The preliminary analyses were performed using simplified 3D models created by Lawrence Berkeley National Laboratory. The simplified models were developed from the printed 100% CD drawing set and AutoCAD data files provided by the Smithsonian.
The final analyses were performed with more complex geometry. The geometry for the Radiance 3D model as modeled by Interface Multimedia was provided by the Smithsonian. The FormZ drawing files were initially exported into Radiance through the Wavefront/OBJ file format and tested extensively for light leaks and consistency with the 100% CD documents. Significant differences between the 3D model and the drawings were resolved where possible, but some differences remain. For example, the 3D model contained geometry for only 1/2 of the Space Hangar. We presumed that the hangar was bilaterally symmetrical and mirrored the north side to the south. This has since been discovered to be incorrect. However, due to time and budget constraints and the fact that the implications for the lighting analysis are nominal, this was not corrected for the final renderings. On the other hand, the 3D model was discovered to contain windows along the northwest side of the Aviation Hangar. These windows were replaced with opaque wall material for the final analysis.
The approximately 200 aircraft suspended in the Aviation Hangar were presumed to have a significant impact on the overall lighting levels. To address this concern, 3D models of many aircraft were provided by the Smithsonian. A model of the SR-71 was purchased from Viewpoint Digital. Models of the vast majority of the aircraft, however, were not available. The AutoCAD drawings were then used as a template to extrude the plan outlines into "pancake" representations of actual aircraft. The thickness of these extrusions varied from 1200cm to 400cm depending upon the overall size of the aircraft. The aircraft were then "hung" in the hangar at four different heights depending upon the "level" they belong to.
Assumptions of the Model
Surface Reflectances were measured from samples provided by the Smithsonian. Total hemispherical reflectance was measured with a Minolta CM2002 spectrophotometer. The spectral color and specular reflectance at 7 degrees was also measured using this device for the Radiance model.
The final Radiance simulations were performed using the following simulation parameters:
These parameters instruct Radiance to calculate two ambient bounces of light (-ab 2) and to perform a very accurate estimate of the ambient component (-ad 1200 -as 600 -aa .175). In addition, the specular threshold was set to 1% (-st .01) to account for the small semi-specular component measured in the material samples. The sub-sampling of direct light sources was also turned on to improve the accuracy of the luminaire distributions a close proximity to other surface geometry (-ds .5). The background ambient value was empirically determined through an iterative Radiance simulation and set accordingly (-av .02 .02 .02).
For more information about these parameters, please see the Radiance Reference Manual for the rpict command.
Calculation of Energy Savings
Energy Savings were based upon the assumption that the quartz halogen emergency lighting would be required to stay on at all times. This has since been discovered to be incorrect.
Limitations of this lighting simulation are related to the input, calculation method, and analysis techniques.
The luminaires were modeled with far-field photometric data but are used in a near-field condition, especially in the cieling areas immediately above the catwalk. Near-field photometrics are not generally availble and extremely costly to measure. Using far-field photometry is a common practice in lighting simulation and is not expected to introduce significant error.
The luminaires were also simplified as equal-energy (color balanced) light sources in order to avoid problems with color balancing the final images. Because the surface colors are predominantly various tones of grey and because the vast majority of light is provided a single lamp type (the Metal Halide), this again is not expected to introduce significant error.
The Radiance Synthetic Imaging System has been used extensively throughout the building industry for architectural research and lighting analysis. It has been validated by numerous independent sources. However, errors are possible due to a few general limitations of the backward ray-tracing process. The primary limitation is the "bright surface through a narrow aperture" problem. The problem is that lighting levels will be under estimated if there is a source of significant illumination that is obscured by an opaque surface with only a small aperture to allow the light to pass through. The rays of light are followed "backward" from the eye, to the surfaces, and then to the light sources and surfaces of the model. Whereas Radiance knows where to find those surfaces which are pre-defined as "lights", it does not know a-priori which general model surfaces are likely to contain high luminance levels. This could be the case for the clerestory lites at the perimeter of the Aviation Hangar because the skydome represents a very bright diffuse source of illumination that is not identified as a light source for the Radiance calculation. This limitation was mitigated in two ways. First, the ambient calculation parameters were set very stringently such that it would be quite unlikely for a brigh surface to be missed. Secondly, the Radiance "mkillum" program--a hybrid semi-forward ray-tracer--was used to pre-compute the distribution of the clerestory lites. However, for the final simulations, the mkillum-created distributions were slowing down the calculations significantly. A preliminary test was performed without the mkillum sources and the overall illumination levels were determined to be equivalent. The final renderings of the daylit conditions proceeded accordingly.
The "raw" Radiance simulations represent a prediction of the lighting levels to be found in the modeled space. These light levels are stored as a luminance "measurement" of the intensity of light at a specific pixel location from a specific viewpoint. These luminance can vary from the very dimmest starlight to brightnesses vastly exceeding the solar disk, a dynamic range of over 1,000,000 to 1. A typical scene can often contain luminances ranging from 10,000 to 1. If it were possible to view Radiance renderings using a device which could reproduce these actual luminances or with a device which produced the corresponding luminance directly onto the human retina, then there would be a perfect one-to-one correspondence between the rendered image and one's physical perception of the displayed image.
However, today's display devices can only reproduce luminance ratios on the order of 100 to 1. The best color LCD projector can perhaps acheive 400 to 1 ratios. This is two orders of magnitude less than what is required to exactly display the physical world.
The mapping of real-world luminances to computer display brightnesses is known as the "gamut" or "tone mapping" problem. Greg Ward Larson, the author of Radiance, developed an algorith called pcond which attempts to solve the gamut problem. The research behind this algorithm is documented in his paper, A Visibility Matching Tone Reproduction Operator for High Dynamic Range Scenes. While this research is arguably the most advanced of its kind, there has yet to be any significant validation of the pcond display mapping algorithms. Human subject studies are necessary to ensure the closest correspondence between displayed images and physical perception of the represented space.
The pcond program was used to adjust the brightness of the perspective images and QuickTimeVR images that are part of this analysis. Images should be viewed on a standard computer VDT display under the viewing conditions of normal office spaces (between 50 and 150 lux on the horizontal workplane) Images will appear dark if pcond predicts that the appropriate human adaptation level (exposure) would cause such an experience of the space. Images will appear bright or even "glary" if pcond predicts that there are significant sources of glare relative to the adaptation level for the current view. The fact that the pcond algorithm has not been validated should be taken into consideration when making judgements about the lighting conditions based upon these images.
Horizontal workplane illuminance levels are also an indicator of the overall brightness of the space. As you become familiar with the relationship between illuminance levels and the perceived brightness of typical spaces, these values can be used to verify or contradict in your own mind the predicted brightnesses displayed with pcond.