Airflow Networks vs CFD

An airflow network (AFN, also called a multizone or nodal model) represents the building as a graph. Each thermal zone becomes a single node at one uniform pressure and temperature. Each opening, crack, door, window or vent becomes a link with a known resistance to flow. Outdoor points carry wind pressure coefficients (Cp) that vary with wind direction. The solver then does something conceptually simple: it finds the set of zone pressures at which the air flowing into every node exactly equals the air flowing out - mass conservation across the whole network - and reports the flow rate through each link.

Because each zone is just one point, the maths is tiny. An AFN can solve the airflow for a whole multi-storey building in a fraction of a second, and crucially it can do so for every hour of an EPW weather file - 8,760 hours - coupled to the thermal simulation. This is exactly what tools like EnergyPlus's AirflowNetwork and the standalone CONTAM (NIST) do. They answer bulk questions: annual air-change rates, whether stack and wind ventilation deliver enough fresh air, how a contaminant or smoke spreads between zones, infiltration heat loss. What they cannot tell you is anything about air movement inside a room - because inside a node, by definition, there is no 'inside'.

Computational fluid dynamics takes the opposite approach. It divides a single space (or a cluster of them, or an outdoor domain) into thousands to millions of tiny cells and solves the governing equations of fluid motion in every one, producing a full velocity field - speed and direction of air at every point - plus temperature, pressure and turbulence. Where the AFN says 'this room receives 4 air changes per hour', CFD says 'a 0.4 m/s jet enters here, hugs the ceiling, spills down the far wall, and leaves a stagnant 0.05 m/s pocket in this corner where the occupant sits'.

That resolution answers questions no network can: draught risk at a specific desk, whether a supply diffuser actually reaches the occupied zone, thermal stratification in an atrium, pedestrian wind comfort around a tower, smoke movement in a specific fire scenario. The price is steep. A single CFD run models one moment or one steady condition - one wind speed, one direction, one set of temperatures - and can take hours to set up and hours to solve. You cannot run 8,760 hours of CFD for a design study; you run a handful of carefully chosen representative cases. Tools include OpenFOAM (free, powerful, steep learning curve), Butterfly (a Ladybug Tools plug-in that drives OpenFOAM from Grasshopper), and commercial packages like Autodesk CFD or Ansys Fluent.

It is tempting to think CFD is simply 'more accurate' and AFN is a rough approximation. That is the wrong mental model. Each is accurate for the question it is built to answer and misleading for the other. An AFN gives a trustworthy annual air-change estimate in seconds; forcing CFD to answer that question would be absurdly expensive and, run for one hour only, less representative. CFD gives a trustworthy in-room velocity picture; asking an AFN for it is impossible because the information does not exist in a single-node model.

The real axis is resolution versus coverage. AFN has coarse spatial resolution (one point per room) but huge temporal coverage (every hour, whole building). CFD has fine spatial resolution (millions of points) but tiny coverage (one condition, one space). A crude rule: if your question is about quantities of air between spaces over time, use an AFN; if it is about the pattern of air within a space at a moment, use CFD. And be warned - CFD's beautiful colour images carry an authority they have not earned. A CFD result is only as good as its mesh, boundary conditions and turbulence model, all of which are easy to get wrong, and the next lesson is devoted to exactly how it misleads. Treat the two as different instruments, like a thermometer and a thermal camera - neither is 'more accurate' than the other in the abstract; each is the correct instrument for its own measurement.

It helps to picture the concrete outputs, because they reveal the gulf. Run an airflow network on a naturally-ventilated house and you get a table: for each zone and each hour, an air-change rate, a flow rate through every opening, an infiltration heat gain or loss, and - if you asked - how a tracer contaminant spreads room to room. Plot it over the year and you can say 'the living room averages 6 ACH in summer but drops below 2 ACH for 400 hours when the wind is calm'. That is a decision-grade answer about whether the strategy works, and it costs almost nothing to compute.

Run CFD on one room of that same house, at one chosen hour, and you get something entirely different: a spatial map. Every cell carries a velocity vector and a temperature, so you can see the supply jet, trace where it goes, find the 0.05 m/s stagnant pocket in the corner, and measure the air speed exactly where an occupant sits to judge draught. What you cannot get from that CFD run is the annual picture - it is one frozen moment - and what you cannot get from the network is the in-room map. The two outputs are not more and less detailed versions of the same thing; they are answers to different questions. Recognising which output your design decision actually needs - a yearly table of quantities, or a spatial map at a moment - is the fastest way to pick the right tool before you spend a minute modelling.

In practice the two are complementary, and mature workflows use both in sequence. You start with the fast, whole-building AFN coupled to your energy model: it tells you, across the year, whether natural ventilation is even viable, how many air changes each zone gets, and which hours fall short. That is a design-steering answer you get cheaply and early. Then, for the one or two critical spaces where the in-room air pattern actually matters - the naturally-ventilated classroom, the double-height lobby, the operating theatre, the plaza between towers - you invest in a handful of targeted CFD runs at the worst-case conditions the AFN flagged.

There are even coupled approaches where a CFD domain feeds boundary values back to a zonal or network model, and lighter zonal models that sit between the two (dividing a room into a few dozen cells rather than millions). But the everyday decision is simpler than the theory suggests. Ask what your design question actually is. If you cannot name the specific space and the specific moment you care about, you do not need CFD yet - you need the network model first. Reaching for CFD prematurely is one of the most common and expensive mistakes in building simulation: hours of meshing to answer a question a five-second network run would have settled. A good habit is to write the design question down in plain words first; if it contains the phrase 'across the year' or 'how much air', it is a network question, and if it names a single room and a single moment, it may be a CFD one - and only then.