Visual comfort analysis of innovative interior and exterior shading systems for commercial buildings using high resolution luminance images.
In recent years, mainstream acceptance of the "green building movement" has 1 led to the construction of a significant number of buildings with highly glazed facades, designed with the dual goals of 1) complying with green building certification programs (for example, LEED daylight and view credits) and 2) the reduction of electrical lighting energy consumption via daylighting. Casual inspection of these buildings in operation often reveals that shading systems are routinely deployed by occupants to maintain comfortable visual conditions, reducing the potential for lighting energy reduction as well as visual connection to the exterior.
This study focuses on evaluating the visual comfort conditions produced from a number of innovative interior and exterior facade shading and light-redirecting systems that have the potential to more effectively balance the goals of visual comfort with daylight transmission in comparison to conventional shading systems (for example, manually operated Venetian blinds or roller shades) used in typical sidelit perimeter zones. In this study, the term "innovative" refers to systems that employ one or more of the following strategies: subdivision of the window wall into an upper daylighting zone and a lower "view" zone, optical light redirection, translucent materials, and automation of slat angle and/or height adjustment.
Prior to the availability of luminance maps to assess lighting conditions in real spaces, the limitations of available photometric monitoring equipment (for example, shielded illuminance sensors, spot luminance meters) resulted in simplified methods of glare assessment in daylit spaces. Average luminances of large window areas using shielded sensors have commonly been used to compute the daylight glare index (DGI) [IEA SHC Task 21, 2000] as well as numerous other discomfort glare indices. In addition, average luminances from shielded sensors have been correlated to subjective response data in the derivation of a predictive model that relates the average luminance of the entire window to the probability of the shade being lowered [Clear and others, 2006].
The primary limitation of these methods is that the boundary of the glare source must be pre-defined. This is a problem in daylit spaces because the boundary and position of glare sources are constantly changing. An additional limitation is that a physical shield must be fabricated with a profile identical to the desired target. This presents a challenge when defining targets with irregular profiles, such as the boundary between the landscape and the sky as viewed through the window. A final limitation is that each pre-defined region requires an individual illuminance sensor, restricting the number of glare sources that can be defined for practical reasons, often leading to the entire window being defined as a single (and very large) glare source. Because the DGI is sensitive to both the size and position of each glare source, the definition of the entire window as a single source can lead to misleading results.
HDR images store luminance data on a "per-pixel" scale, providing far greater resolution for visual comfort analysis, enabling both the definition and analysis of an arbitrary number of pre-defined regions, and the possibility of detecting glare sources throughout the full field of view (FOV) to compute any number of discomfort glare metrics. In computer graphics, the original HDR format (Radiance RGBE) was developed for the lighting simulation engine Radiance in order to record the photometric conditions of synthetic lighting environments [Ward 1991, 1994]. In following years, techniques were developed to produce HDR images from real-world scenes [Debevec 1997, Mitsunaga & Nayar 1999] by compositing multiple, exposure-bracketed, low dynamic range (LDR) images (for example, JPEG) into a single HDR image. Motivated by the possibility of using HDR in real spaces for photometric analysis, methods are now available to produce calibrated HDR images, commonly referred to as luminance maps [Inanici & Galvin, 2004]. Wienold and Christoffersen  derived a new daylight glare prediction model based on HDR data correlated to subjective response.
In this study, two methods of visual comfort analysis using luminance maps are demonstrated on a variety of interior and exterior shading systems, some with reflective surfaces, some with automated controls, and some where the window wall is subdivided into an upper and lower region. These systems have been designed to more effectively balance the competing goals of admitting useful daylight without the associated glare. The Illuminating Engineering Society of North America (IESNA) recommended contrast ratio limits and Hopkinson Cornell Large Source Glare Index (DGI) (1) are used as a basis for evaluation. The analysis focuses on addressing two questions: Do existing metrics and summary indices computed from luminance maps enable differentiation between systems? And, what value can time-lapse imaging add over more conventional methods of visual comfort assessment?
[FIGURE 1 OMITTED]
2 MEASUREMENTS AND PROCEDURES
2.1 TEST FACILITY
Experimental tests were conducted in the 88.4 [m.sup.2] (952 [ft.sup.2]) Advanced Windows Testbed Facility located at the Lawrence Berkeley National Laboratory (LBNL) in Berkeley, California (latitude 37[degrees]4'N, longitude 122[degrees]1'W). The facility consists of three identical side-by-side test rooms (Fig. 1) built to represent a commercial office environment and were unoccupied during this study. Testing was conducted with electric lighting on and dimmed to maintain a setpoint illuminance of 538 lx at the workplane.
2.2 ACQUISITION OF HDR IMAGES
Images were taken vertically at two locations within each room assuming a seated occupant (1.2 m (4 ft) eye height): a) looking at the west sidewall and computer monitor or visual display terminal (VDT) 1.52 m (5 ft) from the window and b) normal and centered on the window, 1.2 m (4 ft) from the window (Fig. 1). Images were acquired from 6 AM to 6 PM Standard Time and therefore some data include nondaylit conditions.
Each HDR image was composited from a set of nine, 1536 x 1536 pixel, LDR images, exposure-bracketed in 1 EV steps using conventional CCD cameras (Nikon 990) with an equidistant fisheye lens (Nikon fc e8, 1831. Compositing was achieved using the program hdrgen, which converts the RGB pixel values of exposure bracketed LDR images into real-world luminances [Mitsunaga and Nayar 1999]. Automation of acquisition and exposure bracketing was achieved using a series of scripts written for this application. A more general script (HDRcapOSX) was written later for the higher quality Canon EOS 5D digital camera series for a related project, which was then ported and tested on the Mac OSX 10.5.8 operating system by [Fan and Mardaljevic 2009]. This script is backwards compatible to OSX 10.4.x and available free at: (http://www.iesd.dmu.ac.uk/~m/doku.php?id=resources:hdrcaposx).
[FIGURE 2 OMITTED]
2.3 POST-PROCESSING OF HDR IMAGES
To correct for the light fall off typical of wide-angle camera systems, a "digital filter" was applied to each HDR image as a post-process in Radiance using the method developed by Inanici and Galvin . During the time required to acquire a bracketed set of images (~1 minute), the luminous environment is rarely stable and large swings in outdoor sky conditions can introduce significant measurement errors. To control for unstable lighting conditions, the global vertical illuminance computed from the image was compared to the average of two readings from an adjacent vertical illuminance sensor (Licor LI-210, cosinecorrected with a nominal accuracy of 3 percent and a range of 0-15,000 lx): immediately before and after acquisition of the bracketed set of images (Fig. 2). The image was then uniformly scaled by adjusting the exposure line of the image file header so that the global vertical illuminance computed from the image matched the average of the two illuminance readings.
2.4 MEASUREMENT ERRORS
To quantify errors associated with changes in vertical illuminance during acquisition, a test was performed to examine the accuracy of luminance maps under dynamic sky conditions. The camera was positioned adjacent to a shielded luminance sensor. A mask was defined for the luminance maps that matched the FOV of the shielded sensor (Fig. 3). The average luminance computed from the luminance map was then compared to the shielded sensor and reported in terms of percent error and magnitude error. Figure 4 presents data for one of the six cameras tested. Each of the six cameras was tested for one day resulting in a total of -144 images per camera (867 images total). During this test, the average luminance of the masked region ranged from 0 to 11,990 cd/[m.sup.2] (average of all observations=2997 cd/[m.sup.2], std.dev=3011cd/[m.sup.2]), and the global vertical illuminance at the lens ranged from 0 to 11,900 lx.
Results (presented in Fig. 5 and Table 1) show that measurement accuracy was affected by both the magnitude of the change in global vertical illuminance during acquisition (Fig. 5) and when the global vertical illuminance at the lens was very low (< 50 lx). When the global vertical illuminance at the lens was above 50 lx, the average error of the six cameras combined was 0.21 percent, with a standard deviation of 7.6 percent. Within this set of data, accuracy was shown to decrease when the change in global vertical illuminance during acquisition exceeded 5 percent (avg. error= 5.46 percent std. error=13.59 percent), compared to when the change was less than 5 percent (avg. error= -0.21 percent std. error= 5.79 percent). When the global vertical illuminance at the lens was less than 50 lx (N = 67 images), the percent error of all cameras was significantly larger, however the magnitude error was not significant (avg. magnitude error=8 cd/[m.sup.2], max magnitude error=20 cd/[m.sup.2]). Because this study focused on glare analysis, the small magnitude errors introduced by images acquired at global vertical illuminance levels below 50 lx were not considered a problem, and therefore images acquired under these conditions were retained.
[FIGURE 3 OMITTED]
2.5 TEST AND REFERENCE CONDITIONS
Two, six month, solstice-to-solstice phases of field testing were defined. Phase I evaluated interior shading systems (Fig. 6) and Phase II evaluated exterior shading systems (Fig. 7). For Phase I, the period of measurement was from December 21, 2007 to June 21, 2008. For Phase II, the period of measurement was from June 21, 2008 to June 21, 2009. To serve as a simple benchmark to judge test conditions against, two reference conditions were defined: a "manuallyadjusted" interior Venetian blind or fabric roller shade. A description the characteristics of each system related to interior daylight distribution is given in Appendix Table 1, and operation and material characteristics are provided in Appendix Tables 2 and 3. A detailed description of hardware and operation is given in [Lee and others, 2009].
2.6 CLASSIFICATION OF SKY CONDITIONS
Sky conditions were expected to have a significant influence on visual comfort outcomes Therefore, a sky classification for each test day was determined using the method developed by [Darula and Kittler, 2004]. The classification is based on two parameters; the fraction of time, s, that the direct normal irradiation (measured using a pyroheliometer) exceeds 120 w/[m.sup.2], and U, the natural logarithm of the average of the absolute values of the change in global horizontal illuminance (lux) over a one-minute span. If s is less than 0.03, then conditions are considered to be overcast. If s is greater than or equal to 0.75, conditions are characterized as clear. For solar fractions between this range, if U is less than (10 - 6s), conditions are considered cloudy, otherwise they are considered dynamic. The resulting number of test days and sky conditions are summarized in Table 2.
TABLE 1. Measurement Errors During Unstable Sky Conditions Global Camera (All Vertical Name Data) Illuminance During Acquisition N Avg % SD % Avg Magnitude (Images) Error Error Error (cd/[m.sup.2]) >50lux Av 130 -4.12% 5.39% - 59.2 Aw 132 3.29% 7.36% 11.7 Bv 133 -0.61% 7.22% -76.7 Bw 133 3.98% 6.63% 71.6 Cv 136 -2.92% 6.72% - 87 Cw 136 1.63% 8.64% - 46.4 All 800 0.21% 6.99% -31 cameras <50lux Av 15 168.62% 205.40% 5.33 Aw 14 266.54% 317.61% 9.14 Bv 10 223.05% 218.13% 8.5 Bw 10 180.59% 140.61% 11 Cv 10 162.58% 171.59% 6.9 Cw 8 127.30% 83.82% 10.5 All 67 188.11% 189.53% 8.56 cameras Global Camera Change in Vertical Name Global Illuminance Vertical During Illuminance Acquisition During Acquisition (<5%) SD N (Images) Avg % SD % Magnitude Error Error Error >50lux Av 47.7 119 - 4.72% 4.14% Aw 34.2 125 2.63% 5.50% Bv 303 118 - 6.23% 1.23% Bw 273 114 3.26% 4.47% Cv 231 118 - 5.22% 3.05% Cw 807 121 1.28% 8.59% All 283 715 -0.21% 5.79% cameras <50lux Av 3.39 Aw 5.91 Bv 2.92 Bw 3.92 Cv 5.22 Cw 2.56 All 3.99 cameras Global Camera Change in Vertical Name Global Illuminance Vertical During Illuminance Acquisition During Acquisition (>5%) Avg Magnitude SD N (Images) Avg % Error Magnitude Error (cd/[m.sup.2]) Error >50lux Av - 59.8 48.2 11 - 3.89% Aw 12 34.9 7 15.11% Bv -77.8 294 10 7.14% Bw 73.4 214 11 7.31% Cv - 82.1 231 13 0.02% Cw -53.9 854 10 7.04% All -31.4 279 62 5.16% cameras <50lux Av Aw Bv Bw Cv Cw All cameras Global Camera Vertical Name Illuminance During Acquisition SD % Avg Magnitude SD Error Error Magnitude (cd/[m.sup.2]) Error >50lux Av 10.58% - 52.8 42.52 Aw 19.58% 6 19.18 Bv 12.92% - 86.8 355.3 Bw 14.65% 5.36 528.1 Cv 14.29% -57.5 96.15 Cw 9.50% 35.5 85.97 All 13.59% 25 187.9 cameras <50lux Av Aw Bv Bw Cv Cw All cameras
[FIGURE 4 OMITTED]
[FIGURE 5 OMITTED]
[FGIURE 6 OMITTED]
3 IESNA REGION-BASED CONTRAST RATIO LIMITS
The IESNA has developed guidelines for acceptable levels of nonuniformity expressed in terms of luminance ratios between surfaces in the FOV. The basis for the guideline is the physiological phenomenon called transient adaptation, a phenomenon associated with reduced visibility after viewing a higher or lower luminance than that of the task [IESNA 2005]. To avoid visual performance decrements, a maximum allowable contrast of [1:3:10] is permitted between the task, surroundings, and background.
[FIGURE 7 OMITTED]
[FIGURE 8 OMITTED]
Luminance maps are particularly well suited to the analysis of luminance ratios, as any number of arbitrarily shaped regions can be defined during analysis. In this study, regions were defined by creating a unique bitmap mask (Fig. 8) and a process was implemented using Radiance to compute the average luminance for each of the masked regions. The visual task was defined as a visual display terminal (VDT) with a constant luminance of 200 cd/[m.sup.2], leading to discomfort thresholds of 600 cd/[m.sup.2] for "surrounding" regions (that is, 1:3), and 2000 cd/[m.sup.2] for "background" regions (that is, 1:10).
TABLE 2. Summary of Phase I and Phase II Sky Conditions Phase I Orientation Number of Days Test Condition N Clear Cloudy Dynamic Overcast Auto-split-mir-VB VDT 42 27 5 7 3 Window 45 30 5 7 3 Auto-VB VDT 16 10 1 4 1 Window 15 9 1 4 1 Split-opt-VB VDT 35 21 3 9 2 Window 35 21 3 9 2 Split-VB VDT 29 22 0 7 0 Window 31 21 0 7 3 Diffiuse-VB VDT 26 19 1 4 2 Window 26 19 1 4 2 Auto-RS VDT 26 12 3 4 7 Window 26 13 3 4 6 Average Phase II Test Condition Orientation N Clear Cloudy Dynamic Overcast VB-E1n VDT 45 30 2 7 6 Window 44 29 2 7 6 VB-E1n-auton1 VDT 76 31 9 22 14 Window 72 30 9 20 13 VB-E2n VDT 33 22 2 8 1 Window 32 21 2 8 1 VB-E2n-auton1 VDT 24 14 3 4 3 Window 24 14 3 4 3 RS-E-auton1 (vs. VDT 49 28 5 13 3 ref) Window 47 26 5 13 3 VB-E3opt VDT 45 25 6 10 4 Window 45 25 6 10 4 RS-1-autol1 VDT 56 32 5 14 5 Window 47 29 3 13 2 Average Phase I Orientation Fraction of All Sky Conditions Test Condition Clear Cloudy Dynamic Auto-split-mir-VB VDT 0.64 0.12 0.17 Window 0.62 0.13 0.18 Auto-VB VDT 0.62 0.06 0.25 Window 0.60 0.07 0.27 Split-opt-VB VDT 0.60 0.09 0.26 Window 0.60 0.09 0.26 Split-VB VDT 0.76 0.00 0.24 Window 0.68 0.00 0.23 Diffiuse-VB VDT 0.73 0.04 0.15 Window 0.73 0.04 0.15 Auto-RS VDT 0.46 0.12 0.15 Window 0.50 0.12 0.15 0.63 0.07 0.21 Phase II Test Condition Orientation Clear Cloudy Dynamic VB-E1n VDT 0.67 0.04 0.16 Window 0.66 0.05 0.16 VB-E1n-auton1 VDT 0.41 0.12 0.29 Window 0.42 0.12 0.28 VB-E2n VDT 0.67 0.06 0.24 Window 0.66 0.06 0.25 VB-E2n-auton1 VDT 0.58 0.12 0.17 Window 0.58 0.12 0.17 RS-E-auton1 (vs. VDT 0.57 0.1 0.27 ref) Window 0.55 0.11 0.28 VB-E3opt VDT 0.56 0.13 0.22 Window 0.56 0.13 0.22 RS-1-autol1 VDT 0.57 0.09 0.25 Window 0.62 0.06 0.28 0.58 0.09 0.23 Phase I Orientation Test Condition Overcast Auto-split-mir-VB VDT 0.07 Window 0.08 Auto-VB VDT 0.06 Window 0.07 Split-opt-VB VDT 0.06 Window 0.06 Split-VB VDT 0.00 Window 0.10 Diffiuse-VB VDT 0.08 Window 0.08 Auto-RS VDT 0.27 Window 0.23 0.10 Phase II Test Condition Orientation Overcast VB-E1n VDT 0.13 Window 0.14 VB-E1n-auton1 VDT 0.18 Window 0.18 VB-E2n VDT 0.03 Window 0.03 VB-E2n-auton1 VDT 0.12 Window 0.12 RS-E-auton1 (vs. VDT 0.06 ref) Window 0.06 VB-E3opt VDT 0.09 Window 0.09 RS-1-autol1 VDT 0.09 Window 0.04 0.10 N varies for VDT and WIN views as a result of a varying errors in HDR acquisition between the two acquisition systems.
[FIGURE 9 OMITTED]
3.1 ANALYSIS OF REGION-LUMINANCE DATA
The analysis of region luminance data focused on differentiating between shading systems based on the frequency and degree of threshold exceedance produced by each pre-defined region. This method was used to evaluate 12 regions for each of 14 test conditions under four different sky conditions to quickly identify the regions and sky conditions that produced the greatest discomfort conditions.
As an example of this method (using Phase I data for the Window-facing orientation), Fig. 9 illustrates the frequency and degree of threshold exceedance resulting from exceeding the IESNA recommended contrast ratio limits for the window regions under clear sky conditions. For each window region in Fig. 9, there are two accompanying figures immediately below. The bottom figure indicates the percent of time (6 AM to 6 PM Standard Time) that the luminance threshold (2000 cd/[m.sup.2]) was exceeded under all clear sky conditions. Data for all Phase I shading systems, including the reference conditions (indicated in brackets), are shown on the same plot (identified by a letter code).
As an indicator of the magnitude of threshold exceedance, the middle figure summarizes the distribution of values observed above the luminance threshold. The data are ordered and the first (Q1), second (Median) and third (Q3) quartiles are identified by horizontal lines. The median is indicated by a bold line within the box that is drawn around the inner quartile range (IQR). The whiskers extend to the smallest observation (that is, 2001 cd/[m.sup.2]) and largest observation (that is, the maximum luminance recorded for that region).
This method enabled the identification of general trends in relative performance between systems, for example, the automated Venetian blind systems (A, B) controlled window luminance to within desired limits better than the static (that is, seasonally adjusted) blind systems for all four regions (Fig. 9). In addition, the method enabled the differentiation between systems based on the magnitude of "failure" conditions. As an example, the upper window region of the split-opt-VB, split-VB, and diffuse-VB test conditions resulted in significantly greater luminance levels when the luminance threshold was exceeded compared with other systems.
Finally, this method enabled the performance of individual systems to be assessed in greater detail spatially. For example, the performance of the split-opt-VB was better than the split-VB for the upper and lower window regions, but significantly worse for the mid and lower sky regions. The substantial variation in performance between regions (both within and across systems) illustrates the need for detailed analysis of innovative facade systems when the system is divided into different functional roles.
4 THE DAYLIGHT GLARE INDEX
4.1 DISCOMFORT GLARE
The Hopkinson-Cornell large-source glare index (daylight glare index or DGI) is a metric commonly used to evaluate discomfort glare for large-area sources of glare such as windows [Hopkinson and Bradley, 1960; IES 1962]. DGI is computed from a formula that was developed to account for the effects of source brightness, size, location relative to line of sight, and adaptation luminance. Table 3 lists the mean subjective response to values from the DGI formula as measured in environments with windows.
[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII] (1)
[L.sub.s] Source luminance (cd/[m.sup.2])
[L.sub.b] Background luminance (cd/[m.sup.2])
[pi] Solid angular subtense of source modified for the effect of the observer in relation to the source (sr)
w Solid angular subtense of source at the eye of the observer (sr)
4.2 DETECTION OF ARBITRARILY LOCATED GLARE SOURCES
Although a luminance map provides a highly detailed representation of the visual environment, there remains the challenge of determining what regions in the image constitute a glare source in order to compute glare discomfort metrics such as the DGI. Addressing this issue is presently dealt with by specifying a threshold value for glare: either an absolute value (for example, all sources above 1000 cd/[m.sup.2]) or a ratio between a given pixel and the average luminance of the task or entire scene (for example, 4:1). There is currently limited guidance for what this threshold value should be, or what method to use in order to represent the glare sensitivity of a human observer. In the development of evalglare, a tool to predict the probability of discomfort glare (DGP) from hemispherical luminance maps, Wienold and Christoffersen  defined glare as any pixel four times greater in luminance than the average luminance of a circular visual task zone with an angle of ~0.53 sr. However, Wienold and Christoffersen did not indicate how the ratio of 4:1 was arrived at in favor of other possible ratios. Evalglare allows the user to select from multiple methods of glare detection (for example, scene-based, task-based, and absolute) as well as specify the threshold. If the task-based method is chosen, the tool enables the user to define a circular task zone by specifying the centroid and solid angle.
TABLE 3. Subjective Correlation to DGI Glare Criterion DGI Just perceptible 16 Just acceptable 20 Just uncomfortable 24 Just intolerable 28
In this study, glare sources were detected using the Radiance program findglare [Ward 1992]. The approach used by findglare to identify glare sources is called "thresholding," where the image is divided orthogonally into equal samples and if a particular sample is above a designated threshold value, then findglare assumes it must be part of a glare source. In this study a threshold value of 1000 cd/[m.sup.2] was used to detect glare sources. This value corresponds to a ratio of 5:1 relative to the task defined in this study (a VDT of average luminance=200 cd/[m.sup.2]). When a sample above the threshold value is found, it is merged with neighboring contiguous glare samples. Two glare samples are considered contiguous if they are separated by at most one nonglare sample. This allowed separation is to avoid the breakup of something like a window with Venetian blinds into an unreasonable number of sources. A glare source may have any number of holes in it and still be considered contiguous. The sample parameter was specified to generate a sample size of 2 pixels by 2 pixels, enabling the differentiation of glare sources located relatively close together. The output of findglare is the centroid, solid angle, and average luminance of each glare source identified. In this study, these values were then processed by another Radiance program glarendx to compute the DGI.
4.3 WEIGHTING OF TIME-SERIES DGI DATA
Because the DGI in a highly daylit space can cover a wide range of subjective values over the course of a day (from imperceptible up to uncomfortable or even intolerable glare), simple averaging of all observed values can easily result in a summary value of "imperceptible" glare and therefore lead to misleading conclusions. There is no agreed upon procedure for handling this type of problem.
To address this problem, a weighted daylight glare index ([DGI.sub.w]) sensitive to infrequent periods of perceptible-to-uncomfortable glare was implemented [Carmody 2004]. The weighting formula is relatively sensitive to extreme values of discomfort glare yet capable of discriminating between cases with single large excursions and cases with multiple but just slightly smaller excursions from the base value. The returned [DGI.sub.w] value gives an average that is sensitive to excursions from a base value if they occur at a frequency of 1 percent or more and has a scale that is the same as that defined for an individual DGI value (for example, 20="just noticeable", 22="just uncomfortable" etc.). We therefore constrained the minimum and maximum values of DGI and calculated a [DGI.sub.w] from the constrained values for each day using the equation below. An example of the weighting method applied to 144 observations of the DGI is shown in Fig.
[FIGURE 10 OMITTED]
If DGI < 6, then set DGI to 6, or
if DGI > 32, then set DGI to 32,
else, [DGI.sub.w] = 4*[log.sub.10] [mean([10.sup.(DGI/4)])] (2)
where, DGI = Measured 5-min DGI data over the 12-h period from 6 AM to 6 PM.
4.4 ANALYSIS OF [DGI.sub.w] DATA
The [DGI.sub.w] analysis focused on two goals 1) differentiating system performance across seasonal variation in sun and sky conditions and 2) providing an "overall" summary of visual comfort performance for the solstice-to-solstice interval. Although the [DGI.sub.w] analysis indicated discomfort conditions under nonclear sky conditions, the unequal number of sky conditions recorded for each system combined with the uneven distribution of nonclear conditions throughout the test intervals resulted in the analysis focusing on clear sky data.
Using the auto-split-mir-VB as an example, Fig. 11 illustrates both the daily and seasonal variability of time-series DGI data that was recorded for clear sky conditions during Phase I testing. For each day (indicated on the x-axis in MMDD format) the 5-minute time-series DGI data is plotted on the y-axis and color-coded by magnitude of discomfort (for example, yellow corresponds to "just uncomfortable"). The green lines indicate the times when seasonal slat adjustments were made to the ref-VB. This graphic technique enables one to visualize the time and magnitude of visual discomfort for each test condition and provided feedback to improve the operation of dynamic systems by indicating specific times for more detailed diagnostics.
As an example, the periods of discomfort indicated around 9AM on March 18 for the auto-split-mir-VB (Fig. 11) were investigated in greater detail by compiling the entire day of images into a time-lapse movie using a falsecolor tonemapping (Fig. 12). This technique enabled us to directly visualize the scenes that resulted in glare discomfort throughout the day and to identify the causes. In the above example (Fig. 12), the glare discomfort was a result of direct sun on the sidewall and an unshaded view of bright sky. These conditions existed while the direct solar irradiance on the facade was insufficient to trigger the deployment of the blind due to position of the sun at an oblique angle to the facade, but sufficient to cause discomfort for an unacceptable period of time. This information was later used to improve the control algorithm used to control this system.
[FIGURE 11 OMITTED]
Figure 13 illustrates the seasonal distribution of [DGI.sub.w] values for the auto-split-mir-VB and ref-VB pairing. In this figure, [DGI.sub.w] values for nonclear sky conditions are included for illustrative purposes. The increasing trend in [DGI.sub.w] values from winter to summer reflects the increasing periods of visual discomfort in the morning and evening (indicated in yellow in Fig. 9) that occurred as the system remained retracted until the exterior vertical illuminance threshold was exceeded and the blind deployed.
Figure 14 summarizes the [DGI.sub.w] values for Phase I and II shading systems for the window-facing orientation under clear sky conditions. The summary was created by taking the average of the [DGI.sub.w] values over the solstice-to-solstice interval for each test condition, using an equal number of test days from each interval of seasonal slat-angle adjustment (for example, "winter, "equinox," "summer") to not bias the results toward a system that performed better or worse during one particular time of year. The standard error is provided as an indicator of the variability in the observed [DGI.sub.w] values. Data for this figure and letter-code for each system are given in Table 4. All results include equal groups of test days from each third of the solstice-to-solstice test interval except for test condition "B" (auto-VB), which due to operational issues, resulted in 2 test days in the first third, no days in the middle, and 7 days in the final third. Because glare conditions were generally more severe at lower sun angles, it is expected that an equal sampling of test days would result in a greater summary [DGI.sub.w] score for this test condition.
[FIGURE 12 OMITTED]
This article illustrates two methods to achieve summary indicators of visual comfort based on performance over solstice-to-solstice conditions using existing metrics. The analysis focused on addressing two questions:
1. Do Existing Metrics And Summary Indices Computed From Luminance Maps Enable Differentiation Between Systems?
[FIGURE 13 OMITTED] Both methods enabled general trends in performance between groups of shading systems to be identified and these trends were generally consistent between the two methods. For example, of all Venetian blind systems tested, the interior automated shading systems (auto-split-mir-VB, auto-VB) resulted in the lowest summary [DGI.sub.w] scores (Fig. 14) and the least luminance threshold exceedance for the window regions (Fig. 9). In addition, both methods indicated that the ref-VB performed the worst of all systems tested, with the highest [DGI.sub.w] score (Fig. 14) and in general the greatest periods of exceedance for most pre-defined regions (Fig. 9). However, the methods were not consistent in assessing the performance of the static interior subdivided systems (split-opt-VB, split-VB), where the split-VB resulted in a lower summary [DGI.sub.w] score than the split-opt-VB, yet exceeded the luminance threshold for a greater period of time for the upper and lower window regions.
[FIGURE 14 OMITTED]
2. What Value Can Time-Lapse Imaging Add Over More Conventional Methods Of Visual Comfort Assessment?
TABLE 4. Summary of DGIW Data Window-Facing Camera Test Condition Code N Mean. test SE. Mean. SE. N test ref ref Auto-split-mir-VB A 24 19.5 0.41 25.6 0.12 21 Auto-VB B 9 18.5 0.53 24.6 0.34 10 Split-opt-VB C 18 23.4 0.32 25.1 0.20 18 Split-VB D 12 21.2 1.11 25.5 0.26 12 Diffuse-VB E 12 24.6 0.40 25.2 0.21 12 VB-Eln G 24 23.7 0.16 25.1 0.17 21 VB-E1n-auton1 H 12 24.0 0.19 25.0 0.15 12 VB-E2n I 15 23.8 0.20 25.3 0.23 18 VB-E2n-auton1 J 9 23.7 0.32 25.1 0.26 9 VB-E3opt K 18 23.2 0.33 25.2 0.17 18 RS-l-autoll F 30 21.8 0.22 18.0 0.52 33 RS-E-auton1 L 18 22.3 0.14 17.9 0.73 18 VDT-Facing Camera Test Condition Code Mean. test SE. Mean. SE. test ref ref Auto-split-mir-VB A 11.9 0.27 13.5 0.36 Auto-VB B 7.2 0.19 12.3 0.50 Split-opt-VB C 12.2 0.32 13.2 0.40 Split-VB D 10.3 0.52 13.5 0.57 Diffuse-VB E 13.8 0.42 13.2 0.47 VB-Eln G 11.9 0.27 13.5 0.36 VB-E1n-auton1 H 11.6 0.27 13.4 0.29 VB-E2n I 12.8 0.36 13.6 0.40 VB-E2n-auton1 J 12.6 0.74 13.8 0.65 VB-E3opt K 12.3 0.33 13.5 0.31 RS-l-autoll F 13.2 0.27 10.2 0.43 RS-E-auton1 L 13.7 0.89 10.0 0.62
Two benefits were found in this study. The first benefit, demonstrated in Section 3, is the ability to subdivide regions of the image into an arbitrary number of zones of any arbitrary shape. This capability enabled us to differentiate performance between systems with greater spatial detail. For example, the performance of the split-opt-VB was better than the split-VB for the upper and lower window regions, but significantly worse for the mid and lower sky regions (Fig. 9). The substantial variation in performance between regions (both within and across systems) illustrates the need for detailed analysis of innovative facade systems when the system is divided into different functional roles.
The second and perhaps more significant benefit is that an HDR image can be used to directly visualize the magnitude and distribution of light throughout the day. Used primarily as a diagnostic technique in this study, daily sets of HDR images compiled into time-lapse movies enabled the performance of dynamic shading and light redirecting systems to be evaluated and improved by visually identifying the conditions that caused discomfort (Fig. 12).
Six-month, solstice-to-solstice experimental tests were conducted of both conventional and innovative interior and exterior shading systems on south-facing, large-area windows at a full-scale daylight test facility, located in a predominantly sunny climate in Berkeley, California. These systems used a variety of strategies, including optics, subdivision of the window wall into a lower view and upper daylighting zone, and automated controls, with the goal of achieving a more balanced and comfortable luminous environment.
The objective of this study was to explore how calibrated high dynamic range (HDR) images (luminance maps) acquired in real world daylit environments can be used to characterize, evaluate, and compare visual comfort conditions of innovative facade shading and light-redirecting systems. Detailed (1536 x 1536 pixel) luminance maps were time-lapse acquired at 5-min intervals over a 12-h period from two view positions in an unoccupied full scale testbed facility. These luminance maps were analyzed using a weighted Hopkinson Cornell Daylight Glare Index (DGI) and the IESNA recommended contrast ratio limits to quantify how innovative interior and exterior shading systems compared to conventional systems under real sun and sky conditions over a solstice-to-solstice test interval. Glare sources for the DGI calculation were detected using the Radiance findglare tool assuming a threshold value of 1000 cd/[m.sup.2]. Contrast ratios were based on the luminance of user-defined masked regions of the luminance map, assuming a computer-based task with a visual display terminal (VDT) average luminance of 200 cd/[m.sup.2].
Based on the results of this study, the following conclusions were made:
1. Both the weighted DGI and the IESNA recommended contrast ratio limits enabled general trends in performance between groups of shading systems to be identified and these trends were generally consistent between the two indicators.
2. Compared to existing methods of analysis that define the entire window as a single glare source, luminance maps indicate that the magnitude and duration of visual discomfort varied significantly for many systems across the "masked" regions defined for the window and illustrate the need for more detailed spatial analysis when assessing the visual comfort performance of innovative facade systems.
3. Daily sets of luminance maps compiled into movies with a false-color tone-mapping can provide a significant advantage over conventional visual discomfort indicators when used as a diagnostic technique to visualize, assess, and improve the performance of control algorithms for automated shading and light redirecting systems.
Luminance maps, corrected for lens light fall off (vignetting) and calibrated against an adjacent illuminance sensor, are reasonably accurate on average under dynamic sky conditions when the change in global vertical illuminance during acquisition is below 5 percent (avg. % error of the six cameras= -0.21 percent, standard deviation= 5.79 percent, N = 715 images). However, luminance data becomes significantly less accurate when the change in global vertical illuminance at the lens during acquisition exceeds 5 percent (avg. % error of the six cameras= 5.46 percent, standard deviation 13.59 percent, N = 62 images).
We would like to thank our LBNL colleagues, Daniel Fuller, Dennis DiBartolomeo, and Chad Howdey Goudey for their assistance on this project. This work was supported by the Assistant Secretary for Energy Efficiency and Renewable Energy, Building Technologies Program, of the U.S. Department of Energy under Contract No. DE-ACO2-05CH11231 and the California Energy Commission through its Public Interest Energy Research (PIER) Program on behalf of the citizens of California.
APPENDIX TABLE 1. Summary of Phase I and Phase II Test and Reference Conditions Phase I: Interior Code Daylighting/Control Test Conditions Strategy Automated, split Auto-split-mir-VB Scheduled slat optical Venetian adjustment to block blind direct sun and redirect daylight to the ceiling and upper wall regions. Retracted when exterior vertical illuminance decreased below fixed value. Automated, Auto-VB Automated slat matte-white venetian adjustments to blind regulate daylight transmission in response to workplane illuminance and exterior solar conditions. Manually-operated, Split-opt-VB Reflective slats optical venetian with a prismatic blind function to redirect daylight onto the ceiling and upper wall regions. Manually-operated, Split-VB Semi-gloss white split venetian coating on top blind surface of slats to diffusely reflect incident daylight. Lower surface is brushed silver with low-e. Automated, Auto-RS Automated height fiberglass/PVC adjustments to limit fabric roller shade daylight penetration to a depth of 0.91 m (3 ft) from the interior face of glazing. Translucent Diffuse-VB Translucent panel diffusing panel and (VLT ~30%) in upper manually-operated window region. Lower venetian blind section identical to the reference-VB. Phase II: Exterior Test Conditions Manually-operated, VB-Eln 2" slats with full height venetian semi-gloss white blind coating adjusted seasonally to block direct sun and diffusely reflect daylight. Automated, VB-Eln-autonl Identical to the full-height venetian VB-Eln, but blind automated based on a manufacturer's schedule to block direct sun, resulting in two tilt adjustments/day. Manually-operated, VB-E2n Zoned version of the full height two-zone VB-Eln. The upper venetian blind zone was set at a more open position relative to the lower zone to increase daylight. Automated, VB-E2n-auton1 Identical to the full-height two-zone VB-E2n, but venetian blind automated based on a manufacturer's schedule to block direct sun, resulting in two tilt adjustments each. Manually-operated, VB-E3opt Upper region three-zone mirrored blocking direct view horizontal louver of the sky, middle system and lower allow increasing levels of daylight transmission and view to the exterior. Automated, fabric RS-E-autol1 Exterior version roller shade ofauto-RS. Reference Conditions Seasonally-operated Reference-VB Slat angle adjusted interior venetian at two-month blind intervals to block direct sun for the majority of the day. Static fabric roller Reference-RS Set at a fixed shade height of 0.76 m (30 inch) above the floor for the duration of testing. APPENDIX TABLE 2. Description of Venetian Blind Systems Shade Type Zone Slat Slat Slat Slat Slat Concave Width Width Spacing Spacing (mm) (in) (mm) (in) Interior Systems Reference-VB None Down 25.4 1.0 20.0 0.79 Split-VB Upper Down 25.0 1.0 20.0 0.79 Lower Down 25.0 1.0 20.0 0.79 Split-opt-VB Upper Up 25.0 1.0 17.0 0.67 Lower Up 25.0 1.0 17.0 0.67 Auto-split-mir-VB Upper Up 82.5 3.2 71.4 2.81 Lower Up 82.5 3.2 71.4 2.81 Auto-VB None Down 25.4 1.0 20.0 0.79 Diff-VB Lower Down 25.4 1.0 20.0 0.79 Exterior Systems VB-E1n, None Down 100.0 3.94 85.0 3.35 VB-E1n-auton1 VB-E2n, Upper Down 100.0 3.94 85.0 3.35 VB-E2n-auton1 Lower Down 100.0 3.94 85.0 3.35 VB-E3opt Upper Down 77.0 3.03 70.0 2.76 Middle Down 77.0 3.03 70.0 2.76 Lower Down 77.0 3.03 70.0 2.76 Shade Type Zone Slat Slat Top Slat Bottom Concave Surface Surface Interior Systems Reference-VB None Down Semi-gloss Semi-gloss white white Split-VB Upper Down Semi-gloss Reflective white metal Lower Down White Reflective metal Split-opt-VB Upper Up Matte Prism white Lower Up Prism Matte white Auto-split-mir-VB Upper Up Mirror Matte gray Lower Up Shiny Matte gray white Auto-VB None Down Matte Matte white white Diff-VB Lower Down Semi-gloss Semi-gloss white white Exterior Systems VB-E1n, None Down Semi-gloss Semi-gloss VB-E1n-auton1 white white VB-E2n, Upper Down Semi-gloss Semi-gloss VB-E2n-auton1 white white Lower Down Semi-gloss Semi-gloss white white VB-E3opt Upper Down Polished Matte light aluminum gray Middle Down Polished Matte light aluminum gray Lower Down Polished Matte light aluminum gray Shade Type Zone Slat Upper + Concave Lower Slats Ganged? Interior Systems Reference-VB None Down No Split-VB Upper Down Yes Lower Down Yes Split-opt-VB Upper Up No Lower Up No Auto-split-mir-VB Upper Up Yes Lower Up Yes Auto-VB None Down No Diff-VB Lower Down No Exterior Systems VB-E1n, None Down No VB-E1n-auton1 VB-E2n, Upper Down No VB-E2n-auton1 Lower Down No VB-E3opt Upper Down Yes Middle Down Yes Lower Down Yes APPENDIX TABLE 3. Operation of Venetian Blind Systems Shade Type Zone Operation Auto? SA SA Equinox Winter (Deg) (Deg) Interior Systems Reference-VB No 58 12 Split-VB Upper No 35 10 Lower No 55 34 Split-opt-VB Upper No 28 6 Lower No 50 12 Auto-split-mir-VB Upper Early AM Yes 15 5 late PM Lower Early AM Yes 75 63 late PM Upper 10:00-14:00 Yes 28 28 Lower 10:00-14:00 Yes 78 78 Auto-VB Yes Diff-VB Lower No 58 12 Exterior Systems VB-E1n No 56 16 VB-E2n Upper No 31 16 Lower No 56 16 VB-E1n-auton1 Yes Manufacturer control of slat angles: solar exclusion VB-E2n-auton1 Upper Yes Manufacturer control of slat angles: daylight Lower Manufacturer control of slat angles: solar exclusion VB-E3opt Upper No 63 63 Middle No 36 36 Lower No 17 17 Shade Type Zone Operation SA BA BA Summer Winter Equinox (Deg) (Deg) (Deg) Interior Systems Reference-VB 0 0 30 Split-VB Upper 0 20 35 Lower 28 0 16 Split-opt-VB Upper 0 20 35 Lower 0 0 30 Auto-split-mir-VB Upper Early AM 0 20 35 late PM Lower Early AM 54 0 0 late PM Upper 10:00-14:00 8 6 6 Lower 10:00-14:00 70 0 0 Auto-VB LBNL control of slat angles Diff-VB Lower 0 0 30 Exterior Systems VB-E1n 16 4 32 VB-E2n Upper 16 22 32 Lower 16 4 32 VB-E1n-auton1 VB-E2n-auton1 Upper Lower VB-E3opt Upper 63 4 4 Middle 36 20 20 Lower 17 31 31 Shade Type Zone Operation BA Height Summer Above (Deg) Floor (m) Interior Systems Reference-VB 35 0 Split-VB Upper 35 0 Lower 20 0 Split-opt-VB Upper 35 0 Lower 35 0 Auto-split-mir-VB Upper Early AM 30 0, late PM 2.74 Lower Early AM 2 0, late PM 2.74 Upper 10:00-14:00 25 0, 2.74 Lower 10:00-14:00 0 0, 2.74 Auto-VB 0 Diff-VB Lower 35 0 Exterior Systems VB-E1n 32 0 VB-E2n Upper 32 0 Lower 32 0 VB-E1n-auton1 0 VB-E2n-auton1 Upper 0 Lower 0 VB-E3opt Upper 4 0 Middle 20 0 Lower 31 0 Positive slat angle: Occupant can see the exterior ground from the interior. Auto: automated; SA: slat angle; BA: blocking angle; deg: degrees. Blocking angle is defined as the profile or cut-off angle between two slats at normal incidence to the glass. Slat angle was defined as the angle between horizontal and the plane defined by the two outside edges of the slat.
Aschehoug O, Christoffersen J, Jakobiak R, Johnsen K, Lee ES, Ruck N, Selkowitz SE, editors. 2000. Daylight in buildings: A source book on daylighting systems and components. A Report of IEA SHC Task 21 / ECBCS Annex 29.
Carmody J, Selkowitz SE, Lee ES, Arasteh D, Willmert T. 2004. Window systems for high-performance buildings. New York: WW Norton & Co. 400 p.
Clear RD, Inkarojrit V, Lee ES. 2006. Subject responses to electrochromic windows. Energy Build. 38(7):758-779.
Darula S, Kittler R. 2004. Sunshine duration and daily courses of illuminance in Bratislava. Inter J Climat. 24(14):1777-1783.
Debevec PE, Malik J. 1997. Recovering high dynamic range radiance maps from photographs. In: SIGGRAPH 97. pp 369-378.
Fan D, Mardaljevic J. 2009. HDRcapOSX. How to automate capture using a Mac Mini less k/b and screen. <http://www.iesd.dmu.ac.uk/~jm/doku.php?id=resources:hdrcaposx> Accessed 2011 10 Jan.
Hopkinson RG, Bradley RC. 1960. Glare from very large sources. Illum Eng. 55(5):288-297.
Hopkinson RG. 1972. Glare from daylighting in buildings. Applied Ergonomics. 3(4):206-215.
Inanici M, Galvin J. 2004. Evaluation of high dynamic range photography as a luminance mapping Technique. Lawrence Berkeley National Laboratory. <http://escholarship.org/uc/item/9h61f5h8> Accessed 2011 10 Jan.
Lee ES., Selkowitz SE, DiBartolomeo DL, Klems JH, Clear RD, Konis K, Hitchcock R, Yazdanian M, Mitchell R, Konstantoglou M. 2009. High performance building facade solutions. California Energy Commission PIER publication: CEC-500-06-041.
Mitsunaga T, Nayar SK. 1999. Radiometric self-calibration. In: IEEE Conference on computer vision and pattern recognition (CVPR). pp 1374-380.
Nabil A, Mardaljevic J. 2006. Useful daylight illuminances: A replacement for daylight factors. Energy Build. 38(7):905-913.
Rea M, editor. 2000. The IESNA Lighting Handbook. 9th ed. New York: The Illuminating Engineering Society of North America. pp 11.16-11.18.
Sutter Y, Dumortier D, Fontoynont M. 2006. The Use of shading systems in VDU task offices: A pilot study. Energy Build. 38(7):780-789.
Tuaycharoen N, Tregenza PR. 2007. View and discomfort glare from windows. Light Res Tech. 39(2):185-200.
US Green Building Council. 2009. LEED 2009 for new construction and major renovations rating system. Version 3. 88 p.
Ward G. 1991. Real Pixels. In: Graphics Gems II. James Arvo, editor. New York: Academic Press. 643 p.
Ward G. 1994. The RADIANCE Lighting Simulation and rendering system. In: Computer Graphics, Proceedings of SIGGRAPH 94. pp 459-472.
Ward G. 1992. Radiance visual comfort calculation. <http://radsite.lbl.gov/radiance/refer/Notes/glare.html> Accessed 2011 10 Jan.
Wienold J, Christoffersen J. 2006. Evaluation methods and development of a new glare prediction model for daylight environments with the use of CCD cameras. Energy Build. 38(7):743-757.
(1.) We did not use the Wienold and Christoffersen's glare prediction model due to an error in the software tool at the time, which prevented analysis of our HDR images.
K. Konis, E.S. Lee, R.D. Clear, Ph.D
|Printer friendly Cite/link Email Feedback|
|Author:||Konis, K.; Lee, E.S.; Clear, R.D.|
|Date:||Jan 1, 2011|
|Previous Article:||The influence of color interreflections on lighting simulations.|
|Next Article:||Proposing measures of flicker in the low frequencies for lighting applications.|