Heat pumps: defrost, wrong mode, capacity, staging and the source loop

Workbook exercise plant-heat-pump · heat-pump guides (no PNNL re-tuning chapter covers heat pumps) · about 50 minutes

Goal

Find the heat-pump faults that no open dataset in the catalog shows: a unit that defrosts far too often, a unit that heats a room that is already too warm, a unit that falls behind in a cold snap, a compressor stage that hunts, two units that fight in one room, and a source loop that pumps water without carrying heat. Then find out what a 15-minute trend hides, and how much the defrost verdict depends on the limit you choose.

Learn more

No PNNL re-tuning guide or chapter covers heat pumps. Read these instead (free):

• Heating and cooling with a heat pump (Natural Resources Canada): its sections on the defrost cycle, on supplementary heat and the balance point, and on thermostats explain the defrost, capacity and mode parts of this exercise.
• Guide to Installing Air-Source Heat Pumps in Cold Climates (NEEP): its thermostat section says where a unit's thermostat belongs, which is what goes wrong in the room with two units.
• Variable Flow Pumping (EPRI technical brief TB-114673): why a water-source loop whose pumps run at constant speed moves water it does not need.
• Chapter 3 of the small-building re-tuning course (PNNL): its packaged-unit and thermostat sections cover the settings that make a compressor cycle or stage.

Datasets and licence

• synthetic-heat-pump: the synthetic heat pumps of one small building, which CAMBER generates on your computer. Nothing is downloaded and no real building is behind it. It covers a fortnight of late winter, every 5 minutes: damp days around freezing, a three-day cold snap, two days of mild fog and a sunny spell, so the units both heat and cool. Licence Apache-2.0 (CAMBER's own; the data may be shared). About 3 MB. See synthetic datasets for how CAMBER generates it.

Twelve pieces of equipment are ingested into the facility ds-synthetic-heat-pump:

• four air-source units (class ASHP), each trended with its zone and discharge-air temperatures, fan, occupancy, heating and cooling setpoints, compressor status and reversing valve (1 = heating): ASHP__healthy, ASHP__defrost_sensor, ASHP__defrost_timer and ASHP__no_aux_heat;
• six water-source units (class WSHP), trended with the same points but the compressor stage (0, 1 or 2) instead of the compressor and valve: WSHP__classroom and WSHP__classroom_twin share one room, WSHP__window_side and WSHP__door_side share another, and WSHP__stuck_heating and WSHP__stage_hunting each serve a room of their own;
• the water-source units' source loop (class HPLOOP), trended with its supply and return water, pump status, pump speed and differential pressure: LOOP__loop_ok and LOOP__loop_flat.

What is wrong with each is the exercise; the instructor key describes how each was built.

Setup

In the lab

1. Run camber lab and open the URL it prints (http://127.0.0.1:8765/lab?token=...).
2. Tick synthetic-heat-pump (its kind reads synthetic) and press Fetch & ingest. CAMBER generates the file instead of downloading it, then ingests it like any other dataset.
3. When the job is done, the row links to trends (the trend viewer) and to report, the audit report of the dataset's template at the trend's own 5-minute resolution. The lessons in steps 7 and 8 below use the command line.

On the command line

camber datasets fetch synthetic-heat-pump
camber datasets ingest synthetic-heat-pump --store lab_store
camber datasets config synthetic-heat-pump --exercise plant-heat-pump --store lab_store --out hp.json
camber run hp.json --out hp_out
camber datasets score synthetic-heat-pump --store lab_store --findings hp_out/findings.json

The exercise's config (--exercise plant-heat-pump) runs six rules at their defaults on the 5-minute trend: heatpump_defrost (reversing-valve changes per day), hp_mode_vs_need (heating a warm room or cooling a cold one), hp_capacity_shortfall (a room out of its band while its unit works), hp_room_imbalance (units in one room that fight), compressor_staging (stage changes per day) and source_loop_deltat (the loop's delta-T while it pumps). The units that share a room are declared in the config's topology (each unit's parent is its room), since nothing in a trend says which units share a room. Two variants run fewer rules: one on the same data resampled to 15 minutes, one with a limit of 20 reversals a day instead of 24:

camber datasets config synthetic-heat-pump --exercise plant-heat-pump--15min --store lab_store --out hp15.json
camber run hp15.json --out hp15_out
camber datasets config synthetic-heat-pump --exercise plant-heat-pump--20rev --store lab_store --out hp20.json
camber run hp20.json --out hp20_out

Steps

1. Look before you judge. In the trend viewer, open ASHP__healthy and ASHP__defrost_sensor and plot the reversing valve and the compressor over one foggy day (26 February) and one day of the cold snap (20 February). Zoom in until you can see single 5-minute samples. Note the outdoor temperature when each unit defrosts.
2. Run the exercise's config (the commands above). For the four air-source units, read heatpump_defrost in hp_out/findings.json: the severity, reversals_per_day and n_reversals.
3. Mode and capacity. For every unit, read hp_mode_vs_need (heating_while_warm_hours, cooling_while_cold_hours, wrong_mode_pct_of_running, mode_source) and hp_capacity_shortfall (heating_outside_pct, cooling_outside_pct, verdict, side). Plot the zone and discharge-air temperatures of each unit that fires.
4. The rooms. Read the hp_room_imbalance finding (its equipment is <fleet>): rooms has one entry per room, with fight_pct, active_pct, zone_sensor_spread_f and a severity. Plot the zone temperatures of WSHP__window_side and WSHP__door_side together.
5. Stages and the loop. Read compressor_staging (stage_changes_per_day) for the six water-source units and source_loop_deltat (deltat_median_f, deltat_p90_f, flat_pct, diff_press_median) for the two loops.
6. Score. Run camber datasets score and read the rates for the five scored rules, then the per-equipment lines.
7. Sample it every 15 minutes. Run the 15min variant and compare reversals_per_day and stage_changes_per_day with steps 2 and 5.
8. Choose the limit. Run the 20rev variant. Note which defrost verdicts change.

Questions

1. Which air-source unit does heatpump_defrost flag, and how badly? At what outdoor temperatures does it defrost, and why is that impossible for a coil carrying frost?
2. ASHP__defrost_timer passes, just under the limit. What makes it defrost more than the healthy unit, especially in the dry cold snap? With a limit of 20, what changes, and is it a fault?
3. Which unit heats a room that is already too warm? How does CAMBER know the unit is heating when no compressor or valve is trended? What does defrost_hours say for the air-source units, and why are those hours not cooling?
4. Which unit falls out of its heating band while it runs flat out? What does the verdict say, and what does it say for WSHP__stuck_heating, and how did that unit spend its time?
5. Which room's units fight, and for what share of the occupied time? Why does hp_mode_vs_need not see the fight, and what verdict does hp_capacity_shortfall give its two units?
6. Which unit's stages hunt? What do the stage changes per day of the other water-source units say about normal staging?
7. Which loop carries almost no heat, and what do its delta-T and DP say?
8. What does the label score say for each rule? Which unit is the false alarm for each rule that has one?
9. Resampled to 15 minutes, what happens to the defrost and stage counts and their verdicts? Why can a 15-minute trend not vouch for a 5-minute defrost?

What CAMBER shows

• Findings. camber run prints one line per finding: one per rule and unit, and one <fleet> finding for hp_room_imbalance.
• Defrost. heatpump_defrost counts every change of the reversing valve (into a defrost and out of it, and every heating/cooling changeover) and reports reversals_per_day, n_reversals and max_reversals_per_day. It warns at the limit (24) and faults at twice it. It needs a trend of 5 minutes or finer (max_interval_min): coarser, the count is a floor, and a count below the limit declines to info.
• Mode. hp_mode_vs_need reports heating_while_warm_hours, cooling_while_cold_hours and wrong_mode_pct_of_running, in occupied hours. mode_source says how it knew the mode: status (the compressor and the reversing valve) or discharge_vs_zone (discharge air at least 15 °F above the zone reads as heating, at least 10 °F below as cooling). With status, a reversal of at most 15 minutes between two heating samples is a defrost and counts as heating; defrost_hours reports it. It warns at 2 hours and 2 % of the running time, and faults at 10 hours and 10 %.
• Capacity. hp_capacity_shortfall reports, per side, the share of occupied time out of band (heating_outside_pct, cooling_outside_pct) and the share of that time the unit was working in the right mode (heating_working_pct, cooling_working_pct). Its verdict is capacity (working at least 70 % of that time), control (at most 30 %) or mixed; heating_opposite_pct and cooling_opposite_pct give the share spent in the other mode. A unit in a room whose units fight gets room_fight instead of capacity or mixed. It warns at 10 % of occupied time and faults at 25 %.
• Rooms. hp_room_imbalance reports per room fight_pct (one unit heating while another cools), the units' active_pct and its spread, and zone_sensor_spread_f. A room faults at a 10 % fight and warns at 2 %, a 40-point active spread or a 3 °F sensor spread.
• Stages. compressor_staging reports stage_changes_per_day (changes of the stage while the compressor runs; starts and stops are not counted) and max_stage. It warns at 24 changes a day and faults at 48, and like the defrost count it needs a trend of 5 minutes or finer.
• Loop. source_loop_deltat reports the loop's |delta-T| while it pumps: deltat_median_f, deltat_p90_f and flat_pct (the share under 1 °F), against a 10 °F design value. It faults when 80 % of the pumping time is flat and the 90th percentile is under a quarter of the design.
• Score. camber datasets score prints the true- and false-positive rates for the entry's five scored rules. hp_room_imbalance is a fleet rule with no per-unit finding, so it is not scored: read its rooms instead.

From this exercise's synthetic units ASHP__defrost_sensor (top) and ASHP__healthy (bottom) on 26 February: the reversing valve (1 = heating) and the compressor. Each shaded sample is a defrost: the valve reversed with the compressor running.

Caveats

• The data are synthetic: CAMBER's own model of one building and its heat pumps, built to show these faults. They show what the rules see, not how real heat pumps behave; never cite them as evidence about a building.
• A 5-minute trend still hides a defrost or a stage change shorter than 5 minutes, so the counts are a floor.
• The rules judge occupied hours (the trended occupancy, weekdays 07:00 to 18:00 here).
• Which units share a room comes from the config's topology, not from the data. On a real site it comes from the drawings or a walk-down.
• The limits are generic screening defaults (camber rules params heatpump_defrost). A real unit's defrost interval and staging delays come from its documentation and controls.

Going further

• Read the boiler cycling exercise: its resolution lesson is the same one, for a boiler on a 15-minute trend.
• Run camber rules params heatpump_defrost, camber rules params compressor_staging and camber rules params source_loop_deltat to see where each default comes from and how to calibrate it.