Energy Mark

How it works

Built on historical test results, Energy Mark predicts energy performance as each new test progresses.

Ten plain steps, from the logger on your rig to the decision on your screen.

Logger dataYour existing exportsQuality gateBad data stops hereTest statePhases, cycles, settlingPredictionFinal result, with a rangeRating formulaYour standard, as publishedEarly readOne of four readsEach certified result is compared with the prediction and feeds the learning loopLogger dataYour existing exportsQuality gateBad data stops hereTest statePhases, cycles, settlingPredictionFinal result, with a rangeRating formulaYour standard, as publishedEarly readOne of four readsEach certified result feeds the learning loop.

Step by step

From the rig to the result.

  1. 01

    Readings come in

    Your data logger records power, voltage, temperatures and humidity every few seconds. Energy Mark picks up its export files or live feed.

    DetailNo new sensors or hardware. See the integrations page for supported instruments.

  2. 02

    The test is identified

    Each run is linked to one unit, its model and the band it is designed for.

    DetailUnit number, model, rated values, lab and rig.

  3. 03

    The data is checked

    Gaps, faulty sensors and ambient conditions outside the test limits are caught before any prediction.

    DetailIf the data fails, the test goes to Engineer review.

  4. 04

    The test phase is found

    It works out where the test is: pull-down, steady running, defrost or a load point.

    DetailCounted in cycles and phases, not just hours.

  5. 05

    The behaviour is measured

    From the trace it measures how the product behaves: how long the compressor runs, how fast it cools, how much energy each cycle uses.

    DetailThe measures are chosen for each product family.

  6. 06

    The final number is predicted

    A model trained on your own completed tests predicts the final energy result, with a range.

    DetailFor example: 412 kWh/yr, plus or minus 9.

  7. 07

    It checks it is sure enough

    If the range is too wide or the design is unfamiliar, it says Continue testing instead of guessing.

    DetailIt checks again at the next point in the test.

  8. 08

    The band is worked out

    Your standard's published formula turns the number into a band. No model guesses the band.

    DetailThe formula version is recorded with every result.

  9. 09

    Your team sees the decision

    Target match, Outside target, Continue testing or Engineer review, with the reasons behind it.

    DetailEvery input, model version and decision is logged.

  10. 10

    It learns from the certified result

    When the full test ends, the certified result is compared with the prediction.

    DetailModels are retrained on your premises, with your sign-off.

When is it ready to predict?

Found in your data, not promised in advance.

Readiness differs by product and test. Energy Mark checks several points in the run and keeps the earliest one where predictions hold up on units the model has never seen.

Time checkpoints

30 min60 min90 min2 hLonger for long tests

Event checkpoints

Compressor cyclesLoad points completedProgramme phasesThermal settlingCharge cycles

How accuracy is measured

On whole units the model has never seen.

  • Prediction error on the final energy result.
  • Range coverage: how often the final result falls inside the predicted range, and how wide the range is.
  • False target matches: how often an early Target match was contradicted by the full test.
  • Early-read coverage: the share of tests that get a confident early call.
  • Time to a reliable read, as a median and 90th percentile.

Data it needs

Mostly what the lab already has.

  • Exports from your existing data logger.
  • Completed tests with their final certified results.
  • Basic product details: rated values, intended band and revision.
  • Unit identity, lab and test station for each test.

Data it will not use

Excluded on purpose, and logged as excluded.

  • Corrupt, unreadable or duplicate sessions.
  • Sessions with an unknown unit or a missing final result.
  • Sensor error values treated as real readings.
  • Sessions with unresolved timestamp gaps.
  • Readings from one unit split across training and testing.

Questions

Plain answers.

What is an early read?

An early read is Energy Mark's prediction of a product's final energy result, made while the energy test is still running. It comes with a prediction range and one of four decisions: Target match, Outside target, Continue testing or Engineer review.

What is a prediction range?

The prediction range is the band of values the final certified result is expected to fall within, for example 412 kWh/yr plus or minus 9. A narrower range means more certainty. Energy Mark only gives an early read when the range is narrow enough to act on.

How early can Energy Mark predict the result?

It depends on the product and the test. Energy Mark checks at set points in the run and gives an early read only when there is enough evidence. The POC measures how early predictions hold for your own products.

Does Energy Mark replace the certified energy test?

No. The full test still runs and certifies the result. Energy Mark is a predictor that tells your team where the result is heading, so they can act sooner.

What data does Energy Mark need?

Exports or live feeds from the data loggers and power analysers you already use, your completed tests with their certified results, and basic product details such as rated values and the intended band.

How is the energy band worked out?

The model predicts the energy value. Your standard's published formula then turns that value into a band. No model guesses the band, and the formula version is recorded with every result.

Proof of concept

See it predict your own tests.

A POC on your completed tests shows how early the final result becomes clear for your products.

Start a POC