Customer results / Case study
Cooling appliances · Asia
How a leading manufacturer in Asia saw the final energy result before the test ended.
Energy Mark read the early part of each energy test and predicted the final annual energy consumption, so the team could act sooner.
The goal
Answers sooner, without changing the test.
Every unit ran a multi-phase energy test lasting several hours. The final annual energy figure, and the label band it falls in, normally arrives only when the last phase ends.
The lab already logged power, voltage, current and temperatures every five seconds. The goal was to use that data to see where each test would land while it was still running, so engineering and quality could act earlier and shorten the time to an answer.
How we predicted early
Four steps, on data the lab already had.
- 01
Read the live test data
Energy Mark took the readings straight from the lab's existing data logger. No new sensors and no change to the test procedure.
- 02
Found where the test was
It separated the pull-down, warm and cold phases and measured every compressor cycle: how long it ran, how often and at what power.
- 03
Predicted the final result
From the early phases, it predicted the final annual energy consumption. Accuracy was checked on units the model had never seen.
- 04
Showed the band
The standard's own published formula turned each prediction into a label band, so the team could see whether a unit was on target.
How early, how accurate
The prediction sharpened as the test ran.
Mean prediction error at five points in the test, compared with simply using the average for that model.
| Point in the test | Mean error | Largest single-unit error |
|---|---|---|
| End of pull-down | 5.2% | 10.1% |
| Warm phase, 30 min | 6.5% | 11.5% |
| Warm phase, 60 min | 5.5% | 10.8% |
| Cold phase, 30 min | 2.7% | 7.0% |
| Cold phase, 60 min | 2.5% | 6.2% |
| Model average, no test data | about 8.4% |
Unit by unit
Four of five units within 4%.
Prediction error for each unit, within the first hour of the final phase.
What it means
A shorter path to every answer.
Know the outcome before the test ends
Within the first hour of the final phase, predictions were within 2.5% on average. The team no longer has to wait for the last reading to know where a unit is heading.
Act on off-target units sooner
Units heading below their intended band showed up early in the data, so engineering can start on them while the test is still running.
Read the right moment
Compressor cycle behaviour carried the signal. Energy Mark reads at completed cycles, which is why accuracy improved sharply once cycling settled.
Trust every prediction
Unusual readings, such as logger saturation and brief voltage dropouts, were flagged before any prediction, so a bad reading never becomes a bad answer.
Accuracy measured on units held out from training. Customer name, product models and absolute energy values are withheld.
Proof of concept
See it predict your own tests.
A POC on your own completed tests, run inside your network, shows how early the final result becomes clear for your products.