Energy Monitoring: Why Site kWh Is Useless

What is energy monitoring in manufacturing?

Energy monitoring in manufacturing is the continuous measurement of electrical, thermal and compressed-air energy consumption across a production system, correlated with what that system is actually doing at the time. The textbook definition stops at the first half of that sentence. From the architecture chair, the second half — the correlation with production context — is what decides whether the data is useful or just expensive telemetry. A site meter that records 42,000 kWh yesterday tells you nothing. The same 42,000 kWh broken down per machine, per order, per state of the machine, and compared against the units produced, tells you where 5–10 % of that consumption is structurally avoidable. Same electricity. Completely different conversation.

I spend most of my time building the data infrastructure that makes that second conversation possible — time-series ingestion at scale, per-machine attribution, correlation engines that tie kWh to production orders, alerting for peak-load excursions. After connecting more than 15,000 machines in 18 countries, the pattern for energy monitoring is the same everywhere: the plants that save energy are not the ones that buy better meters; they are the ones that get the metering hierarchy, the data model and the correlation right. The hardware has been cheap for a decade. The architecture is what is underinvested. This article is about that architecture, the design decisions that matter, and the specific metric — energy per produced unit — that separates real energy programmes from sustainability theatre.

The metering hierarchy — and why site-level data is useless

Every energy-monitoring system I have ever seen is built on a hierarchy of meters. The hierarchy has four levels, and the level at which a plant meters determines almost everything about what the plant can do with the resulting data. Most plants meter at levels one and two, wonder why their energy reports do not produce actionable insights, and conclude that energy monitoring does not work. The real problem is that they are measuring the wrong thing at the wrong granularity.

The economic argument for level-4 metering is usually misunderstood. People assume sub-metering is expensive. It used to be; it no longer is. A modern IoT energy meter with current transformers costs €200–500 per machine installed, goes in during planned downtime in 1–2 hours, and needs no PLC modification. For a plant with 50 machines that is a €15,000–25,000 one-off investment with a payback period typically under six months — because the first 2–3 % energy reduction from per-machine visibility alone usually funds the rest. The argument against sub-metering is not cost; it is habit. Plants have been running on site-level data for decades and have rationalised the blindness into a feature.

Why total kWh is the wrong metric

Here is the single most important idea in this whole article: total kWh is a volume metric, not an efficiency metric. A plant that reduces its total kWh by 8 % could have become more efficient, or it could simply have produced less. Without normalisation to production output, the number is unreadable. The only energy metric that is actually useful for operational management is kWh per produced unit (or kWh per kg, per metre, per run-hour, depending on product).

The kWh-per-unit metric is where energy monitoring collides with MES. You cannot compute it without knowing exactly how many units were produced during the interval the kWh was consumed, ideally at the cycle level. That correlation is trivial if the energy meter and the production-counting system live in the same data model; it is a multi-month integration project if they do not. This is the reason energy-only systems sold by utility companies rarely move the needle operationally — they have excellent kWh data and no production context. A plant can watch its total consumption rise and fall beautifully and still have no idea whether it is getting more or less efficient per part.

Once you have kWh per unit, two things happen immediately. First, you can compare the same product across shifts, lines and plants on a genuinely like-for-like basis — because you are comparing efficiency, not volume. Second, you can see structural energy waste that was previously invisible: a machine that produces the same part at 0.42 kWh/unit on the day shift and 0.58 kWh/unit on the night shift is not a random measurement artefact, it is a 38 % efficiency gap that has a root cause somebody can find and fix.

The four energy-loss categories

Productivity loss has its well-known categories; energy loss has a parallel structure that is less widely understood but equally useful for framing improvement work. These four categories cover almost every opportunity a real manufacturing plant has, and they show up in roughly the proportions listed below across the automotive, metal-processing and food customers we work with.

The most actionable category by far is idle consumption. A surprising fraction of industrial energy is burned by machines that are technically "off" — they are in standby, keeping heaters warm, running auxiliaries, holding hydraulics pressurised for fast restart. On a Sunday, a plant should ideally draw 10–20 % of its weekday load; in reality many plants draw 40–60 %, because half the machines were never actually shut down. Detecting this is trivial once you have per-machine metering and machine-state data from the MES side-by-side. Fixing it is usually a procedural change, not a capital project.

Architectural observation from the platform telemetry across 15,000+ connected machines: across the SYMESTIC installed base in automotive, food and metal processing, the typical energy reduction achievable in the first 12 months of per-machine monitoring is in the 5–10 % range of total plant consumption — not from a single intervention, but from four or five small, data-driven actions stacking on top of each other. The biggest single lever is almost always idle-load elimination on weekends and between-order periods. The second biggest is peak-load shaving through sequenced start-ups. Both of these require sub-minute resolution — the 15-minute utility interval data that most plants rely on is averaging over exactly the transients that matter. Site-level kWh at 15-minute resolution will never show you these patterns. Per-machine kWh at 1-minute resolution, correlated with machine state and production orders, makes them unmissable. The difference is not the meter; the difference is the data model.

Architecture: how energy monitoring scales across plants

Energy monitoring looks simple on a single machine — measure volts and amps, multiply, integrate over time, store. The problems start when you go from one machine to five thousand, from one plant to eighteen countries, from daily reports to real-time correlation with production context. The architecture decisions that matter at scale are almost all about the data layer, not the measurement layer.

The three design decisions I would flag as non-negotiable: 1-minute resolution as the default sampling rate (anything coarser loses peak transients and idle-state boundaries), co-location with production data in one data model from day one (not a side-car database that has to be joined later), and per-machine granularity from the first rollout (upgrading a level-2 deployment to level-4 afterwards is twice the work of doing it right the first time). Every time we have seen an energy project stall out at 1–2 % savings instead of 5–10 %, at least one of those three decisions was wrong.

What this looks like in practice — the typical deployment pattern

Across the SYMESTIC installed base, energy monitoring is almost never the first module a customer deploys. The typical sequence is production metrics and OEE first (months 0–6), then event and alarm analysis (months 3–9), and then energy monitoring added as an extension on top of the existing data foundation (months 6–12). This sequence is not accidental. It works because the MES foundation already contains the machine-state and order-context data that energy monitoring needs for its correlation — so adding energy is a sensor retrofit and a module activation, not a new data project.

The deployment pattern for a mid-size plant is roughly consistent across industries. Installation of per-machine IoT energy meters takes 1–2 hours per machine during planned stops, so a 50-machine plant is fully instrumented in 2–3 weeks of staggered downtime windows — no production interruption. The first usable kWh-per-unit dashboards are running within 4–6 weeks, typically covering the plant's highest-consumption machines first (melt units, ovens, presses, compressors). The first round of actions — weekend shutdown protocols, start-up sequencing to shave peak loads, compressed-air leak audits — is usually implemented in months 2–4 of the energy rollout. By month 12 the compounded effect across those actions is in the 5–10 % total-consumption range, which is the SYMESTIC-documented benchmark for energy-monitoring-enabled reductions.

The arithmetic of those rows explains why the overall benchmark lands at 5–10 % rather than in a single headline number. No single action produces a double-digit reduction in a mature industrial plant. What produces the 5–10 % total is the stacking of four or five coordinated small actions, each of which requires per-machine visibility correlated with production context. That is the fundamental architectural argument for energy monitoring done properly, and it is why the module lives inside the MES rather than alongside it.

FAQ

What is energy monitoring in manufacturing?
Energy monitoring is the continuous measurement of electrical, thermal and compressed-air consumption across a production system, correlated in real time with the machine states and production orders that produced the consumption. The correlation is what distinguishes operational energy monitoring from utility billing — the goal is not to know what the plant used last month, but to understand what each machine used while it was doing what, so that avoidable consumption can be identified and eliminated.

Why is site-level energy data not enough?
Because site-level kWh is a volume number, not an efficiency number. It tells you what the plant consumed without telling you whether that consumption was justified by the output. A plant that reduces its site kWh by 8 % could be more efficient, could simply have produced less, or could be hiding inefficiency behind product-mix changes. Without normalisation to production output — ideally per machine — the number cannot drive operational decisions. Site metering is appropriate for the utility bill and nothing else.

What is kWh per unit and why does it matter?
kWh per produced unit — sometimes called specific energy consumption or SEC — is the energy divided by the output. It is the only energy metric that allows genuine like-for-like comparison across shifts, lines, plants and products. A machine that produces the same part at 0.42 kWh/unit on day shift and 0.58 kWh/unit on night shift has a 38 % efficiency gap that is invisible in total-kWh reporting but obvious in per-unit reporting. Computing kWh/unit requires energy and production data in the same data model at the same time resolution, which is why it lives naturally in an MES with an energy module rather than in a standalone energy system.

How much energy can a plant save with per-machine monitoring?
Across the SYMESTIC installed base the typical reduction in total plant consumption in the first 12 months is 5–10 %. The figure is not produced by any single intervention; it stacks from idle-load elimination (2–5 %), peak-load shaving (10–20 % of demand charges, not kWh), cross-shift best-practice propagation (1–3 %) and compressed-air leak detection (1–3 %). Plants that have already done surface-level energy work see the lower end; plants starting from site-level metering see the higher end. The savings compound in subsequent years as more use cases are added.

What sampling resolution do you need for useful energy monitoring?
One minute is the minimum useful resolution for operational decisions, one second is the right resolution for peak-load analysis, and one hour is the standard that most plants default to and that systematically hides the transients that matter. The 15-minute interval used by utilities averages over exactly the peaks and idle-state boundaries that contain the actionable information. If an energy-monitoring vendor proposes hourly or 15-minute data as the default, they are optimising for storage cost at the expense of insight.

How expensive is it to retrofit per-machine energy metering?
€200–500 per machine for a modern IoT energy meter with current transformers, installed in 1–2 hours during planned downtime with no PLC modification required. For a 50-machine plant that is €15,000–25,000 of hardware and a 2–3 week staggered installation. Payback is typically under six months — the first 2–3 % of savings from idle-load elimination alone usually funds the entire rollout. The argument against sub-metering on cost grounds reflects 2010 hardware prices, not 2026 prices.

How does energy monitoring connect to ISO 50001?
ISO 50001 is the international standard for energy management systems, and it requires structured energy-performance indicators (EnPIs), regular review cycles and demonstrable improvement. A properly architected energy-monitoring system delivers the EnPI data natively — kWh per unit, per product family, per shift, per plant — and the audit trail needed for ISO 50001 recertification. The harder ISO 50001 requirements around review governance and continuous improvement are organisational, but the measurement foundation that makes them possible is exactly what an MES-integrated energy module provides.

Why should energy monitoring be part of the MES and not a separate system?
Because the useful metric — kWh per unit, per order, per product — requires energy data and production data in the same time-aligned data model. Separate systems can be integrated after the fact, but the integration cost and latency are an order of magnitude higher than having the two modules share a common data layer from day one. The MES already knows which order was running on which machine at which second; plugging energy data into that context is a sensor retrofit and a module activation. Starting from a standalone energy system and trying to add production context is a multi-month integration project.

How does SYMESTIC implement energy monitoring?
Per-machine IoT energy meters via Modbus/MQTT into the same cloud data model that holds the production and event data, with 1-minute resolution as default and 1-second available for peak-load analysis. kWh automatically correlated to machine state and active production order, so kWh-per-unit is a computed field rather than an export-and-join exercise. Dashboards for idle-load, peak-load and kWh-per-unit comparisons at plant and group level. ISO 50001-ready EnPI reporting layer. Typical per-plant reduction after 12 months sits in the 5–10 % range across the SYMESTIC installed base. See SYMESTIC Process Data.

Related: Production Parameters · Performance Measurement · Machine Data Acquisition · OEE · ISO 50001 · OPC UA · MES · SYMESTIC Process Data

About the author

Mark Kobbert

CTO of SYMESTIC GmbH. Responsible for the Cloud MES architecture since 2014 — from the Microsoft Azure microservice platform and IoT gateway connectivity to real-time data processing for 15,000+ machines across 18 countries. B.Sc. Business Informatics, SRH Heidelberg. Led the platform's cloud-native rebuild: instead of lifting on-premise software to the cloud, built the architecture from scratch around scalability, real-time capability and maintainability. · LinkedIn