Chiller efficiency: calibrating a design ceiling, and chiller fouling

Workbook exercise plant-chiller-efficiency · PNNL re-tuning chapter 8 · about 45 minutes

Goal

Judge a chiller's kW/ton against a ceiling that fits this chiller. First see what a generic ceiling does to a healthy plant. Then calibrate the ceiling from the plant's own fault-free run and find the runs that really cost energy: a fouled chiller and a leaking tower bypass. Finally, score the rule against the dataset's published labels.

Learn more

Read these first (PNNL, free):

Central Utility Plant Cooling Control, the re-tuning guide, for how a water-cooled plant is controlled and which points to trend. It judges the plant by its controls (reset, delta-T, pumping; see the reset and pumping exercise) rather than by kW/ton, which is what this exercise adds.
Chapter 8: Central Utility Plant of the re-tuning training, for the central plant as a whole.

Datasets and licence

lbnl-chiller: LBNL's simulated water-side chiller plant (three chillers, three cooling towers, primary and secondary chilled-water loops), a year at one-minute resolution, as a fault-free run plus labelled faulted runs. Licence CC-BY-4.0 (open; cite it). The download is about 1.6 GB; this exercise ingests the full subset (all 24 runs). The catalog lists the problems found in the published data and how CAMBER handles each: see its data issues.

Each run becomes one piece of equipment, PLANT__<run>, in the facility ds-lbnl-chiller (chiller 1 and tower 1 are the mapped units). This exercise reads PLANT__fault_free, the two chiller-fouling runs (PLANT__chiller_fouling_065, the severe one, and PLANT__chiller_fouling_095), the tower-fouling runs (PLANT__coolingtower_fouling_065 ...), the tower-bypass runs (PLANT__bypass_leakage_025 ... PLANT__bypass_stuck_075) and, for the score, every other run.

Setup

In the lab

Run camber lab and open the URL it prints (http://127.0.0.1:8765/lab?token=...).
Choose the full subset, tick lbnl-chiller and press Fetch & ingest.
When the job is done, the row links to trends (the trend viewer) and report (the dataset's default config, which is this exercise's second run). The first run uses its own config: use the command line for it.

On the command line

camber datasets fetch lbnl-chiller
camber datasets ingest lbnl-chiller --subset full --store lab_store
camber datasets config lbnl-chiller --exercise plant-chiller-efficiency--generic --store lab_store --out generic.json
camber run generic.json --out generic_out
camber datasets score lbnl-chiller --store lab_store --findings generic_out/findings.json
camber datasets config lbnl-chiller --store lab_store --out chiller.json
camber run chiller.json --out chiller_out
camber datasets score lbnl-chiller --store lab_store --findings chiller_out/findings.json

The first config (--exercise plant-chiller-efficiency--generic) runs chiller_efficiency with the rule's own generic ceiling of 0.85 kW/ton. The second is the dataset's default config: the same rule with a ceiling calibrated from the fault-free run (its value is in the config), because the simulated chiller's design curve is not published, plus cooling_tower_approach (the subject of the cooling-tower exercise). Read the _comment in each config file.

Steps

Look before you judge. In the trend viewer, open PLANT__fault_free and plot the chiller power with the chilled-water supply and return temperatures over a summer week. When is the chiller loaded, and when does it idle?
Run the generic config and read what camber run prints for PLANT__fault_free. Score it with camber datasets score.
Run the calibrated config and read chiller_efficiency for the fault-free run again.
Chiller fouling. Compare kw_per_ton_median and pct_hours_inefficient in chiller_out/findings.json for PLANT__chiller_fouling_065, PLANT__chiller_fouling_095 and PLANT__fault_free.
Other faults. Find the other runs chiller_efficiency flags, and look at the worst tower fouling, PLANT__coolingtower_fouling_065.
Score. Run camber datasets score on the calibrated findings and read the per-detector rates and the list of which rule fired on which run.

Questions

With the generic ceiling, what does CAMBER say about the fault-free plant? What true- and false-positive rates does the score give chiller_efficiency, and why is a detector that catches every fault still useless here?
With the calibrated ceiling, what does the fault-free chiller read, and why is that the right reference for this plant?
How far do the two chiller-fouling runs move the kW/ton? Why does CAMBER flag one and not the other, and what in the finding still shows the milder one?
Which other runs does the rule flag, and which fault barely moves the chiller's kW/ton at all? Where would you look for that one instead?
What does the label score give chiller_efficiency with the calibrated ceiling, and which runs count as its false alarms?

What CAMBER shows

Findings. camber run prints one line per finding with its severity (fault, warn, ok) and the median kW/ton against the ceiling. findings.json holds the metrics: kw_per_ton_median, design_kw_per_ton, tons_median, load_factor_median_pct and pct_hours_inefficient (the share of loaded hours more than 15 % above the ceiling).
Severity bands. warn from 1.2 times the ceiling, fault from 1.5 times, judged only on hours when the chiller draws power and makes at least 5 tons of cooling.
Score. camber datasets score prints the true- and false-positive rates, with 95 % intervals, for the dataset's declared detectors, and which rules fired on each labelled run.

Chiller kW per ton against cooling load; most part-load points sit above the warn ceiling

Synthetic illustration, not this exercise's dataset: kW/ton against load, drawn with CAMBER's diagnostic scatter and a band up to the rule's warn ceiling (1.2 times the 0.85 kW/ton design). The degraded chiller crosses it at part load.

Caveats

The data are simulated. A real chiller's ceiling comes from its schedule (kW/ton at design and part load), or from its own commissioned, healthy operation, never from a generic number.
Calibrating on the fault-free run assumes that run is healthy. On a real plant that is the claim to check first.
The kW/ton is computed from the chilled-water flow and temperatures. A biased temperature sensor changes the computed tons, not only the power; see the sensor exercise.
The chiller-fouling runs are in the archive but not in the dataset's inventory; their severity is read from the file names (see the data issues).
One median over a year hides when the fouling started. A drift comparison against a baseline period is the tool for that (camber drift).

Going further

Edit chiller.json and try ceilings between the two: at what ceiling does the milder chiller fouling appear, and what else appears with it?
Open the report (camber report chiller.json --out chiller.html) and compare the evidence for the two chiller-fouling runs.
Read the cooling-tower exercise for the fault this rule misses.