Factory Energy Management: From IIoT Metering Hardware to kWh per Part

Energy is one of the few major manufacturing costs still managed at the level of a monthly invoice. The bill arrives, it is larger than last month, somebody asks why, and nobody can answer with precision because the meter that produced it measures the entire site as a single number. Every improvement discussion that follows is therefore a discussion about theory. Factory energy management exists to replace that theory with measurement.

The hardware layer is where this starts and it is less daunting than most manufacturers expect. Sub-metering at the level of individual machines, cells or distribution boards can usually be added without disturbing production — current transformers clamp around existing conductors, and IIoT energy meters report over Modbus, MQTT or OPC UA into the same network your other shop-floor data already uses. You are not rewiring the plant; you are instrumenting it.

What those meters should record is more than kilowatt-hours. Active and reactive energy, power factor, per-phase current and voltage, and peak demand each answer different questions. Power factor and reactive energy often reveal charges you are paying without knowing it. Peak demand explains tariff penalties that a total-consumption view cannot see. Per-phase data catches imbalance that shortens equipment life long before it causes a failure.

The data becomes actionable only when it is tied to production context, and this is where most energy projects either succeed or quietly stall. A chart of consumption over time tells you when energy was used. A chart of consumption joined to MES data tells you what was being produced, on which machine, under which order, at what output rate. That join transforms the metric from kilowatt-hours per day into kilowatt-hours per part — and kWh per part is the only energy number that survives a change in production volume.

Once that metric exists, comparisons become possible that were previously arguments. The same product made on two machines can be compared directly. The same machine across two shifts can be compared. Consumption before and after a maintenance intervention can be compared. Factories almost always find at least one machine that consumes materially more energy per part than its peers, and the cause is usually mechanical and fixable rather than inherent.

Idle and non-productive consumption is the most reliable early win. Equipment drawing significant power while producing nothing — over breaks, between shifts, across weekends — is common, invisible without sub-metering, and often addressable with sequencing changes rather than capital investment. Compressed air systems are the standard example: leaks that cost nothing visible and a great deal in aggregate, detectable as baseline consumption that never falls to zero.

Anomaly detection adds a maintenance dimension that pays for the instrumentation separately. A motor whose current draw creeps upward over weeks is usually telling you something about bearing wear or misalignment before any vibration threshold is crossed. Energy data is a cheap, continuous condition signal — you installed the meters for cost reasons and get early failure warning as a by-product.

For manufacturers with reporting obligations — ISO 50001, CSRD, customer sustainability requirements, carbon accounting — the same data serves compliance without a parallel measurement exercise. Per-product energy figures underpin product carbon footprint calculations that are otherwise estimated from industry averages. Increasingly, industrial customers ask for those numbers directly, and estimated answers are becoming harder to defend.

The pragmatic sequence: meter the largest consumers first rather than everything at once, join the data to production context immediately rather than later, and pick one metric — kWh per part on your highest-volume product line — to publish and improve. Sites that instrument comprehensively before deciding what question they are answering tend to end up with a great deal of data and no decisions.