Variable Frequency Drives in Data Center Cooling Systems

Overview

Variable frequency drives (VFDs) are now the default method of controlling fans and pumps in data center cooling systems. They earn that position twice over: cube-law energy savings at part load, and the controllability that modern cooling plants need — precise speed references, network integration, soft starting and configurable failure behavior.

In a mission-critical facility, however, a VFD is not just an efficiency device. It sits directly in the cooling continuity path. Every drive serving a chilled water pump or CRAH fan is a component whose failure modes, restart behavior and network fallback strategy must be engineered deliberately, not left at factory defaults.

This guide covers the full engineering life cycle of VFDs in data center cooling: where they are used, how to select and size them, harmonics and EMC, BMS integration, redundancy architecture, emergency operation and commissioning. The content is vendor-neutral until the final section, which examines one example drive platform against the requirements developed here.

Where VFDs are used

Data center cooling systems concentrate variable-torque loads that run continuously — the ideal application profile for variable-speed control:

DATA CENTER BMS

BACnet / Modbus

VFD-1

Chilled WaterPump

VFD-2

CRAH / AHUFan

VFD-3

Cooling TowerFan

The BMS provides supervisory speed commands and monitoring over BACnet or Modbus; each drive powers and protects its own motor.

The common pattern: the BMS provides a supervisory speed or setpoint command over BACnet or Modbus, while the drive powers, protects and (often) locally controls its motor.

Fan and pump affinity laws

Centrifugal fans and pumps obey the affinity laws. For a speed change from N₁ to N₂:

Flow:      Q₂ / Q₁  =  N₂ / N₁
Pressure:  P₂ / P₁  =  (N₂ / N₁)²
Power:     Pw₂ / Pw₁ =  (N₂ / N₁)³

The cube relationship between speed and power is the economic core of the technology. At 80% speed a fan delivers roughly 80% flow at 64% pressure — but consumes only about 51% of rated shaft power. Because cooling equipment in a data center is sized for design-day conditions plus redundancy margin, most equipment runs well below 100% most of the time, and the cube law converts that oversizing into permanent energy savings.

Use the Fan Affinity Law Calculator to explore the relationships, and the VFD Energy Savings Calculator to estimate annual savings against damper or throttling control.

Drive selection

Selection starts with the application requirements, not the catalog. For data center cooling duty, the baseline requirement set is:

• Variable torque rating matched to the motor, with ~110% / 60 s overload — fans and pumps do not need constant-torque (150%) overload capability, and paying for it wastes money and panel space.
• HVAC feature set: internal PID control, multi-pump/fan staging, skip frequencies, automatic restart, real-time clock functions.
• Embedded communications: BACnet and/or Modbus without external gateways.
• Integrated harmonic mitigation (DC choke or line reactor) and EMC filtering appropriate to the installation environment.
• Configurable failure behavior for communication loss, sensor failure and power events.
• A manufacturer with local support and spare availability for the facility's design life.

General-purpose industrial drives can serve HVAC loads, but HVAC-dedicated platforms typically package these requirements natively — which matters when you are configuring and commissioning hundreds of drives.

Motor current vs motor kW

Drives are current-limited devices, and sizing on nameplate kilowatts is a classic error. Two motors with the same rated power can draw meaningfully different currents depending on efficiency class, power factor and design voltage.

I = P / (√3 × U × η × cos φ)

A 30 kW motor at 400 V with η = 0.93 and cos φ = 0.85 draws about 55 A; an older motor with η = 0.89 and cos φ = 0.80 draws about 61 A — an 11% difference on identical nameplate power.

Drive sizing

Beyond matching FLA, data center sizing practice should account for:

1. Ambient and altitude derating (Section 8) at the installed location, not the catalog reference conditions.
2. Switching frequency: raising carrier frequency for acoustic reasons reduces available output current per the manufacturer's derating tables.
3. Future motor changes: where fan walls or pump impellers may be upgraded, confirm margin or select the next frame.
4. No blanket oversizing: routinely selecting drives two sizes up wastes capital, panel space and standby losses. Size correctly and document the basis.

IP20 vs IP55

Enclosure selection is an installation architecture decision:

IP20 units concentrated in electrical rooms simplify maintenance access and keep drives in a controlled environment, but require panel thermal design. IP55 units eliminate the panel but put the drive in the mechanical environment — acceptable when the platform is rated for it, and often the pragmatic choice for retrofit and distributed fan applications.

Ambient temperature and derating

Drive ratings are published at a reference ambient, typically 40 °C, sometimes 50 °C. Electrical rooms in data centers are usually benign, but drives mounted in plant rooms, on mezzanines above heat-producing equipment, or in rooftop enclosures can see far higher local temperatures.

• Apply the manufacturer's derating curve for ambients above the reference (commonly 1–2% output current per °C).
• Altitude derating applies above 1000 m due to reduced air density.
• Inside panels, calculate internal temperature rise from drive losses (roughly 2–3% of drive throughput power) and verify with the panel builder.

Harmonics

VFDs rectify AC to DC on their input, drawing non-sinusoidal current rich in 5th, 7th, 11th and 13th harmonics. In a data center this adds to the distortion produced by UPS rectifiers and IT power supplies, all sharing the same buses, transformers and — during utility failure — standby generators.

Consequences of excessive harmonic distortion include additional heating in transformers and cables, nuisance operation of protection, voltage distortion affecting other loads, and control instability on generator power where source impedance is several times higher than utility.

Mitigation options, in ascending order of effectiveness and cost:

IEEE 519 considerations

IEEE 519 provides the standard evaluation framework: voltage and current distortion limits assessed at the point of common coupling (PCC) — the point where the utility supplies other customers, or as defined in the project's electrical design criteria.

Three points engineers repeatedly get wrong:

1. The limits apply at the PCC, not at individual drive terminals. A drive with 35% input THDi can be perfectly acceptable if the aggregate distortion at the PCC is within limits.
2. Current limits scale with the ratio of short-circuit current to load current (Isc/IL). Stiff utility connections tolerate more distortion; weak ones tolerate less.
3. Generator operation is the critical case. On standby generators the source impedance rises sharply, so voltage distortion that was acceptable on utility can become a problem. The harmonic study must cover both operating modes.

EMC

Fast-switching drive outputs generate high-frequency emissions that can couple into control wiring, sensor circuits and communication networks. The IEC 61800-3 framework defines emission categories:

• C1: first environment (residential/commercial), unrestricted distribution.
• C2: first environment, restricted distribution — installed by professionals.
• C3: second environment (industrial) — the usual minimum for data center plant.
• C4: high-power systems, engineered installations with an EMC plan.

Specify the category matching the installation environment, and prefer drives with integrated EMC filters — retrofit external filters consume panel space and are frequently value-engineered out. Filter effectiveness also depends on installation practice, which leads to the next section.

Motor cable considerations

The cable between drive and motor is part of the EMC and reliability design:

• Length limits: every drive has maximum motor cable lengths (screened and unscreened) beyond which reflected-wave voltage peaks and capacitive charging currents become problematic. Long runs may require dV/dt filters or output reactors.
• Screening: use symmetrical screened motor cable with 360° screen termination at both ends — pigtail terminations destroy screen effectiveness at high frequency.
• Segregation: maintain separation between motor cables and control/network cabling; cross at right angles where unavoidable.
• Bearing currents: inverter-fed motors can suffer bearing damage from common-mode voltages. Inverter-duty motor insulation, shaft grounding rings or insulated bearings mitigate this on larger machines.

BMS integration

The BMS relationship defines how the drive participates in the cooling system. Three integration patterns are common:

1. Full network control — the BMS writes speed references and start/stop over the network. Maximum flexibility, but cooling becomes dependent on network integrity unless fallback behavior is configured.
2. Local PID with network supervision — the drive holds a pressure or temperature loop locally; the BMS writes setpoints and monitors. Control survives network loss naturally.
3. Hardwired control with network monitoring — run commands and references are hardwired; the network is monitoring-only. Simple and robust, less flexible.

For critical cooling, patterns 2 and 3 push autonomy toward the equipment — a principle worth applying wherever the sequence of operations allows.

A minimum monitoring point set per drive: run status, speed feedback, output current, output power, energy, DC bus voltage, drive temperature, fault code, Hand/Auto status. Standardize the point list across the facility — see the planned VFD BMS Point List.

BACnet

BACnet (ANSI/ASHRAE 135) is the dominant integration protocol for data center mechanical plant. Practical engineering points:

• Device profile: HVAC drives typically conform to the B-ASC profile. Request the PICS (Protocol Implementation Conformance Statement) during submittal review, not at commissioning.
• MS/TP vs IP: MS/TP (RS-485) remains common for drive-level networks; BACnet/IP suits larger architectures. On MS/TP, plan segment lengths, device counts and baud rates; a flooded segment shows up as sluggish control.
• COV vs polling: use Change-of-Value subscriptions where supported to reduce traffic on large drive populations.
• Object mapping: agree the object list per drive early. Retrofitting extra points across 200 drives after handover is expensive.

Communication failure

Every networked drive must have an explicit, configured, tested answer to the question: what do you do when the network goes silent?

BMS✕VFDSAFE FALLBACKCOOLING CONTINUES

Trigger

BACnet communication loss — the watchdog timer expires with no valid master traffic.

Required response

Maintain last commanded speed, or transition to a predefined fallback frequency, as selected in the drive configuration and coordinated with the cooling sequence of operations.

Engineering objective

Cooling continuity — a monitoring-network failure must never become a cooling failure.

The correct fallback depends on the application. A CRAH fan might hold last speed or ramp to a conservative fixed speed; a secondary pump running local PID simply continues its loop. The wrong answer — and a factory default on some products — is to stop.

PID sensor failure

Drives running local PID loops depend on their feedback sensor. Sensor failure (broken wire, out-of-range signal) must be detected and answered:

BMS✕VFDSAFE FALLBACKCOOLING CONTINUES

Trigger

Feedback sensor failure — the PID feedback signal falls outside its valid range (e.g. below 3.5 mA on a 4–20 mA loop).

Required response

Detect loss-of-feedback, alarm to the BMS, and transition to a predefined safe output: preset speed, last valid output, or transfer to the redundant unit per the sequence of operations.

Engineering objective

Prevent a single sensor from driving the loop to zero or maximum and destabilizing cooling.

Prefer sensor ranges that make failure detectable (4–20 mA over 0–10 V), and specify the failure response per application rather than accepting a global default.

Redundancy

Cooling redundancy is expressed at system level — N+1 CRAH units, N+1 pumps — but it is delivered at component level. The drive is part of the failure domain of the motor it feeds.

The core principle: one drive per critical motor. Each redundant motor should preferably have its own dedicated VFD to maintain independent failure domains. Architectures that route two redundant motors through shared drive hardware, transfer switches or shared control power create hidden common-mode failures that quietly defeat the redundancy the mechanical schematic promises.

BMS / COOLING CONTROLLER

HV600-A

PUMP A

HV600-B

PUMP B

N + 1

Each critical redundant motor should preferably have its own dedicated VFD to maintain independent failure domains.

Scrutinize the auxiliary systems too: control power supplies, network segments and control panels should not aggregate multiple redundant units into one failure domain.

N+1 architecture

In an N+1 group, N units carry the design load and one is surplus. For VFD-driven groups this has specific implications:

• All units should normally run at reduced speed rather than N running with one idle. Thanks to the cube law, N+1 units at partial speed consume less total power than N units at higher speed — redundancy that pays for itself in energy.
• Failure response is a speed increase: when one unit fails, the remaining units ramp up to restore capacity. The sequence of operations must define who commands this — plant controller, or autonomous per-drive logic.
• Alternation equalizes run hours across the group; drive-integrated or BMS-based alternation both work if failure cases are handled.

Bypass philosophy

A bypass contactor arrangement allows a motor to run direct-on-line if its drive fails. Whether to specify it deserves engineering thought rather than habit:

Bypass makes sense when the motor is unique (no redundant unit), full-speed operation is thermally acceptable, and the facility accepts across-the-line starting current.

Bypass is questionable when genuine N+1 drive-per-motor redundancy exists — the redundant unit already covers drive failure, and the bypass adds contactors, control complexity and new failure modes (welded contactors, accidental DOL starts into closed valves, loss of speed control precisely when the plant is stressed).

In a properly redundant data center cooling plant, money spent on bypasses is often better spent on the redundancy itself. This question is examined in depth in Should Data Center VFDs Have Bypass?

Emergency operation

Critical facilities need defined drive behavior in emergency modes:

• Emergency override input: a hardwired digital input that forces a predefined speed (often maximum) regardless of network commands — used for smoke control or maximum-cooling scenarios. It must bypass normal PID and network logic by design, not by workaround.
• Fire mode ride-through: in override mode, drives typically suppress non-critical protection trips and auto-reset critical ones, accepting drive sacrifice to maintain airflow where the life-safety strategy requires it.
• External 24 V control power: keeping the control board and communications alive from an external DC supply while mains is removed preserves BMS visibility during switching operations and speeds recovery.

Power restoration

Utility loss and transfer to generator is a routine event in a data center's life, and the mechanical plant's response is a designed behavior:

BMS✕VFDSAFE FALLBACKCOOLING CONTINUES

Trigger

Supply interruption — utility loss, transfer to generator, or retransfer to utility.

Required response

Drive rides through brief sags where energy permits; on longer interruptions it trips cleanly, then automatically restarts with a configured delay once supply returns and the run command is present. Flying-restart catches coasting fans without waiting for standstill.

Engineering objective

Automatic, orderly recovery of full cooling capacity without operator intervention.

Key configuration items: automatic restart enabled with an application-appropriate number of attempts, restart delay per the staggering plan (next section), flying restart (speed search) for high-inertia fans, and undervoltage trip levels coordinated with the transfer scheme.

Staggered restart

When power returns to dozens of drives simultaneously, simultaneous restart creates problems: inrush and charging currents, a block-load step on generators, and hydraulic/airflow transients from every machine accelerating at once.

The solution is deliberately staggered restart:

t = 0 s     Power restored
t = 2 s     CHW pumps group A restart
t = 6 s     CHW pumps group B restart
t = 10 s    CRAH group A restart
t = 14 s    CRAH group B restart
t = 18 s    Cooling tower / heat rejection fans

Implement stagger with per-drive restart delays (works even if the BMS is still booting) or plant-controller sequencing — ideally both, with the drive delays as the safety net. Verify the total restart timeline against the thermal ride-through analysis: the white space must stay within ASHRAE envelopes throughout.

Commissioning

Drive commissioning in a critical facility goes beyond making motors spin:

1. Installation verification — cable types, screen terminations, segregation, fusing/breaker coordination, panel cooling.
2. Parameter configuration — motor data, protection settings, ramps, skip bands, PID tuning, communication mapping, failure behaviors. Work from a facility-standard parameter template, not defaults.
3. Functional testing per drive — Hand/Off/Auto, local and network control, alarm reporting to BMS.
4. Documentation — electronic parameter backup for every drive, archived in the O&M system, updated after any change.
5. Point-to-point verification of every BMS point against the agreed point list.

A drive population commissioned to a documented standard template is dramatically cheaper to maintain than one where every unit is a snowflake.

Failure testing

Configuration is a claim; testing is evidence. Failure behaviors must be physically demonstrated during controls integration testing and, where cooling continuity is claimed, during integrated systems testing:

• Disconnect the network cable → verify fallback behavior and alarm.
• Disconnect PID feedback sensor → verify detection and safe response.
• Kill and restore power → verify restart delay, sequence and flying restart.
• Trip one drive in each N+1 group → verify remaining units ramp and capacity recovers.
• Activate emergency override → verify forced-speed operation ignores network commands.

Record pass/fail criteria in advance. The tests that feel inconvenient to run during commissioning are exactly the events the facility will experience in service.

Recommended specification

The requirements developed in this guide are captured as a ready-to-adapt template: the Data Center HVAC VFD Specification. Its structure:

• Ratings and construction — variable torque duty, enclosure classes, integrated DC choke and EMC filter
• Protection and reliability — dedicated drive per redundant motor, auto-restart, external 24 V control power
• Control and operation — HOA keypad, internal PID with sensor-failure handling, emergency override
• Communications — native BACnet/Modbus, minimum point list, communication-loss watchdog
• Harmonics — IEEE 519 evaluation at the PCC, both utility and generator modes
• Motor cables — length limits, screening, inverter-duty motors
• Commissioning — parameter backups, witnessed failure testing

Specify requirements, not brands: it keeps procurement competitive and makes compliance verifiable.

Example drive platform

The guidance above is vendor-neutral, and several manufacturers offer HVAC drive platforms capable of implementing it. As a concrete example: one drive platform suitable for implementing the architecture described in this guide is the Yaskawa HV600, designed specifically for HVAC fan and pump applications.

Mapped against the engineering requirements developed in this guide:

The HV600's embedded BACnet MS/TP and Modbus RTU, integrated DC choke and EMC filter, configurable communication-loss and sensor-failure behaviors, emergency override input and external 24 V control power option align with the requirement set in the recommended specification. Application detail for data center use is covered on the HV600 data center page.

Disclosure: The author works in the industrial automation industry and has professional experience with Yaskawa drive technology. Product-specific information should always be verified using the latest manufacturer documentation.

References

1. [1]

   IEEE. IEEE 519 — Standard for Harmonic Control in Electric Power Systems (external link) (2022).

2. [2]

   IEC. IEC 61800-3 — Adjustable speed electrical power drive systems: EMC requirements (external link) (2023).

3. [3]

   ASHRAE. ANSI/ASHRAE Standard 135 — BACnet: A Data Communication Protocol for Building Automation and Control Networks (external link) (2024).

4. [4]

   ASHRAE TC 9.9. Thermal Guidelines for Data Processing Environments (external link) (2021).

5. [5]

   Yaskawa America. HV600 AC Drive for HVAC Applications — Technical Documentation (external link) (2024).