Skip to content
gigabiome
Try Plant Check

Plant Check · free · no login

Plant Check: read your chiller plant from a BMS export

Upload a trend CSV. AI identifies your points, physics checks the result, you confirm, and Plant Check calculates kW/ton, lift, tower approach, fan and standby energy. The written explanation can only cite those calculated numbers.

What it calculates

Efficiency
  • Chiller efficiency (energy-weighted)
  • Chiller kW/ton, best 10% of running intervals
  • Chiller kW/ton, median running interval
  • Chiller kW/ton, worst 10% of running intervals
  • Whole-plant efficiency (chillers + towers + pumps mapped)
Tower
  • Tower approach to wet-bulb (median while running)
  • Tower approach, worst 10% of running intervals
  • Tower fan energy per ton-hour
  • Share of running hours with the mapped tower fan at ≥95% speed
Loop
  • Lift (median while running)
  • Chilled-water delta-T (median while running)
  • Condenser-water range (median while running)
Standby
  • Share of plant electricity used in standby hours
  • Hours with chillers on below 5% of peak load

25 metrics in all · calculated from your data, never by AI

01Pick data

Sample data: LBNL's public simulated chiller plant. About the data Run a sample, or upload an export from your own plant.

Checking your browser…

Sample data

The sample data

Where the two samples come from, and what the physics check finds in them.

Source
LBNL's open fault-detection datasets for a chiller plant. Granderson, J., Lin, G., Chen, Y., Casillas, A., et al. LBNL Fault Detection and Diagnostics Datasets. Lawrence Berkeley National Laboratory, 2022. Open Energy Data Initiative. https://doi.org/10.25984/1881324 (CC BY 4.0)
Licence
CC BY 4.0
Processing
July 2018 slices, resampled from 1-minute to 15-minute averages; the original 77 point names, as exported, not renamed.
  • Sample plant, no faults (July)Three chillers and three cooling towers serving a 12-storey Chicago office, fault-free case.2,976 rows · Jul 1, 2018 – Jul 31, 2018
  • Sample plant, cooling-tower fouling (July)The same plant with cooling tower 1's heat-transfer coefficient reduced to 65% (LBNL's 'fouling' fault).2,976 rows · Jul 1, 2018 – Jul 31, 2018

What the tower-fouling case shows

The OA_TEMP / OA_TEMP_WB finding

Mapped by their names, as the LBNL inventory labels them, the two outdoor-air points fail the physics check “Wet-bulb is not above dry-bulb”. It reports:

Wet-bulb is above dry-bulb in 98% of rows, which is physically impossible. The two columns may be swapped or mislabelled in the export.
PointGoing by the labelReference mapping
OA_TEMPOutdoor dry-bulb temperatureOutdoor wet-bulb temperature
OA_TEMP_WBOutdoor wet-bulb temperatureOutdoor dry-bulb temperature

The check flags; it never changes a mapping. The reference mapping swaps the pair because of the plant's control sequence: LBNL inventory §1.2, Eq. 3 (tower leaving-water setpoint = wet-bulb + 8 °F). The tower setpoint CT_SW_TEMPSPT minus each column, over all 2,976 rows:

  • CT_SW_TEMPSPT − OA_TEMP: mean 7.93 °F, sd 0.76
  • CT_SW_TEMPSPT − OA_TEMP_WB: mean 1.16 °F, sd 4.86

In Plant Check, the AI proposes a mapping for these files like any other, and you confirm it or swap the pair.

What we store

Uploaded files are processed in memory and never stored. A profile of each point (its name, range and a few sample values), your mapping and the calculated results are kept, so running the same file again reuses them.

Methods and data: how every number on this site is calculated →

Method

How Plant Check works

AI does two jobs: it proposes what your points are, and it writes the explanation. It never produces a number. Every figure is calculated directly from your data.

Map: AI proposes, you confirm

AI reads your point names, value ranges and a few sample rows, and proposes what each point measures and its unit, with a confidence level and a one-line reason. You confirm or change every row. Nothing is calculated until you do.

Check: physics before math

Before anything is calculated, fixed engineering rules test the mapping against physics: supply colder than return, wet-bulb never above dry-bulb, plausible temperatures and fan speeds, no negative power. A check can flag a mapping; it never changes one.

Calculate: standard engineering math, no AI

Every number is calculated directly from your data, using the mapping you confirmed: energy-weighted kW/ton, lift, tower approach, fan energy per ton-hour and standby energy. The same data always gives the same numbers.

Explain: AI writes it, but can't invent a number

AI writes a short explanation of the results. It can only cite figures we calculated; we insert the values. If it writes a number of its own, or cites a figure we never calculated, the whole explanation is rejected.

Where AI is used in Gigabiome, and where it isn't

  • Identify your points

    In Plant Check
    Done by
    AI proposes what each point is; you confirm every row
    Does it produce numbers?
    No. It only labels points. Nothing is calculated until you confirm.

  • Physics checks

    In Plant Check
    Done by
    Fixed engineering rules
    Does it produce numbers?
    No. They flag a suspect mapping; they never change it.

  • Performance figures

    In Plant Check
    Done by
    Standard engineering formulas, applied to your data
    Does it produce numbers?
    Yes. This is the only step that produces numbers.

  • Written explanation

    In Plant Check
    Done by
    AI, citing only figures we calculated
    Does it produce numbers?
    No. If it writes a number of its own, the explanation is rejected.

  • Baseline and verified savings

    In the monthly service
    Done by
    A weather- and load-normalized baseline, verified to IPMVP
    Does it produce numbers?
    Yes, calculated from each pilot site's own data each month

  • Recommendations

    In the monthly service
    Done by
    An optimizer within guardrails
    Does it produce numbers?
    Will propose changes; none reaches the plant without operator approval