Where to Start with Factory Energy Management: From Sub-Metering to Benchmark Analysis

Five stumbles that recur in the early stage

• Metering the whole plant at once: capital spend balloons first, and more data arrives with no agreement on how to read it, so it sits unused
• Meters installed but no baseline defined: with nothing to compare against, you cannot say whether consumption went up or down
• A dashboard nobody opens: the screens are not broken down along dimensions the floor can act on, such as line, machine and shift
• Targets set by feel: intensity targets are fixed without separating the effect of output volume and outdoor temperature, so neither hitting nor missing them can be explained
• No named owner: a number moves, but who checks it and who proposes a countermeasure is unclear, and all that remains is the monthly report

All five share the same root: before the measurement discussion began, nobody had settled what to compare against or who acts. If numbers are collected only to fill in an ISO 50001 report, improvement rarely follows.

A working order: metering design, baseline, benchmark analysis, alarms, improvement cycle

Following the order alone removes a lot of wasted spend. Each step produces the input for the next, and skipping any one of them leaves the later conclusions without support.

• Metering point design: decide how deep to split electricity, compressed air, water, gas or steam and the chiller plant, working backwards from what you actually intend to improve
• Baseline measurement: build a baseline from several months of actual readings, tied to output volume, shift pattern and outdoor temperature; skip this and later evaluation has nothing to stand on
• Benchmark analysis: line up machines within the same process, days and shifts on the same line, and sites across the group, then start where the gaps are widest
• Anomaly alarms: turn idle-period consumption, overnight compressed-air flow that indicates leaks, and demand trending against contracted capacity into calculable alarm definitions
• Improvement cycle: compare before and after against the same baseline, and keep tracking in the following months to see whether the effect holds

How to lay out metering points

More metering points does not automatically mean more useful. Split down to the level where you can say what happens if this number moves and who does it, and stop there; anything finer becomes operating load. Three tiers keep it manageable: the whole plant, then by end use (production equipment, HVAC, lighting, utilities), then per machine for major equipment.

It is also worth checking early whether existing instruments can be reused. Meters with pulse output, Modbus or BACnet can go straight into data collection. Older meters with no communication port can sometimes still be read through an add-on converter or a pulse tap, which avoids replacing the entire fleet. In plants where compressed air and the chiller plant account for a large share, this check tends to pay off quickly.

From monitoring to operational improvement, then to load adjustment

The early stage is about visibility. Once sub-metering and an energy dashboard are running, the items that surface first are usually the ones that need no equipment replacement: idle consumption, HVAC running longer than the process requires, and compressor sequencing that does not match actual demand. Most of the plant and building energy SCADA work we take on also starts at this layer.

The next stage is embedding it into daily operation. Monitoring the utilities (water, electricity, gas, HVAC, pure water and effluent) as one FMCS puts energy data and equipment status on the same screen, which shortens the time needed to isolate an anomaly. Further out sit combined solar and storage management, bringing generation, storage and CO₂ accounting into one view, and a traceable data foundation for ISO 50001 reporting and carbon accounting. For sites overseas we have delivered interface language switching (Chinese, English, Japanese) on a redundant OPC UA architecture so headquarters and the plant read the same figures; the language set is fixed per project.

AI-generated load adjustment suggestions and large-language-model operating assistance sit in our research and joint-validation track and are not offered as a standard delivery commitment. Where they are included in a project, the premise is that a responsible person reviews the suggestion before it is executed.