Redundancy: what happens when a unit fails in week 7 of flower

Lose one dehumidifier out of a bank of five, mid-afternoon, in a 200 m² room running dense colas through week 7, and the room does not drift gently. Research on greenhouse cannabis reports inflorescences destroyed rapidly above 70% RH at 17–24 °C ^[1], and this room can cross from a 50% RH setpoint into that range within about 17 minutes, modelled below. That number is the entire argument for this article: redundancy planning is not really about hardware, it is about whether your detection and response can beat a clock that short.

Time to cross 70% RH after one unit fails at peak load

~17min

modelled example below, 200 m² room, one of five units lost

RH above which greenhouse cannabis inflorescences were destroyed rapidly, 17–24 °C

>70% RH

greenhouse cannabis disease research, cited below

Late-flower RH ceiling common commercial sizing guidance targets

40–50% RH

vendor sizing guidance, cited below

Typical spare capacity commercial N+1 sizing adds over bare design load

20–30%

vendor sizing guidance, cited below

What a lost unit does to the room, minute by minute

This is deliberately the worst moment to lose a unit, not a random one. A dense late-flower canopy under full light is close to its highest transpiration rate of the whole cycle, the same load that drives the lights-off humidity spike once photosynthesis stops and the room's sensible cooling drops with it; this page starts from that transpiration load rather than re-deriving it, and asks a narrower question: what happens when the equipment meant to handle it loses a piece, mid-afternoon, at the load's peak.

Take a realistic case: a 200 m² canopy of dense, late-flower colas, ceiling height 4.5 m to the return-air grille, so roughly 900 m³ of room air. A moderate design load for that stage, well short of the roughly 10 L/m²/day peak one industry source gives for maximum canopy demand ^[2], is about 5 L of applied water per square metre per day, of which roughly 95% ends up as vapour in the room air rather than in the runoff ^[2]. That is 950 L (950 kg) of water vapour a day. Concentrate it, as transpiration genuinely does, into the 12-hour lights-on window and the room needs to shed around 80 kg of water vapour an hour at the photoperiod's peak.

Suppose that load is met by five identical dehumidification units, 16 kg/h each, installed with no spare: five units for an 80 kg/h load, capacity matched exactly to demand. That is an N system, and it is a common way rooms actually get built, because on paper the maths balances.

One unit trips. The other four keep running, removing 64 kg/h. The room still needs 80 kg/h shed. The shortfall, 16 kg/h, has nowhere to go but into the room air.

The room's own air holds very little spare capacity for that surplus. Using Tetens' approximation for saturation vapour pressure ^[3] and a simple mass balance on the room's 900 m³ of air:

Vapour mass in room air (kg) = W × dry-air mass
W = 0.622 × e / (101.325 − e)            (humidity ratio, kPa)
e = RH/100 × SVP(T)                       (Tetens, kPa) [3]

At 26 °C and 50% RH, that 900 m³ of air holds about 11 kg of water vapour. At 70% RH it holds about 16 kg. The gap is under 5 kg, against a 16 kg/h surplus.

By that mass balance, the room crosses 70% RH at about 17 minutes and is approaching 85% by 30 minutes, at which point real rooms stop climbing as fast as the model suggests: surplus vapour starts condensing on the coldest surfaces in the room (duct runs, coil fins, cool wall sections) rather than staying airborne, which is itself a hazard, since it puts free water where dense flowers can pick it up. Two things make a real room somewhat more forgiving than this simplified model: a running sensible-cooling unit often condenses some moisture incidentally even without a dedicated dehumidifier online, and a larger, less densely planted room has more air to buffer the same surplus. Neither changes the direction of the result. A meaningful fraction of a room's dehumidification capacity, lost at a genuinely peak moment, buys you tens of minutes, not hours.

Why dehumidification failures outrank almost every other HVAC fault in late flower

Not every unit failure carries the same stakes, and treating them as equivalent is how facilities over-spend on redundancy in the wrong place. Rank them by what they threaten:

A cooling-only fault raises temperature and shifts VPD, which is unpleasant and yield-relevant but rarely destroys a crop inside a single shift. A dehumidification fault in dense, tightly packed colas is different in kind: moisture gets trapped between bracts, and the internal microclimate of a mature inflorescence has been measured running about 15 percentage points of RH and 2.5 °C above the surrounding room air by day 49 of flower ^[1] (reported), so what the room sensor reads is already an underestimate of the risk inside the bud. That is the failure mode worth engineering around first; what the resulting disease actually looks like on the plant is covered in bud rot.

A third fault deserves a mention here only to rule it out: two units that are both running but working against each other, one heating while the other cools, is a controls and sequencing problem rather than a capacity one, and it is covered on its own terms in why HVAC units fight each other. It will not save you from the failure modelled here, and a poorly sequenced room can mask a genuine capacity loss for longer, since nobody notices one unit has stopped when the others always seemed to be fighting anyway.

How much time you actually have to respond

Detection and response have to beat the clock the model above draws, and most facilities are not set up to do that with people alone. A member of staff on a normal round, if one happens to be walking that room, might notice a hygrometer reading, work out what it means, and get to the panel within 15 to 30 minutes on a good day. That is already close to the 17-minute mark this scenario crosses into risk. A facility running unstaffed overnight, which many licensed operations do deliberately to control labour cost, might have nobody on site for eight hours or more.

Put plainly: manual response, even from an attentive, well-staffed team, is racing a clock measured in minutes. It usually loses. That is the case for building the response into the equipment and the alarm system rather than into a rota.

N, N+1 or 2N: how much spare capacity the risk justifies

Three commonly used tiers, and what each one actually buys:

There is no default answer here, and a page that gave you one would be doing your risk assessment for you badly. N accepts the 17-minute clock as a cost of doing business, which only makes sense if the exposure is small (a small canopy, an early flower stage, a facility that can absorb the occasional lost room). N+1 is standard commercial practice for a room where dense late flower is the normal operating state, because the cost of one spare unit is trivial next to the value of the canopy it protects. 2N is rarely justified by a single flower room; it earns its cost when losing an entire mechanical system would take out more than one room, or more than one harvest, at once.

Vendor guidance on commercial installations backs the same shape: for loads above roughly 280 L/day (600 pints per day, the unit most dehumidifier nameplates use), a bank of smaller units is generally recommended over one large unit specifically for redundancy, alongside better air distribution and easier servicing ^[6]. A multi-unit N+1 layout gets you that spare almost for free, since you were already splitting capacity across several machines.

A spare unit is not redundancy until the controls prove it

The single biggest gap between nameplate redundancy and real protection is automatic failover. Take the five-unit room from the worked example and add a sixth as its spare: if that sixth unit sits idle, wired to start only when someone walks over and switches it on, it is not N+1 during the 17 minutes that matter; it is N with an extra step sitting in the corner. Real N+1 protection needs:

• A sensed trigger. The controller must detect the failed unit directly (a run-status signal, a current sensor, a fault contact), not infer it from a slowly climbing room RH reading ten minutes later.
• An automatic start sequence, not a manual switch, so the spare is already removing moisture before a human has been notified anything is wrong.
• A confirmed run, not just a start command: the controller should verify the spare is actually producing airflow and condensate, since a spare that starts and immediately faults is worse than no spare, because it delays the human response too.

This is also where [!SAFETY]-relevant power planning belongs, since a spare unit is worthless if the event that took out the first one, a breaker trip, a phase loss, a site-wide power dip, took out the spare's supply too.

Alarms and escalation: who gets called, and how fast

An alarm that nobody sees until the next shift is not protection, it is a timestamp for the post-mortem. A workable escalation path for a late-flower dehumidification fault needs:

• A threshold set below the danger zone, not at it. Alerting at 55% RH, with 70% as the hard-action ceiling this article has been modelling against, buys real response time instead of firing the alarm only once the room is already in trouble.
• A channel that reaches a person outside working hours: an SMS or automated call to an on-call rota, not only an email or a dashboard tile nobody is watching at 2 a.m.
• A defined second contact if the first does not acknowledge within a set window, commonly 10 to 15 minutes, so a missed call does not silently become a missed night.
• A record of the response, logged automatically where possible: when the alarm fired, when it was acknowledged, when the room returned to setpoint. That record is what lets you check, honestly, whether your actual response time beats the clock this article models for your own room and load.

The failure pattern this library keeps seeing

Undersized dehumidification for the plant density a room eventually reaches is a problem cannabis-facility HVAC engineers say they see constantly ^[7]: a room designed and commissioned for a lighter early canopy, then planted out to full commercial density without anyone re-running the latent-load numbers or adding the spare capacity that density now justifies. Pair that with no monitored alarm at all and it becomes one of the most expensive HVAC-D mistakes in commercial cultivation, because nobody is watching for the gap it opens. It is not usually one dramatic equipment failure that causes the loss. It is ordinary equipment ageing into a load nobody resized, with no alarm to catch the gap before a grower notices humidity by smell.

The fix is not exotic: size to the canopy you will actually run, not the one you started with, add the spare unit while the electrical and mechanical rough-in is cheap to change, and wire the alarm before the first flower room ever reaches peak density. Facility design decisions belong with the grow-spaces section, and the load-sizing maths itself is covered in full in HVAC-D sizing: latent vs sensible load; this page assumes you have already done that sizing and is asking what happens when the equipment you sized still fails.

A spare that auto-starts and an alarm that reaches someone are also the two things a proper commissioning process checks before a room is ever signed off; see the commissioning checklist for the fuller pre-handover list this article's failover test draws from.

Testing redundancy before week 7 forces the question

A spare that has never been made to actually take over is a guess, not a plan. Test it deliberately, during a lower-stakes stage:

1. Pick a low-risk windowplan ahead

   Run the test in vegetative growth or early flower, not in the last three weeks of a valuable late-flower room. The plants are more tolerant, and a real failure during the test costs you far less.

2. Isolate one unit, not the room5 min

   Take one dehumidification unit offline at its controller or breaker, the way an actual fault would, rather than simply switching it to standby. You are testing failure, not a manual toggle.

3. Time the automatic response

   Confirm the spare starts without human input, note how long it takes to confirm a run, and note whether the alarm actually fires and reaches the on-call contact.

   Warning If nothing starts within your alarm's stated response window, you do not have N+1, whatever the nameplate capacity says.
4. Watch the room, not just the equipment30–60 min

   Log RH through the test. Compare the room's real trajectory against a model like the one above, built for your own canopy, room volume and installed capacity.

5. Log it and restore N5 min

   Record the response time achieved, fix anything that underperformed, and bring the tested unit back online before ending the test.

Do this once, honestly, and you replace a guess about your redundancy with a measured number, the same kind of number this article has been modelling throughout, except now it is yours. Compliance and environmental record-keeping requirements vary by jurisdiction; check the relevant page in the law section for what your regulator expects you to log and retain. For comparing specific dehumidifier or AHU models against each other on capacity, footprint and rated conditions, that vendor-neutral comparison work lives in the equipment section, not here; this page is about what happens once the unit you already own goes offline, and whether anything catches it before week 7 notices.