Airports of Thailand commissioned a consortium that included HOK as lead designer to design the new Suvarnabhumi International Airport Midfield Satellite Concourse in Bangkok. The 216,000m2, four-story concourse building will add 28 contact gates to Thailand’s main airport, which will serve more than 50 million passengers per year. The design references the existing architectural language of the main terminal. Using a similar barrel-vaulted section and alternating bands of glass and solid material, the interior of the midfield blends harmoniously with the spatial quality of the main terminal. The roof at the centre of the concourse is elevated and separated from the rest of the concourse; a reference to the stacked and layered roof forms of traditional Thai structures.
Initial conceptual design studies were first undertaken using McNeel’s Rhinoceros 5.0. These studies culminated in a wireframe model which defined the structural setting-out of the entire concourse. Due to a lack of computational skills within the project team, this was manually produced as opposed to using Grasshopper. Once the initial Rhino conceptual design studies were complete, the entire model was rebuilt from scratch using Autodesk Revit 2013. As it will be shown, this proved to be very tedious and ultimately unsuccessful, as Revit was unable to replicate the Rhino setting-out geometry, which was critical to the project.
Detailed Rhino study model
Rhino structural wireframe model
Essentially, there were three main geometric components to the project:
- The structural trusses of which there were two types, a ‘bow truss; and a ‘flat truss’;
- The overall shell massing which would form the basis of the roof, glazing and skylights; and
- The ceiling battens.
The structural trusses were generally pretty straightforward to model due to the geometric rationalisation that had already been undertaken in Rhino. The setting-out geometry was based on a series of tangential arcs in section which taper in plan. Key points were defined along these tangential arcs, which represented the ‘top of the top chord’ and the ‘bottom of the bottom chord’. This area was the structural zone with the roof build-up and ceiling situated above and below the region, respectively. This process resulted in each bow truss and flat truss being standardised with only the middle truss varied to allow the building to taper in plan. Using a Revit generic model family, a series of type parameters were defined which allowed the truss to flex and the key setting-out points to be scheduled. Special attention was paid to the graphic representation of the complex geometry to aid legibility of the design intent.
Final Design Development drawings showing setting out of the bow truss
Overall structural model showing different truss types
Overall shell massing
The concourse shell massing was initially rebuilt parametrically within a Revit conceptual mass family. Using the tangential arc profiles as defined above, these were lofted together to create various masses. There were several Revit conceptual mass families generated, including:
- Bottom chord massing
Overall bottom chord massing
Transverse elevation of conceptual mass showing parametric constraints
- Top chord massing with a series of void extrusions to subtract away various zones including window, gutters and transition zones.
Top chord massing showing window void
Longitudinal elevation of conceptual mass showing parametric constraints
- Assembly massing. This was simply a compilation of all the masses into one file.
Parametric mass with top and bottom chord massing
The assembly mass family was then loaded into the Revit project. Since the family defined the structural zone, a roof thickness needs to be applied above this zone. The ‘wall by face’ command was adopted, and initially, this proved to be successful in that Revit was able to read the segmented surfaces and produced individual walls. However, upon further investigation, it was discovered that all the surfaces needed to be tangential; otherwise, it resulted in inconsistent geometry, as shown below.
Detail of wall by face normal problem
Due to an anomaly in our setting-out geometry, the two surfaces shown weren’t tangential. Without completely changing the master geometry, this proved to be a limitation of Revit that was unable to be overcome. Therefore, as an alternative to the ‘wall by face’ command, the project team explored building the entire roof massing as a single element within a conceptual mass family. All that was required was to use the top chord massing and offset the surface. Unfortunately, while this is a relatively straight forward procedure in Rhino, the Revit family editor wasn’t able to do this automatically. In what proved to be a painstakingly slow process, the project team offset all the reference lines and created new parameters before generating the new offset surface.
However, this only fixed the first problem. The next challenge was that is was impossible to subtract the windows volumes accurately. Using the Rhino’ offset surface’ command resulted in the edge of the window being perpendicular to the normal of the top chord surface, as shown below. However, using the void forms in Revit to perform a Boolean subtraction after the roof thickness had been generated would result in the edge of the window being parallel to the ground plane. This result was unacceptable, and consequently, this method was abandoned.
Rhino massing showing window void intersecting the top of top chord surface
Eventually, after numerous unsuccessful tests in attempting to rebuild the Rhino shell geometry in Revit, it was decided to use the actual Rhino massing as the master geometry and import it directly into Revit using a *sat file via a conceptual mass family.
Imported *sat mass within Revit 2014
Of note is that when using Revit 2013, the *sat file was unable to be fully exploded, and if attempted, the imported geometry would disappear completely. However, when testing with Revit 2014, the geometry was able to be fully exploded. Since the project team was contractually tied to Revit 2013, the option of upgrading to Revit 2014 was unavailable. While exploding a *sat file within the conceptual mass family is not a pre-requisite for it to be functional, it does limit its adaptability if required.
Exploded conceptual mass within Revit 2014
The other major computational issue the team faced was documenting the ceiling. Since in plan the building was tapering, the ceiling varied from structural bay to structural bay. A method needed to be developed to express and represent these variations. To achieve this, Chameleon and Grasshopper were adopted as there were no out-of-the-box Revit commands that would be able to replicate the design intent accurately.
Revit view showing ceiling
The base surface was first created within Rhino. The surface was split at grid lines. This subdivision was done to compartmentalise troubleshooting as the script took approximately 10min to run per structural bay.
Rhino base surface for ceiling
Grasshopper was then used to divides the surface with UV divisions. Since the concourse tapered in plan, the divisions were calculated from the shortest length and based on critical dimension (batten width and min spacing, etc.). The centre lines were then offset to accommodate batten spacing..
Subdivided Rhino surface for batten setting-out
A four-point surface was then created. However, these surfaces were not planar and needed to be rebuilt. The rebuilt surfaces were then extruded to give them a thickness and a Boolean operation used to cut out the arched window and skylight geometry.
Panelised Rhino surface
A 2D pattern was then created based on a sin curve and controlled through a grasshopper graph mapper component. This curve was then arrayed, offset, extruded and a Boolean operation performed on the battens.
Typical ceiling bay
Due to the Boolean operations described above, some of the resultant battens became quite small while others had more/fewer vertices. The Grasshopper script then culled the battens that were determined to be too small and filtered the reaming battens to be either 3-points, 4-points or 5- points battens (disregarding the batten thickness).
Several adaptive components were then made within Revit to correspond with the number of vertices (3 points, 4 points, and 5 points) and loaded into the project. The thickness of the batten was controlled within the adaptive component as this reduced the number of placement points and hence made the script run faster.
The XYZ coordinates were then exported from Grasshopper via Chameleon. This export needed to be done in batches so that the appropriate adaptive component could be applied (3 points, 4 points, etc.) to the XYZ points being exported. It was discovered that the list of coordinates needed to be ‘cleaned’, as any empty lines would crash the export.
Final adaptive components ceiling in Revit
The above method was very successful, albeit a little time-consuming. In all, there were approximately 13,000 individual battens for half the concourse. Rather than creating the entire ceiling, only half was created, and then when linked into the main model, it was copied and mirrored to minimise the computational load.
This case study highlighted Revit’s inability to deal with complex geometry. While it is always preferable to have entirely native Revit elements, this isn’t always possible as was shown here. Through the adoption of other software such as Rhino and Chameleon, it is possible to extend Revit’s capabilities and achieve the desired geometric outcome. With a clear workflow, geometric integrity and rigour can be maintained. Of course, for this to be successful, the project team needs to be sufficiently skilled in the proposed software, and if not, adequate training and support needs to be provided.