Packaged / DX equipment FDD

0.5 extends the vendor-neutral rule library beyond built-up AHUs and central plants to the packaged and refrigerant-side equipment that dominates small–mid commercial stock: rooftop units (RTU), heat pumps / VRF, DOAS/ERV, fan-coil units (FCU), and refrigerant-side chiller degradation. As always, rules key off Roles (not vendor tags), each ships a synthetic fixture, and each is accuracy-scored in the synthetic benchmark.

flowchart LR
    roles["Roles (compressor_status, filter_diff_press, cond_approach_temp, ...)"] --> tmpl["Equipment templates (RTU / HeatPump / DOAS / FCU)"]
    tmpl -- "completeness gating" --> rules
    subgraph rules["camber.rules.builtin — DX detectors"]
        r1[compressor_short_cycle]
        r2[compressor_staging]
        r3[heatpump_defrost]
        r4[filter_fouling]
        r5[chiller_approach_fouling]
    end
    r1 --> find[Findings]
    r2 --> find
    r3 --> find
    r4 --> find
    r5 --> find
    find --> bench["synthetic benchmark (VALIDATION.md)"]

Role-keyed DX detectors, gated by per-unit completeness, fan into Findings scored by the benchmark.

New roles

Status/stage and refrigerant-side signals (all with PHYSICAL_BOUNDS + a Haystack hint):

Role	Meaning
compressor_status	DX compressor running (1) / off (0)
compressor_stage	active DX cooling stage (0,1,2,…)
condenser_fan_status	condenser/outdoor fan running
heat_stage	active gas/electric heating stage
reversing_valve_cmd	heat-pump mode: heating (1) / cooling (0)
filter_diff_press	differential pressure across the air filter (inH2O)
supply_air_humidity / return_air_humidity	air-side relative humidity (%)
cond_approach_temp / evap_approach_temp	chiller approach temperatures (°F)

New equipment templates

RTU, HeatPump (VRF), DOAS (ERV via optional humidity roles), and FCU (now a distinct template, not the VAV alias). Completeness validation gates which rules can run per unit — e.g. HeatPump requires a reversing-valve command; DOAS requires outdoor-air flow.

New rules

• compressor_short_cycle — DX compressor firing in short bursts (starts/day past a min-off-time ceiling). Reuses the generic on/off start counter. Flag: max_starts_per_day. On a trend coarser than max_interval_min (15 minutes) the count is a floor: a verdict below the ceiling declines to info when the resampled status shows the compressor on for part of an interval (see TIME-HANDLING.md).
• compressor_staging — unstable multi-stage DX (excess stage changes/day). Flag: max_changes_per_day. Since 0.106 only changes while the compressor runs count: a start or a stop (to or from stage 0) is compressor_short_cycle's to judge.
• heatpump_defrost — excess heat-pump defrost / reversing-valve cycling (iced coil or faulty defrost termination). Flag: max_reversals_per_day. On a resampled valve, a bin part-way between heating and cooling counts the fewest changes it allows (two for a defrost inside a heating spell).

Both rules take max_interval_min (default 5 minutes, since 0.106): a defrost or a short stage averages away in a coarser bin, so the count is a floor. A count at or above the limit keeps its verdict with a caveat; below it, samples part-way between two states decline the verdict to info ("cannot confirm on a 15-minute trend"). - filter_fouling — air filter at/above its change-out differential pressure (wasted fan energy, starved airflow). Flag: change_dp_inwc (default ~1.0 inH2O). - chiller_approach_fouling — condenser/evaporator approach-temperature degradation (tube fouling / low charge), the refrigerant-side indicator that needs no refrigerant-pressure instrumentation. Flags: cond_design_f, evap_design_f.

Every rule returns info (not a false fault) when its required role is absent, is registered in camber.rules.builtin, and runs unchanged across any building once its points are mapped.

Refrigerant-side DX and heat-pump rules (0.93)

Provisional (GitHub issues #40 and #6). Three detectors read a DX unit's or heat pump's refrigerant circuit and indoor coil. Each has two modes:

• Target mode (analyze, the ordinary camber run): the unit's median in cooling operation is compared with a manufacturer target. Pass targets={"subcooling_f": 10}, or per equipment {"RTU-*": {...}}. Without a target, only wide universal limits apply.
• Baseline mode (analyze_periods, the new dx drift family, camber drift): the unit is compared with its own frozen fault-free period at matched conditions. The normalizer is outdoor air plus return air for an air-cooled unit, or load in tons for a water-cooled machine. The finding reports the current median residual against both a degF floor and a sigma floor.

Rows count only in cooling operation: compressor on, reversing valve in cooling and supply air below return, whichever of those points are mapped. The inputs can be controller-reported or derived from pressures and line temperatures (see Refrigerant properties).

Rule	Reads	Verdicts
dx_refrigerant_charge (built-in)	liquid subcooling; superheat as corroboration (metric="superheat" for a fixed-orifice unit)	undercharge, overcharge
dx_indoor_airflow (built-in)	evaporator temperature split, return minus supply, matched on return air and the return dew point (return_air_dewpoint_temp, or computed from return RH)	airflow_low, airflow_high; a narrowed split with subcooling down is reported as capacity_low (charge), not high airflow
discharge_superheat_drift (dx family, #6)	discharge superheat, two-sided, plus the family's CUSUM for a sustained shift	rose: starved compressor (low charge, restriction); fell: liquid reaching it (overcharge, floodback)

Without targets the universal limits are 2-25 °F of subcooling (below 2 °F the valve is fed flash gas) and an 8-28 °F split. The magnitude floors are screening-grade and were characterized on the NIST heat pumps below. They are constructor arguments.

Equipment classes. A new dx family covers split and packaged AC and condensing units that are not heat pumps (DX, SPLIT_SYSTEM, CONDENSING_UNIT, ACCU, ...). dx_refrigerant_charge runs on heat_pump, dx and air_handler (an RTU's DX section). dx_indoor_airflow runs on heat_pump and dx only: a chilled-water AHU's split follows its valve, not its airflow.

Scored on nist-heatpump-fdd

The data are 56 scoreable labelled runs: two residential R-410A split heat pumps with TXVs, each on two line sets, tested in steady state. The 17 mislabelled EF files, the 6 CF_50 files and the 2 JUNK files are excluded as the catalog documents. Each test file (one steady-state test, 269 files) was scored against the fault-free files of the same unit and line set, leaving the file itself out (baseline mode, one averaged row per test point). The rates are Wilson 95% intervals.

Detector	Target faults	TPR	FPR (fault-free files)	Fires on other faults
dx_refrigerant_charge	under- / overcharge (incl. doubles)	86/95 = 91% [83-95]	7/90 = 8% [4-15]	34/84 = 40%
dx_indoor_airflow	low / high indoor airflow (incl. doubles)	20/93 = 22% [14-31]	2/90 = 2% [1-8]	2/86 = 2%
discharge_superheat_drift	under- / overcharge	30/95 = 32% [23-41]	5/90 = 6% [2-12]	27/84 = 32%

What the numbers say:

• Charge. Every overcharge file (16/16) and 70 of 79 undercharge files are caught. All 34 of its firings on other faults are condenser-blockage tests, alone or with an airflow fault, and mostly on the 16 SEER unit, whose subcooling falls 4-9 °F with a blocked coil exactly as it does with undercharge. A charge verdict is therefore a prompt to check the charge and the condenser coil. Scored per run, with fault-free runs split into two halves by file, it catches 23/24 charge runs at 0/8 false alarms. The run template's target mode, with targets taken from the fault-free runs and so in-sample, catches 24/24 at 0/4.
• Indoor airflow. At 15% or more airflow reduction (EF ≤ 85%) it catches 18/44 files. At 5-14% (0/22) and on increased airflow (2/27) the split change sits inside the fault-free scatter. Most of the labelled airflow tests are double faults with undercharge or condenser blockage, which pull the split the other way. It almost never fires on a non-airflow fault (2%).
• Discharge superheat (#6). Weaker than subcooling on these TXV units, confirming #6's original finding that it is the less sensitive signal. Undercharge raises it (30/79 files); overcharge barely moves it (0/16). Two of its five fault-free false alarms are the two files that run with negative suction superheat (fault-free-points-not-steady-cooling).

All seven charge false alarms are on the 14 SEER unit. Three are files with essentially no subcooling: the two 2016-03-15 files in that data issue, and a two-point long-line-set file at -1.1 °F. The other four sit 3.2-4.3 °F below their reference, just past the 3 °F floor. They are left as published and count against the rule.

Water-source heat pumps with three points (0.93)

Water-to-air heat pumps in a school or office are often trended with only discharge air, fan status and zone temperature: no compressor status, no reversing valve and no refrigerant point. rules.heatpump_ops_rule.infer_hp_mode reads the mode from the discharge air against the zone: 15 °F or more above means heating, 10 °F or more below means cooling. A mapped compressor status and reversing valve take precedence. Three rules follow, all on the heat_pump class:

• hp_mode_vs_need: occupied hours spent cooling a room already below its heating setpoint, or heating one above its cooling setpoint (changeover, thermostat location, setpoints).
• hp_capacity_shortfall: the room outside its band. When the unit was already working in the right mode, the verdict is capacity (unit size, airflow, source-loop temperature, refrigerant). When the unit was idle or running the other way, the verdict is control (thermostat, setpoint, schedule, a stuck changeover), and the summary gives the split. This is the heat-pump counterpart of the VAV "zone below setpoint with reheat saturated" rule, which is for terminal boxes only. When hp_room_imbalance runs in the same config and finds a unit's room fighting, a capacity or mixed verdict on that unit becomes room_fight (the severity stands, verdict_alone keeps the old verdict): the cause is the fight, not the unit's size.
• hp_room_imbalance (fleet): two or more units serving one room that fight (one heats while another cools), split the work very unevenly, or whose zone sensors disagree. Rooms come from the rule's rooms={room: [units]} or from a topology whose parent of each unit is its room.

Setpoints are the mapped heat_sp / cool_sp. Otherwise a stated default band of 68-76 °F applies, with a caveat.

Defrost (0.106). With the mode read from the compressor and the reversing valve, a cooling run of at most max_defrost_min (15 minutes) with heating on both sides is an air-source unit's defrost, and all three rules count it as heating. hp_mode_vs_need reports the occupied defrost time as defrost_hours; heatpump_defrost judges how often a unit defrosts.

Source loop. source_loop_deltat (class family source_loop: GEO_LOOP, GROUND_LOOP, ...) reads a heat-pump loop through the new generic roles source_loop_supply_temp, source_loop_return_temp, source_loop_diff_press and source_loop_pump_speed. These are kept apart from the chilled-water and tower roles, whose rules would misread such a loop. While the pumps run, it flags a loop that carries almost no heat. The warn condition is a |ΔT| under 1 °F for half the pumping time with a 90th percentile below half the ~10 °F design difference; the fault condition is 80% of the time with a 90th percentile below a quarter of it. That pattern is overpumping: constant speed, no differential-pressure reset, or heat-pump isolation valves left open.

The faultlab scenarios for the five new single-equipment rules (dx_refrigerant_charge, dx_indoor_airflow, hp_mode_vs_need, hp_capacity_shortfall, source_loop_deltat) were signed off as gated synthetic keys for 0.93 (see VALIDATION.md). hp_mode_vs_need and hp_capacity_shortfall decline equipment with a heating or cooling valve and no compressor or reversing-valve signal: that is a hydronic coil unit, and on a terminal box with no recorded class the out-of-reheat case belongs to reheat_capacity_shortfall.