.MEAS.UOM
For example, a power tag can identify site, furnace, energy type, measurement, and unit without an informal lookup sheet. Maintain the structure through an asset model as well; platforms such as PI AF can link tags to equipment and production context.
Mode and state tags are essential. If an EAF specific-energy calculation includes idle and delay periods, reported GJ/t can be inflated by 15–30%. Link a heat-number string tag from MES to the historian so energy data can be joined directly to production events. For gas energy, link a calorific-value tag based on lab analysis, typically updated each shift, rather than applying a fixed assumed GCV. This is especially important for BFG, COG, and mixed gas.
KPI dashboards that drive same-shift action
A dashboard should show the calculation boundary, unit, time window, and data completeness, not just a number and a coloured dial.
For a reheating furnace, calculate GJ/t rolled against good tonnes, normalise for slab entry and exit temperatures, and exclude idle or soak-hold periods from the production denominator. Pre-aggregate hourly and shift energy tags so a six-month trend can return in under 3 seconds, rather than repeatedly scanning raw readings.[1]
A baseline-deviation alert can be configured to notify the shift manager during the same shift. Reference trigger values of about +5% for EAF and +8% for reheating furnaces are starting points for site-specific configuration, not universal alarm limits.
The following published ranges are illustrative industry references, not targets or guarantees for any specific plant. Actual performance depends on process boundaries, product mix, operating practice, and measurement quality.[1]
- Reheating furnace: 1.0–1.3 GJ/t best-in-class reference; 1.5–2.0 GJ/t fleet-average reference; 2.2–2.8 GJ/t laggard reference.
- Hot rolling mill electrical energy: 40–60 / 70–100 / 110–140 kWh/t across those same reference groupings.
- CCM electrical energy: 8–12 / 14–20 / 22–30 kWh/t.
Failure modes to design out
- Collector buffer overflow: size store-and-forward capacity for 24–72 hours of the site’s credible worst-case outage.
- Archive corruption: test recovery after unplanned shutdowns; verify backups and archive integrity.
- Cardinality explosion: a tag population can jump from 200,000 to 2 million overnight. Govern derived tags and require review before bulk creation.
- Replication lag: size the disaster-recovery link for peak bursts, not average traffic.
- Silent compression mismatch: test data quality and energy reconciliation so operators do not mistake a flat line for a stable process.
Apply the data model to each operation
- In a steel rolling mill, connect reheating-furnace energy with slab identity, entry temperature, discharge temperature, and rolled tonnes.
- For heat treatment furnaces, link cycle records, recipe, batch, and material identifiers to temperature traces for traceability.
- In an aluminum melting furnace, and a melting furnace for steel, align fuel or electrical energy with charge, melt, and tap events.
- A hot dip galvanizing plant benefits from linking zinc and gas metering to line state and production.
- A metal recycling furnace needs charge-level energy accounting tied to material mix and output.
- In the wire and cable industry, connect annealing-line temperature and speed with coil identity and production history.
Keep the data layer stocked
Data quality depends on maintainable plant infrastructure. Keep an engineered inventory of edge collectors and OPC UA gateways, store-and-forward capacity, calibrated flow meters, and server capacity sized for growth and peak loads. Include a clear cadence for GCV lab analysis, dashboard ownership, and tag governance. Review furnace spare parts and accessories as part of lifecycle planning, and discuss control-system integration through furnace automation and control. Equipment design and operating context also matter; see Continental Furnaces’ heat treatment solutions.
A five-phase implementation roadmap
- Tag audit: inventory meters, units, quality codes, timestamp sources, modes, and compression.
Three-layer build: separate raw acquisition from calculated totals and dashboard analytics across your industrial furnace systems. - Baseline and KPI definitions: agree boundaries, production denominators, normalisation rules, and validation checks.
- Alerts and dashboards: introduce same-shift deviation alerts with named owners and clear escalation paths.
- Governance and retention: control tag growth, review access, test recovery, and approve retention by data tier.
A dependable historian is the foundation for an enduring partnership between operations, maintenance, and energy teams. It makes data fit for action and gives later analytics a trustworthy measurement base. Consult Continental Furnaces, an industrial furnace manufacturer focused on data-ready equipment and instrumentation, to plan your approach, or request a quote. With 35+ years of expertise and ISO-certified quality, our team supports customized solutions and responsive service for sustained competitive advantage.
References
- iFactory AI, “Steel Plant Energy Data Historian Design for Analytics Guide,” October 2026.
- IoT Digital Twin PLM, “SCADA Historian Architecture: 2026 Design Patterns from First Principles,” June 2026.
- ANSI/ISA-95.00.01-2010 (R2018), Enterprise-Control System Integration.
- Pelkonen et al., “Gorilla: A Fast, Scalable, In-Memory Time Series Database,” VLDB, 2015.
- OPC Foundation, OPC Unified Architecture Specification, Parts 1–14.
