Detection and early warning

Most data center fires begin small: an overheating power supply, a failing capacitor, a loose connection in a power distribution unit, or a cable damaged during installation. Such fires often smolder for minutes or hours, producing small amounts of smoke before open flame appears. If staff can find and de-energize the faulty equipment during this early stage, the fire never grows and there is no need to discharge a suppression system. Early detection is therefore the most valuable layer of data center fire protection. This lesson explains how fires develop, the types of detectors used, why the high airflow in data centers makes detection difficult, how aspirating smoke detection works, and how alarm sequences connect detection to staff response and automatic suppression.

How fires develop

A fire typically passes through four stages:

• Incipient: overheating materials release invisible particles and gases, with no visible smoke or flame. This stage can last a long time in electrical equipment.
• Smoldering: visible smoke appears, but there is little heat and no flame.
• Flaming: open flame, rising heat and rapid smoke production.
• Intense heat: the fire spreads to nearby materials, and temperatures rise quickly.

Electrical fires in IT equipment often stay in the incipient and smoldering stages for a long time because the energy source is limited and modern equipment uses flame-retardant materials. Detectors that respond to tiny amounts of smoke can give warning during these stages, whereas heat detectors and sprinkler heads respond only once a fire is flaming and hot.

Types of detectors

• Spot-type smoke detectors: individual detectors on the ceiling (and sometimes under raised floors or in ceiling voids). Photoelectric detectors sense light scattered by smoke particles and respond well to smoldering fires; ionization detectors respond to small particles from flaming fires but are now uncommon. Spot detectors respond only when smoke reaches them in sufficient concentration.
• Aspirating smoke detectors (ASD): a fan draws air continuously through a network of pipes with small sampling holes into a central detector with very high sensitivity, often a laser-based particle counter. Known by trade names such as VESDA, they can detect smoke concentrations hundreds of times lower than a typical spot detector.
• Heat detectors: respond to a fixed temperature or a rapid rise; used where smoke detectors would give false alarms (generator rooms, kitchens) and as a reliable confirmation.
• Beam detectors: a light beam across a large open space; smoke reduces the light received.
• Flame detectors: sense ultraviolet or infrared radiation from flames; used for fuel and generator areas.
• Gas detectors: for hydrogen or other gases from batteries (Lesson 7), carbon monoxide, or refrigerant leaks.
• Video smoke and flame detection: software examines camera images for smoke or flame; used in large or tall spaces.

Sensitivity is often expressed as obscuration, the percentage of light lost per meter (or per foot) of smoke-laden air. Spot photoelectric detectors are listed to alarm somewhere between about 1.6% and 12.5% obscuration per meter (0.5% to 4% per foot), commonly around 3% per meter, while aspirating systems can be set to alarm at 0.005% per meter or less.

The airflow problem

Data halls move enormous amounts of air through the IT equipment to cool it. A data hall may change its entire air volume 30 to 100 or more times per hour, compared with a few times per hour in an office. This airflow dilutes smoke, carries it away from ceiling detectors, and can blow it through cooling units where it is filtered or mixed with clean air. In hot-aisle and cold-aisle arrangements, smoke from a server is exhausted into the hot aisle and may travel directly to the cooling unit returns without passing a ceiling detector.

NFPA 72 therefore reduces the floor area covered by each spot smoke detector as air changes increase: from the normal maximum of 900 ft² (about 84 m²) at 7.5 air changes per hour or fewer, down to 125 ft² (about 11.6 m²) at 60 air changes per hour. Detectors must also be located where airflow will carry smoke to them, and not directly in the discharge of supply air.

where \(Q\) is the airflow in m³/s (or the equivalent in ft³/min with 60 minutes per hour) and \(V\) is the room volume.

Aspirating smoke detection

An aspirating smoke detection system consists of a detector unit with a fan, filter and sensing chamber, and a network of plastic pipes with sampling holes drilled at intervals. The fan draws air continuously from all holes into the detector. Because the detector is very sensitive and sees air from many points, it can detect smoke that has been heavily diluted. Sampling pipes are commonly placed:

• At the ceiling, as a replacement for spot detectors (each hole treated like a detector for spacing).
• Across the return air grilles of cooling units, where all air from the room passes (return air sampling).
• Inside hot aisle containment or above rack exhausts.
• Inside individual cabinets or under raised floors.
• In ducts and ceiling voids.

Two parameters matter in design. Transport time is how long air takes to travel from the farthest hole to the detector; NFPA 72 limits this to 120 seconds for aspirating systems, and very early warning designs often target 60 to 90 seconds. Hole sensitivity is the smoke concentration needed at a single hole to trigger the alarm; because air from one hole is diluted by air from all the others, the more holes on a pipe, the less sensitive each hole is.

where \(N\) is the number of holes with equal flow and \(S_{\text{detector}}\) the detector's alarm threshold. Design software from the manufacturer calculates transport times, flows and hole sensitivities for a pipe layout.

Alarm levels and staged response

Very sensitive detection would cause disruption if every small reading triggered an evacuation or suppression. Data center systems therefore use staged alarm levels, each with a defined response:

Requiring two independent detections (called cross-zoning or a double-knock) before releasing a suppression system reduces the risk of an unnecessary discharge caused by a single faulty detector. Clean agent systems also include a pre-discharge alarm with a time delay (often 30 seconds) to allow occupants to leave, an abort switch that holds the discharge while staff investigate, and a manual release.

The investigation stage matters. Staff must be able to find the source quickly in a large hall: zoned aspirating detection that indicates which pipe or area sees smoke, handheld sniffers, thermal imaging cameras and power monitoring that shows which circuit is drawing abnormal current all help. Procedures should define who can de-energize a rack or power distribution unit and how customers are informed.

Integration with other systems

Fire detection interacts with many other systems in a data center: HVAC shutdown or smoke control, door releases on egress paths and access control, elevator recall, power disconnection, suppression release, the building management system and the fire department connection. These interactions must be designed, documented in a cause-and-effect matrix (which inputs cause which outputs) and tested together.

In the US, NFPA 72 covers acceptance and periodic testing of fire alarm systems, and NFPA 4 (Standard for Integrated Fire Protection and Life Safety System Testing) covers end-to-end testing of interconnected systems to confirm that they work together as intended.

Shutting down cooling on a fire alarm is a difficult decision in a data center: it prevents fans from spreading smoke and helps a gaseous agent hold its concentration (Lesson 5), but without cooling, IT equipment overheats within minutes. Designs usually shut down only the units serving the affected room, and often only at the Fire 2 stage or on agent release.

Maintenance and false alarms

Detection systems need regular inspection, testing and cleaning: detectors are tested with smoke or approved aerosols, aspirating systems are checked for airflow faults, transport time and blocked holes, and filters are replaced. False alarms waste staff time, may cause unnecessary fire department responses (some jurisdictions charge for them), and, worst of all, train people to ignore alarms. Common causes include construction dust, cardboard fibers, diesel exhaust drawn in by air intakes, and incorrect sensitivity settings. Temporary disabling or covering of detectors during construction work must follow a permit process with fire watch, and must be reversed when the work ends.

Lesson 4 explains how the fire alarm system turns these detector signals into notification, fire department response and control of other building systems.