Flow Modelling in the Farnsworth House:
Complexity & Simplifications
by Ted Caplow, Columbia University
Contents of this document:
The Engineering for Architecture Project at Columbia is an ongoing, interdisciplinary effort to develop a comprehensive learning and design tool for the teaching of, and ultimately for the practice of, building design. Case studies of several unique and/or well-known buildings are under development; these case studies include design descriptions, structural and stress analysis, and most recently, energy and airflow modelling. The structural modules of these case studies are well-developed (see the Building Technologies homepage). Thermo-fluids work has just begun, however, and as of March 1996 is limited to the model description of the Farnsworth House (1st case study), which follows.
The Farnsworth House was designed by Mies Van der Rohe and completed in 1951. It consists of a two rectangular concrete slabs which form the floor and roof. The walls are large glass windows punctuated by steel support columns, and the total volume enclosed measures approximately 9.5 x 28 x 55 feet. Heat is provided by a radiant coil system in the floor, with hot water circulated from a central boiler. The first owner of the Farnsworth House quarrelled publicly with Van der Rohe, claiming that it was difficult to maintain a comfortable atmosphere in the house. This controversy makes the house particularly suitable for thermal analysis.
By modeling air and heat flow through the Farnsworth House with computational fluid dynamics software, I will be seeking three (3) major pieces of information:
- The net energy consumption of the house for a variety of external weather conditions, assuming a comfortable average temperature is maintained inside.
- The internal temperature distributions corresponding to the solutions in part one above.
- The internal air-flow velocity distribution corresponding to the solutions in part one above.
Thus, once the model is constructed, parameters will be adjusted until the condition in part one is satisfied: a comfortable, steady-state, average internal temperature has been reached. Once this solution is found, parts two and three must be examined to see if thermoclines and drafts would reach uncomfortable levels.
The following are independent variables of interest:
For mass flow:
- Window aperture (how open are the two windows? are they open?)
- Door aperture
- External wind speed
- External wind direction
- Outside temperature
For convective cooling of building envelope:
- External wind speed
- External wind direction
- Outside temperature
For radiation transfer:
- Sky temperature
- Solar radiation flux
- Absorbtion, reflectivity, transmissivity, and emissivity of all surfaces
For conduction through building envelope/convection within building:
- Rate of heat supplied
- Conductivity of all surfaces
Even if it were possible to allow for all of these variables in the Phoenics (c.f.d.) model for the Farnsworth House, the resulting computations would be cumbersome in the extreme. Instead, the first model will be constructed with considerable simplifications. It is believed that the resulting flow simulation will be a more accessible learning tool while still providing reasonably realistic results. Further on in the project, we will attempt to determine the real-world accuracy of any flow solutions.
Currently, the following simplifications are incorporated into the model:
For mass flow:
- External wind is uniform and steady, but direction and velocity may be selected.
For convective cooling of building envelope:
- External surfaces are assigned a fixed temperature (no convection transfer is actually modeled OUTSIDE the house.) This temperature is an "apparent temperature", and is determined as a function of the wind speed (more wind > lower T), outside temperature, sky temperature, and material properties. Note that this temperature may be different on different surfaces.
For radiation transfer:
- At present, no radiation analysis is performed outside the building; rather, the external "apparent" wall temperature is set a little lower to make an apprximate allowance for radiative cooling of the building. (see above)
- Current model is for nightime only; solar radiation is not considered. However, it may be easily introduced as a constant heat flux, since surface temperature has little effect.
For conduction through building envelope:
- All planes are assumed to have uniform composition: glass or concrete, and a conductivity is assigned accordingly. Given the "apparent temperature" of the exterior as a boundary condition, Phoenics then solves for the heat transfer through the building envelope, which has finite thickness in the model and is contained withing the grid.
- The floor is modelled as a constant heat flux per unit area. The floor temperature is the primary independent variable; model logic works most intuitively when all other conditions are determined, and then the model is run recursively to determine the correct floor heat flux for comfortable inside temperatures under steady-state conditions.
The computational grid includes the air in the house, the walls, and the roof. The floor is a heat source with constant flux, and heat passes to the walls and ceiling by buoyancy driven convection, where it is conducted to the outside, which is modelled at a constant temperature (collapsing external radiation and convection into this one figure). Additionally, the doors and/or windows may be specified as open (or partially open), allowing for some exchange of air and forced convection. Once established, these parameters are adjusted to find a reasonable steady-state solution, and then energy use, draftiness, and thermal comfort may be evaluated from Phoenics's output graphs.
Back to Ted Caplow's Engineering Area Homepage