Direct Answer
An old workshop wants submetering, but installing a meter on every machine is impossible — is there another way? NILM continuously samples the current waveform with one high-precision CT at the main feeder, and uses a combination of three features — "startup transient + steady-state power + harmonic signature" — to decompose the total current into per-device load curves; the Tiangyan Engine E-01 and Wanxiang Engine V5.0 load fingerprints follow exactly this path, giving the old workshop sub-metering data without rewiring or stopping production. One meter in exchange for the whole workshop's submetering data — cost reduced by an order of magnitude.
A scene: an old workshop wanted to know each machine's consumption — a meter on every machine: dozens × several hundred yuan + outage wiring — NILM: one CT at the main feeder — decomposed into per-device curves — no rewiring, no production stop — submetering data in hand.
1. Why an Old Workshop Cannot Do Submetering
The status of an old workshop — no submeters: only the master meter — per-device consumption unknown; The cost of a meter on every machine — meters + installation: dozens × several hundred yuan + outage — costly; The trouble of a meter on every machine — production-stop wiring: old-cable retrofit risk — production affected; The value of submetering — per-device consumption visible: which machine consumes a lot and which idles — energy saving with evidence; The contradiction of an old workshop — wanting submetering but not meters: the cost conflicts with the value — a new approach is needed.
In one sentence: an old workshop cannot do submetering = the cost of a meter on every machine is high and production-stop wiring is troublesome — wanting submetering but not meters — the NILM approach is needed.
A cost example — a workshop with 40 devices metered — meters and installation over one hundred thousand yuan plus 2 days of production stop — the cost frightening. A meter on every machine is a heavy investment — meters, installation and production stop — an old workshop needs a lightweight approach.
2. What NILM Is
The meaning of NILM — non-intrusive load monitoring: monitoring at the main feeder — decomposing per-device loads — no submeters; The principle of NILM — decomposing the total current: the total waveform is the superposition of all device waveforms — split by feature; The implementation of NILM — one high-precision CT: continuous sampling at the main feeder — the current waveform; The output of NILM — per-device load curves: when each device runs and how much it consumes — submetering data; The essence of NILM — one meter for the whole workshop: no rewiring and no production stop — lightweight submetering.
In one sentence: NILM = one CT at the main feeder + current-waveform decomposition — split into per-device load curves by feature — one meter for the whole workshop's submetering data.
A NILM example — a workshop installed one CT at the main feeder — three days later the per-device curves of 40 devices were obtained. NILM lands lightly — one CT, no rewiring and no production stop — submetering data automatically in hand.
3. The Three-Feature Set
Feature one: startup transient — the waveform at the device startup moment: inrush current and startup duration — the device fingerprint; Feature two: steady-state power — the power when running steadily: active and reactive power — the device identity; Feature three: harmonic signature — the waveform-distortion pattern: harmonic orders and magnitudes — the device tag; Combining the three features — one alone is not enough: transient + steady state + harmonics — cross confirmation; The value of the three features — the more complete, the more accurate: joint identification — high decomposition accuracy.
In one sentence: the three features = startup transient + steady-state power + harmonic signature — joint identification — the more complete the features, the more accurate the decomposition.
A three-feature example — a motor with a large startup inrush, 15 kW steady-state power and a clear 5th harmonic — locked by the three features. The three-feature combination is the NILM identification logic — one feature is easily confused and three features cross-confirm — decomposition becomes accurate.
4. Which Devices Are Easy to Decompose
The easy type — impact loads: motors and compressors — obvious startup features; The reason it is easy — unique features: large startup inrush and sudden power changes — easy to identify; Periodic loads — air compressors and pumps: regular start-stop cycles — clear decomposition; Large-power devices — strong signals: a large power share — prominent in the total current; The conclusion on easy devices — the more unique the feature, the easier: impact loads are the sweet spot of NILM.
In one sentence: easy to decompose = impact loads (motors/compressors) + periodic loads (air compressors/pumps) — unique features, easy to identify.
An easy example — a workshop with 3 air compressors — regular start-stop — clearly decomposed by NILM. Devices with unique features decompose best — large impact and regular cycles — NILM is naturally suited to such loads.
5. Which Devices Are Hard to Decompose
The hard type — resistive heating: electric furnaces and ovens — steady power; The reason it is hard — plain features: no startup inrush and few harmonics — similar and hard to distinguish; Same-type devices running together — several identical devices on at once: similar waveforms — hard to split; Small-power devices — weak signals: a small share — drowned in the total current; The conclusion on hard devices — the plainer the feature, the harder: resistive heating is the difficulty of NILM.
In one sentence: hard to decompose = resistive heating (electric furnaces/ovens) + same-type superposition + small-power devices — plain features and weak signals.
A hard example — a workshop with 6 electric furnaces — steady and similar power — NILM could only estimate the total. Plain features are the hardest — no inrush and few harmonics — resistive loads are the boundary of NILM.
The table — the decomposability of common devices:
| Device Type | Feature | Difficulty |
|---|---|---|
| Motor/Compressor | obvious startup inrush | easy |
| Air compressor/Pump | regular start-stop cycles | easy |
| Electric furnace/Oven | steady power | hard |
| Small electronic device | weak signal | hard |
6. The Trade-off with "A Meter on Every Machine"
The accuracy of a meter on every machine — direct measurement: one meter per device — the highest accuracy; The cost of a meter on every machine — meters + installation + production stop: a heavy investment — a one-time heavy cost; The accuracy of NILM — decomposition estimation: split by feature — accurate enough but not absolute; The cost of NILM — one CT: no rewiring and no production stop — a small investment; The conclusion of the trade-off — the accuracy need decides the approach: choose meters for absolute accuracy and NILM for economical submetering.
The table — a meter on every machine vs. NILM:
| Approach | Accuracy | Cost | Installation |
|---|---|---|---|
| Meter on every machine | highest | high | production-stop wiring |
| NILM | good enough | low | no rewiring |
In one sentence: the trade-off = choose meters for absolute accuracy and NILM for economical submetering — the accuracy need decides the approach — NILM is the cost-effective choice.
A trade-off example — a workshop first surveyed with NILM — then added precise meters to key devices — the two layers combined. First NILM survey, then precise metering for the key points — accuracy and cost both covered — the two-layer approach is the most pragmatic.
A decision suggestion — with a limited budget, start with NILM: decompose all devices first — then precisely meter the key ones — spend on the blade. NILM is the entry approach — the whole workshop surveyed at low cost — precision added later for the key points — the phased investment is the most reasonable.
7. The Accuracy Boundary of NILM
The error source of decomposition — similar features: same-type devices are hard to distinguish — errors; The boundary of accuracy — small power hard to tell: a small share and drowned — hard to distinguish; Improving accuracy — training the feature library: the more complete the device fingerprints — the more accurate the decomposition; Verifying accuracy — sampled comparison: NILM results compared with meters — verification and calibration; Knowing the boundary — good enough: for energy-saving diagnosis and load forecasting — NILM accuracy suffices.
In one sentence: the NILM accuracy boundary = same-type hard to distinguish and small power hard to tell — the feature-library training improves it — the accuracy suffices for energy-saving diagnosis.
A boundary example — a workshop compared NILM with meters — a 5% error for large devices and 20% for small ones. NILM has a boundary — accurate for large devices and weaker for small ones — choosing the right scenario, the accuracy is good enough.
8. Two Landing Scenarios
Scenario one: energy-saving diagnosis — sub-metering consumption visible: which device idles and which is inefficient — energy-saving retrofits with evidence; Scenario two: load forecasting — per-device patterns: start-stop cycles and energy-use modes — more accurate forecasting; The value of energy-saving diagnosis — finding waste: idling devices and standby losses — saving electricity directly; The value of load forecasting — peak-shifting planning: per-device forecasting aggregated — a more accurate total load; The essence of the two scenarios — monetizing submetering data: diagnosis finds waste and forecasting plans — NILM lands.
In one sentence: the two landing scenarios = energy-saving diagnosis (finding waste) + load forecasting (planning) — monetizing submetering data.
A scene example — a workshop's NILM found 3 air compressors idling on the night shift — after adjustment, the monthly electricity bill fell 12%. Energy-saving diagnosis is NILM's sweet spot — submetering data finds waste — the adjustment shows immediate results.
A forecast example — a workshop used NILM to obtain per-device start-stop patterns — aggregated to forecast the next day's load — peak-shifting planning more accurate. Load forecasting depends on per-device patterns — start-stop cycles and energy-use modes — NILM makes the total-load forecast more accurate.
9. Back to FEXLINK: How to Use NILM
FEXLINK's Tiangyan Engine E-01 and Wanxiang Engine V5.0 load fingerprints: NILM continuously samples the current waveform with one high-precision CT at the main feeder — the "startup transient + steady-state power + harmonic signature" three-feature combination — decomposing the total current into per-device load curves — the old workshop obtains submetering data without rewiring or stopping production. The NILM checklist: ① the principle — one CT at the main feeder: the total waveform split into per-device curves by feature → ② the three features — startup transient + steady-state power + harmonic signature: joint identification → ③ the sweet spot — impact loads easy to decompose: motors, compressors and air compressors → ④ the boundary — resistive heating hard: plain features and same-type superposition → ⑤ the trade-off — meters for accuracy and NILM for economy: survey first and precisely meter later → ⑥ the landing — energy-saving diagnosis finds waste and load forecasting plans: monetizing submetering data — NILM, one meter for the whole workshop's submetering data.
One summary: NILM = one CT at the main feeder + three-feature decomposition — startup transient + steady-state power + harmonic signature — splitting the total current into per-device load curves. A meter on every machine is costly and NILM needs one CT — no rewiring and no production stop — submetering data monetized, energy saving with evidence.
One reminder: when you see "wanting submetering", ask first: what types are the devices? — with many impact loads, NILM works well; with many resistive-heating loads, the accuracy is discounted — look at the device mix first, then choose the decomposition approach. NILM is not a panacea — devices with unique features decompose well — choosing the right scenario, one CT gives the whole workshop's submetering data.
Scope and Limitations
- The conclusions in this article are intended for LV distribution and electrical-safety sites. Criteria are compiled from public codes and FEXLINK product documentation, and product parameters follow the latest selection handbook and product knowledge base.
- Quantitative criteria (protection zone, discharge and residual voltage, earth resistance, thresholds and grading) hold under the preconditions stated in this article. Where the site voltage level, earthing arrangement, environmental category or load profile differs, the figures must be recalculated. Specific configurations involving Tianyan engine E-01, Wanxiang engine V5.0 NILM should be verified against actual circuit parameters.
- Specific product models and parameters are subject to the latest selection handbook and product knowledge base; this article does not constitute selection or procurement advice.
- Standard numbers cited here refer to the currently effective editions; when a standard is revised, the related quantitative conclusions must be re-checked.
- Deployment and implementation for specific projects must be carried out by qualified engineers based on site survey and design drawings.
Frequently Asked Questions
NILM Non-Intrusive Load Monitoring
An old workshop wants submetering, but installing a meter on every machine is impossible — is there another way