- Email: [email protected]
A case study of the properties of marble as building veneer
Int. J. Rock Mech. Min. Sci. & Geomech. Abstr. Vol.30, No.7, pp. 1531-1537, 1993 Printed in Great Britain
0148-9062/93 $6.00 + 0.00 Pergamon Press Ltd
A Case Study of the Properties of Marble as Building Veneer J.M. L O G A N t M. H A S T E D T $ D. L E H N E R T $ M. D E N T O N ~~-
This study brings together documentation of two separate properties of one marble--thermal anisotropy and residual strain. The data focuses on a major building in which31% of the panels developed bows with amplitudes up to 3 cm after 17 years and the tensile strength degraded by an average of 25%. The magnitude of bowing and strength loss were found to be directly proportional. Bowing was determined to be the result of grain-boundary damage due to the thermal anisotropic character of calcite through laboratory thermal cycling for over 600, 24 hour cycles. This resulted in a loss of tensile strength. All panels were found to release residual strain when overcored, with the amount generally increasing as thermal cycling progressed. It is hypothesized that following the loss of strength, residual strain was released from the calcite. The net result of these two processes was an elongation of the panels. The constraint of the panel anchors prevented accommodation of this increased length with the result that bows developed. Inelastic bowing further reduced the panel strengths. INTRODUCTION There are a number of prominent buildings built during the 1960's and 1970's where deterioration of exterior marble facings reflects a lack of architectural and engineering understanding of the mechanical properties of rocks. These non-structural claddings were designed to minimize the weight on the load-bearing framework, utilizing thin panels--often 32 to 38 mm (1.25 to 1.5 inches) thick and more than 1.4 m (4.6 feet) on a side. Panels were independently hung so as not to be in contact with adjacent ones. Unfortunately these architectural and engineering decisions were often made without adequate investigation of the properties of the material, with the result that most of those exteriors have bowed panels. Marble panels are often at an incipient stage of failure, and owners are faced with repairs in some cases costing tens of millions of dollars. This paper presents a case study of one building owned by Amoco Corporation, Chicago in which the marble cladding was replaced. Amoco initiated, funded, assisted and permitted investigation into processes operative in the marble and the conditions of the building affecting deterioration of the exterior. This is the most comprehensive study to date to the knowledge of the authors, with the results relevant to a wide number of similar situations. t Center for Tectonophysics, Texas A&M University, College Station, TX 77843 and J. M. Logan Assoc., Rt 1, Box 1393A, Bandon, OR 97411 1: Ocean Drilling Program, Texas A&M University, College Station, TX 77843 ~fl Amoco Production Co., Houston, TX 1531
Construction on this building was carried out between 1971 and 1973. It is over 83 stories and about 346 m (1136 feet) tall making it presently the fifth tallest, and formerly the tallest marble-clad, building in the world. It is roughly square in plan view with re-entrants at the corners, and the sides containing vertical triangular columns separated by glass windows. The panels are .83 to 1.11 m (33 to 44 inches) wide depending on location and about 1.32 m (52 inches) long and 32 mm (1.25 inches) thick, except above the 42d floor in the re-entrants where they are 38 mm (1.5 inches) in thickness. The panels were hung by anchors inserted into continuous kerfs at the top and bottom of each panel. Although designed to provide freedom for thermal elongation and slight elastic bowing, this condition was compromised during construction when a two-part caulk was introduced into the upper kerf of each panel to inhibit pooling of water. Unfortunately incorrect mixing of the caulk resulted in an elastic compliance of it equal to that of the marble. By 1980 bowing was apparent, and by 1987 bowing of the exterior marble panels was sufficiently severe (Figures 1 and 2) to warrant investigation. A working hypothesis containing two elements was advanced to explain the bowing and associated strength loss. This was that (1) thermal cycling due to exposure to sunlight produced a loss of strength in the marble and (2) this in turn allowed the release of residual strain resulting in elongation of the panel that lead to bowing, and further loss of strength along the hinge line of the bow. This hypothesis developed from a number of independent observations. First, Lord Rayleigh [1] did some experimental studies relating bowing of marble to heating. He noted
1532
ROCK MECHANICS IN THE 1990s
Figure 1. Photograph of a re.entrant corner and chevron of Amoco building, Chicago, 1988. Panels are 1.32 m (52 inches) on a side. Note pervasive bowing of panels. 9-91°/~
t
Northwest [ Reentrant 35.65% 24.58%
N
"07% 39.64%
I Northeast Reentrant
18.75%
I 30.94~
~
30.92%
60.10% 26.67%
.87%
54.44%
I Southeast Southwest 139.98% 57.14%1 Reentrant Reentrant I 12.17%V 2 2 . 5 5 %
Figure 2. Schematic diagram (not to scale) of plan of Amoco building, Chicago showing generalized arc of sun's travel. Percentage of panels bowed 1.13 cm (0.5 inches) or greater at the vertical midline are shown. Values are for the total number of panels for a given side of the building or reentrant. A single chevron is shown to characterize all 14 chevrons on a given side. that after heating, the marble lengthened, became more porous, and its deflection measured in a vertical, cantilever-beam device increased. In addition to producing load-related elastic deflection of a beam, he found larger inelastic deflection occurred with increasing temperature. He attributed this to the thermal anisotropy of calcite. Later measurements by others showed that upon heating from 20°C to 100°C calcite expands 0.189% parallel to the c-axis and contracts 0.042% perpendicular to it [2]. Rayleigh postulated that the individual grains change
shape straining the original grain mosaic, eventually rupturing grain contacts and resulting in an increase in internal friction opposing recovery to the initial grain configuration. Second, observations were made of panels stored outside, which had been exposed to thermal cycling but had never been on the building. Measurements of these panels showed at their mid-point some were bowed in excess of 2 cm (0.8 inches). All panels were measured mechanically with a feeler gage that gives the height of the panel surface normal to an imaginary chord connecting their edges. It was concluded that bowing had no relation to building conditions. Third, when preparing marble specimens for microscopic thin-sections, thin wafers were observed to curl, suggesting strain relaxation. A program designed to explore this hypothesis involved: (1) visits to the source of the marble--Carrara, Italy to determine the geologic setting of the quarries; (2) hydraulically loading full-size panels to failure on the building, by Wiss, Janney, and Elstner Associates, Inc., Chicago, IL (herein refereed to as Test panels); (3) tensile strength measurements on panels that had been in controlled temperature storage since the beginning of the building (Virgin panels) and on panels cut from the building to provide access for the hydraulic testing (Access panels); (4) thermal cycling pieces of panels in the laboratory and testing their tensile strengths at various cycles; (5) overcore measurements of residual strain prior to and during thermal cycling; (6) petrographic observations of the marble; (7) temperature and pressure measurements on the building for one year; (8) compilation of data on historical temperature and wind histories for the area.
MARBLE AFTER 17 YEARS ON BUILDING Measurements of the amount of bowing along a vertical centerline of each panel were completed with the general results shown in Figure 2 where average percentages of bowed panels with amplitudes greater than 1.13 cm (0.5 inches) are given. There are 14 chevrons on each side of the building and the values are averaged for each side. The north side has the least number of bowed panels where only about 7% of the panels are bowed more than 1.13 cm. The east side by contrast has an average of 44% panels bowed more than this amount, with the south and west having intermediate values. The distribution of bowed panels in the reentrants also varies with the orientation of the panel face, with 61% bowed this amount in the southeast reentrant and both the northeast and northwest having about 30%. Even smaller domains of inhomogeneity can be seen with differences in sides of the chevrons and sides of the reentrants. Generally the larger percentages of bowed panels, and the higher bow amplitudes are found on faces receiving the greatest heating from the sun. The influence of panel thickness is also reflected in these data by comparing the measurements above and below floor 42. In the northwest reentrant, for example,
ROCK MECHANICS IN THE 1990s there are one third to one fourth the numbered of panels bowed panels over 1.13 cm above floor 42 as compared to those below. Other reentrants show comparable differences. The panels on the building exterior are nominally "Carrara marble", a term that encompasses about fifty different varieties of varying properties quarried in the Apuane Alps region of Italy [3]. X-ray and petrofabric analyses show that the marble used is nearly 100% pure calcite with traces of quartz, dolomite and fluorite. The average grain size is approximately 220 microns with little preferred orientation of c-axes. As the grains are roughly equant in shape the grain boundaries are also random. Some grains in Virgin panels show minor twinning and a few microfractures; none show preferred orientation. Grain boundaries in the Virgin material are closed and in some cases difficult to recognize. The Virgin panels have an average porosity of about 0.81% and a permeability of about 0.019 md. In contrast, in panels from the building, grain boundaries are sharply defined and in some cases noticeably separated. Average porosity in Access and Test panels has increased to 1.18% and permeability is 0.064 md. To determine current strength of the panels on the building and amount of loss in the 17 years since installation 132 panels were tested. It was assumed that 19 Virgin panels stored inside under controlled temperature conditions closely approximated the strength of the material as it was installed (Figure 3). In addition 55 Access panels and 48 Test panels had tensile strength-tests performed. These panels came from representative locations on the building and had all magnitudes of bows. Tensile strength was measured by point-load tests using 2.5 cm diameter specimens [4], with supplemental uniaxial, Brazilian, and flexural strength tests (ASTM C880). Each given strength value is the average of 12 to 18 individual measurements made on adjacent specimens. Thus the data represent over 1600 measurements. ~
• 1800 MEASUREMENTS
"6,0
"S.O
4.0
"3.0
~a 4 0 0 ' Ira
"2.0 A NO OF pANELS T]KVI~D:
- LO ll
$$
o . . . . . . . . Vifitli~. . . . . . Ac~S . . . . . . TVhT. . . . . . . .
0
Figure 3. Average tensile strength versus panel type, which in turn reflect the panels history. Access panels were cut from the building, Test panels were taken to failure on the building. All panels were then tested in the laboratory for tensile strength. Each data point is an average of at least 12 tests. Mean values and one standard deviation are shown. Percentage of average decrease in strength from that of the Virgin panels is shown.
1533
The Access and Test panels clearly show a strength loss reflecting the building's history when compared with Virgin panels, with an average decrease of 26% for the Access and 41% for Test panels or 34% if the groups are combined. They are divided in all subsequent data as measurements of all properties show a separation of resuits, which is ascribed to microfracture damage inflicted during the testing on the building prior to the tensile strength measurements. We believe that the 26% strength loss is the appropriate figure for 17 years between panel installation and testing. No differences in tensile strength were found between architectural grades on the building (that is varying color and texture), nor were there any discernible variations with height on the building. LABORATORY T H E R M A L CYCLING EXPERIMENTS Thermal cycling experiments were carried out in the laboratory to (1) quantitatively determine strength loss as a function of temperature cycle and number of cycles and (2) establish a base for extrapolating the future condition of the panels on the building. Three temperature cycles above room conditions of 75°F were investigated: 54 °, 150 °, and 225°F (Figure 4). Measurements on linear thermal expansion of marble [5,6] show that it is reasonably linear over the temperature range of interest so that the specific absolute temperatures chosen are not critical to establishing the strength-time curves. A cycle consisted of placing the panel in an oven at 75°F, raising the temperature for 12 hours, removing the panel and allowing it to cool to room temperature for 12 hours. Temperatures were calibrated with a thermister implanted in the middle of a panel with mid-panel temperatures and oven-air temperatures becoming equal after about 60 minutes of heating and within 10% of each other after about 4 hours of cooling. Pieces were cored and tensile strengths measured at various times during the cycling history. Six Virgin panels were used, 12 Access and 8 Test panels. For any given cycle the data were calculated from tests on 2 to 15 panels where the number of measurements ranged from 24 to 180 per panel; in all over 5000 strength measurements were made. Initial pieces of marble were about 45 cm on a side, but became progressively smaller as they were cored for testing. The differences in size produced no noticeable differences in strengths. Some pieces have now been run for over 600 cycles but the bulk of our data is limited to about 200 cycles. Pertinent results of these experiments are: (1) Generally all groups show a relatively rapid initial decrease of strength within the first 15-30 cycles, which is followed by a more gradual loss of strength with further cycling (Figure 4). (2) While this second stage is one of gradual strength decreases, there are intervals when the average strength increases and then declines again; these oscillatory changes are present not only from averages for all panels within a group, but are typical for individual pan-
1534
ROCK MECHANICS IN THE 1990s
els. (3) The strength loss within the first 15-30 cycles is largest (20.40%) for Virgin panels, intermediate (1834%) for Access panels and smallest (16-30%) for Test panels. (4) The strength decrease in all panels is directly proportional to the size of the temperatare cycle; with the smallest decrease found at the lowest temperature cycle of 540F in all cases and generally increasing as the magnitude of the cycle increased. 9~ $4* F
~s%
2 ~'~-~
I-I VIRGIN
F6"0
•
/
~
b.
40@ 3O@ 95 CYCLES
it.
~o oF
700"
1;$
0
VIRGIN ACCESS TEST
~6N. Ba
"5.0
4.0
v
22% 3.0
~I 4@0" ,-t
2.0 2OO
o
9'$ CYCLES
b
195
SU" 2230 F
I"1 VIRGIN • ACCESS O TEST
6N
'$.0
"4.0
z "3.0 A
~a ,=I 41%
z 22%
~
'2.0
% 1.0
c.
0 9'5 CYCLES
195
Figure 4. Average tensile strength as a function of number of thermal cycles at cycle temperatures shown. Percent decreases f~om original strength areshown. Data points are averages of 6 to 12 panels with at least 12 tests for each.
The differences in strength loss with panel type, with the Virgin panels showing the greatest strength and the Test the least, reflects their prior history. Test and Access panels have been subjected to thermal cycling and have already undergone grain-boundary damage and separation. They would be expected to show less response to cycling than the Virgin panels. The fact that all of the curves have the same characteristics regardless of their prior history was initially puzzling as it was expected that the Access and Test panels would have already gone through the initial rapid strength loss during the 17 years on the building. Additional testing was done to see if differences existed between the concave and convex sides of the bows as it was observed that textural differences existed between these two sites. Thermal cycling by itself produces grain-boundary damage and separation. Subsequent bowing exacerbates this condition on the convex side of the bow resulting in the largest grain-boundary separations. Additionally, bowing tends to compact the grains on the concave side, even though they were previously separated due to thermal cycling. Tensile strength measurements on specimens from these two sites show that the initial loss of strength in the Access and Test panels is primarily due to losses in the convex portion of the bow. After 15-45 cycles, both areas react similarly, with a slower rate of strength loss. In Virgin panels, both sides of the panel react uniformly. The transient increase in strength was unexpected. They are increases from a base strength, and they return to about that value with further cycling. Not all of the panels showed this phenomenon, but it is widespread enough to be characteristic. Petrographic observations suggest that this phenomenon is related to grain-boundary microfractures that are in turn opened, closed and reopened. Given that the grain boundaries are randomly oriented, of different lengths, and that the calcite is thermally expanding and contracting in perpendicular directions, the nature of the grain-boundaries will be expected to change as cycling proceeds; mismatches at boundaries will increase, decrease and generally be dynamic with cycling. Large numbers of open grain boundaries can close and lock, yielding higher strengths while with continued cycling these boundaries may open again yielding values slightly lower than initial base strengths. Although these periodic increases in strength are common, lasting from ten to over two hundred cycles, the strengths eventually return to a "base" value that continues to decrease although at a slower rate. Studies of mechanical and thermal fatigue processes in metals show exponential decreases of strengths as a rule [7]. Supporting the argument for similar behavior in marble is our observation that larger strength losses are found as the magnitude of thermal cycle is increased. This is in agreement with other findings that fewer higher temperature cycles produce similar effects as more lower temperature cycles [7, 8]. Finally, the strength losses found in thermal cycling, as noted, are a function of panel history, with Test panels that have already undergone the most grain-
ROCK MECHANICS
boundary damage showing the smallest losses. This supports an interpretation that the largest decreases occur during the first cycles, and once the grain-boundaries achieve a critical separation further strength losses continue at a slower rate. These arguments have led us to fit our data with an exponential curve (Figure 5) in order to assess the long-term changes, although it is recognized that other interpretations are viable. As the shape of the curves is similar over the temperature range investigated, one can interpolate to specific temperatures within the dam set. 850" • O
~750"
ISOOF 2zs*v • 5.0
[.,.
~ 650" III
20%
t...J
t..a
4.0
.~
5~.
z 450"
3.0 45%
350
51%
•
95
195
a.
CYCLES 900
I
'
i
, --
t
I
t
VIRGIN PANELS
[] •
800"~
6.0
54°F 150°F
~
IN T H E 1 9 9 0 s
1535
panels were derived field observations were made in Italy. Because of the large number of panels involved, the material probably came from a number of quarries, of which many are situated in the cores of large folds. Elongated and strained inclusions testify to the insitu strain present in many of the quarries. Additionally, conversations with quarry operators indicate that anecdotal knowledge of the presence of insitu strain is widespread and has existed in a qualitative form for many years so that quarries are identified as to its presence or absence. Unfortunately to our knowledge no insitu measurements have been made. Residual strain measurements were made in each panel by overcoring a strain gage rosette, and monitoring the response until the gage stabilized, generally about 24 hours. This was done in the laboratory with a 5 cm core so air and cooling water temperatures could be controlled. The release of inelastic residual strain was measured in all of the panels (Figure 6) with the average differential strain (algebraic maximum minus minimum, expansion taken as positive) increasing from 20 microstrains (me) in Virgin to 50 in Access and to 85 in Test panels. Maximum differential strains ranged up to 958 microstrains. Although the magnitude of the strain varied widely this consistent separation between the panels based upon their histories suggested the importance of grain-boundary damage. Thus it was postulated that some damage, either as microfracturing or grainboundary separation, through thermal cycling or mechanical stress facilitates the release of the stored strain.
5.0 y = 794.53
x^-5.633Se-2
z
R^2 = 0.844
120 ¸
< sl
~ ~' 26%
~ 28%
<
~._~ 400
lOO
4.0
[]
~
3.0
8o
,.z ~
60
~
40
t
45% y = 706.17 * x^-0.10238
R^2 = 0.812
47%
300 b.
r~
100
200
300
400
500
CYCLES
Figure 5. a) Tensile strength versus thermal cycles for laboratory tests on Virgin panels. Average tensile strength for three temperatures as a function of number of thermal cycles. Percent decreases from original strength are shown. Data points are averages of tests on 6 panels with at least 12 tests for each panel. b) The experimental data points fit with the regression curves. R E L E A S E OF RESIDUAL STRAIN "Carrara" marble is quarried from the hanging wall of a large-scale, low-angle, crustal shear zone. Frictional heating has been postulated to produce the greenschist facies metamorphism observed in the region [9]. In an attempt to determine the geologic setting from which the
c,-
N
t
2o
o
VIRGIN
ACCESS
TEST
Figure 6. Differential (maximum minus minimum; expansion taken as positive) residual strain measured upon overcoring of panels versus panel type. Mean values and one standard deviation are shown. Measurements came from 19 Virgin, 55 Access and 48 Test panels with some panels measured more than once so that a total of about 155 measurements are recorded. To evaluate this, a number of panels were thermally cycled to 90 cycles and overcored at intervals in the cycling process. Although many of the panels, especially the Access and Test, showed a continual release of strain throughout the cycling history, the magnitudes did
1536
ROCK
MECHANICS
not change significantly. There were a number of panels, specifically Virgin and some Access, that released more strain after a period of cycling (Figure 7), although the magnitude appeared to stabilize as cycling continued. In all eases this increased release was correlated with a decrease in the tensile strength (Figure 7). There is, however, not a strong correlation between magnitude of strain release and magnitude of tensile strength for all of the panels tested.
0 ,d me
-200
0
20
a.
40
60
80
I00
NUMBER OF CYCLES
u~m
IN THE
1990s
is often less than 10° F and never over 20 ° F, (4) The temperalare in the middle of a panel usually responds within an hour to changes in the outside air; that is, the panels appear to be close to thermal equilibrium most of the time. (5) The difference between panel and air temperatures varies, with the north side being the closest, the coolest and showing little radiant heating, while the south-facing panels on the east side had the greatest difference and were the warmest with most radiant heating. Although no measurements were made in the reentrants because of accessibility, it is anticipated that the highest thermal heating occurred in the southeast and southwest reentrants. (6) The largest temperature cycle was annual as expected with superimposed major cycles of about 78°F occurring about every 23 days and minor cycles of about 20°F developing about every 37 hours. Deviations from these averages occurred depending upon building conditions and orientation of panel faces but generally were lowest on the north and largest on the east and south. (7) The direct correlation with magnitude of thermal cycles and bowing (Figure 2) is noticeable. 70
SOUTH SIDE 60
0
20
40
60
80
100
NUMBER OFCYCLES
Figure 7. a) Tensile strength difference and b) residual strain difference released versus number of thermal cycles at 540F for panel V. 57. Difference values are cycled value minus uncyeled value. Tensile strength values are the mean of 12 or more tests at a given cycle. Strain values are for one overeore at each cycle.
~'so o
• x 0
OUTSIDE AIR OUTER MIDDLE
A
INSIDE AIR
+ I~NEI P'a ,++aaa
G4' "~" .
~aVq*'A AL.X
•. l ' ' k _ ~ ,a+
i oo 10 ~
IWXmNXSI~N
oN,oN
N~ON TIME
NATURAL T H E R M A L CYCLES While the process of thermal cycling and subsequent loss of strength leading to a release of residual strain are supported by the preceding laboratory data, it remained unclear whether the magnitude of natural thermal cycles was sufficient to produce the same process. To address this question temperature measurements were made for one year (March 1, 1986 through February, 1987) by Amoco Corporation. Eight monitoring stations were located on the 35th floor, two on each side of the building and on opposite sides of the column faces. Thermocoupies were placed to measure air temperatures outside and behind the panels, temperatures at the outside and inside surfaces and in the middle of the panels. Typical dam are shown in Figure 8 from which the following observations are made: (I) Panel temperatures are within 10*F of air temperatures at night but during the day rose by radiant heating. (2) Temperatures in the middle of panels are generally higher than the outside air and cooler than air ~ lhe panels, (3) Although there is a dlffermace in temperatures measured on the outer and inner surfaces of the panel, this difference
Figure 8. Temperatures measured for panels on the south side of the building on the 35th floor. Measurements are for one side of a chevron which faced east. Measurements were made January 24-25. 1987. The measurements were for the outside air, on the outer face of the panel, in the middle of the panel, on the inner face. and the air temperature behind the panel. Finite element modeling by N. Higgs of Amoco Production Co. of thermal expansion in the panels over these temperatures suggests they could result in bows of less than 0.3 mm (0.124 inches) indicating that some additional process must be involved. Some workers have suggested that thermal differences across the panel thickness would produce sufficient thermal elongation of the panel to produce bowing. To demonstrate this large difference, e.g., freezing on one face and steam on the other, have been utilized. But our field data show that differences between temperatures on the inside and outside surfaces of these panels are less than 20 ° F and generally less than 10° F. Our analytical models simulating various steps in thermal heating show that when the temperature steps are about 55 ° F or less, the mid-panel temperatures will come to within 5 ° F of equilibrium in less than one hour and
ROCK MECHANICS IN THE 1990s the inside face in about 3.5 hours. Both results are similar to those measurements for our thermal cycling experiments in laboratory ovens. Additionally, study of the historical records for the 17 years of the building life shows natural temperatures to change gradually without abrupt steps of this magnitude. Thus, it is hard to envision thermal conditions existing with this system of panel installation where large thermal differences would persist for sufficient time to produce the differential thermal expansion resulting in bows of the magnitude found.
1537
that the weak strength of calcite permits insitu and residual strain to be released so that little attention has been given to this lithology. This is clearly false in this case, but to the authors' knowledge systematic studies of other marbles are needed to assess the extent of the phenomena.
Acknowledgment-The authors wish to acknowledge Amoco DISCUSSION AND SUMMARY The field and laboratory data corroborate our initial hypothesis that the bowing is a step-wise process. Thermal cycling and the accompanying anisotropic dimensional changes in the calcite fatigue the grain boundaries eventually causing some to fail. This reduction in tensile strength allows the residual strain within the grains to be released. Both processes contribute to elongation of the panels. There may be sufficient heterogeneities in both processes to produce bowing of panels even when their ends are unconstrained. More commonly, the process requires some end constraints in the form of panel anchors which inhibit the elongation. The bowing is inelastic with the convex outer surface of the bow suffering more grain-boundary damage and separation. All of these processes reduce the panel strength. Although thermal cycling initially produces a rapid loss of strength, this rate decreases as cycling continues. As the amount of strength loss and thus the release of residual strain are functions of the magnitude of thermal cycles, the conditions on any building will be heterogeneous depending upon the sun's travel path and the amount of shading and radiation produced by surrounding buildings. The fundamental character of this process is the loss of strength produced by the anisotropic character of calcite when heated. This is exacerbated in marble by the low initial porosity. In limestones, although of similar composition, usually higher porosity provides accommodation of changes in grain geometry and thus dampens the amount of grain-boundary damage, and subsequent strength loss. Other minerals such as quartz and feldspars, the dominant components of granite, do not possess the unique thermal characteristics of calcite, and thus are not prone to such strength loss. Mitigation of the problem in marble can the achieved by using thicker panels, or making the panels smaller. Another option is to employ marble that is not 100% calcite, but has a matrix of other composition. Such lithologies are found in quarries as breccias. Here the matrix has significant amounts of clay which acts to dampen the thermal expansion. Such material is on the exterior of one building to the authors' knowledge and only shows incipient bowing after 25 years. The second component in the process, residual strain, needs more investigation to determine how common it is in marbles. The assumption has generally been made
Corporation for funding and support of this project. Discussions and assistance from J. Spencer, S. Pierce, S. Holdaway, B. Simmons, J. Dombrowski, N. Higgs, and M. Traugott of Amoco were especially profitable. D. W. Coyne of Amoco directed the temperature measurements on the Amoco building. Amoco Corporation demonstrated that industry can take the "high road" to solving immediate problems, through the integration of practical questions and thorough research. Discussions with Ian Chin and Jack Stecich of Wiss, Janney, and Elstner, Inc. expanded our knowledge of the problem. Provocative reviews by Gene Simmons and an anonymous reviewer significantly improved this paper.
REFERENCES 1. Rayleigh, L., The bending of marble, Proc. R. Soc. A, 144, 266-279 (1934). 2. Clark, Jr., S. P. Ed. Handbook of Physical Constants, 587 p., Geol. Soc. Am., New York (1966). 3. ERTAG, I Marmi Apuani, 55 p., Nuova Grafica Fiorentina, Firenze (1980). 4. ISRM, Suggested method for determining point load strength, Int. J. Rock Mech. Min Sci. & Geomech. Abstr. 22, 53-60 (1985) 5. Johnson, W. H., Parsons, W. H., Thermal expansion of concrete aggregate materials, Nat. Bureau Stds. Research Paper RP 1578 (1944). 6. Rosenholtz, J. L. and Smith, D. T., Linear thermal expansion of calcite, var. Iceland spar and Yule marble, Am. Mineralogist 134, 846-854 (1949). 7. Collins, J. A., Failure of Materials in Mechanical Design, John Wiley & Sons, New York (1981). 8. Rosengren, K. J. and Jaeger, J. C., The mechanical properties of an interlocked low-porosity aggregate, Geotechnique, 18, 317-326 (1968). 9. Carmignani, L., Giglia, G. and Klingfield, R., Structural evolution of the Apuane Alps: an example of continental margin deformation in the Northern Apennines, Italy, J. Geol., 86, 487504 (1978).
0148-9062/93 $6.00 + 0.00 Pergamon Press Ltd
A Case Study of the Properties of Marble as Building Veneer J.M. L O G A N t M. H A S T E D T $ D. L E H N E R T $ M. D E N T O N ~~-
This study brings together documentation of two separate properties of one marble--thermal anisotropy and residual strain. The data focuses on a major building in which31% of the panels developed bows with amplitudes up to 3 cm after 17 years and the tensile strength degraded by an average of 25%. The magnitude of bowing and strength loss were found to be directly proportional. Bowing was determined to be the result of grain-boundary damage due to the thermal anisotropic character of calcite through laboratory thermal cycling for over 600, 24 hour cycles. This resulted in a loss of tensile strength. All panels were found to release residual strain when overcored, with the amount generally increasing as thermal cycling progressed. It is hypothesized that following the loss of strength, residual strain was released from the calcite. The net result of these two processes was an elongation of the panels. The constraint of the panel anchors prevented accommodation of this increased length with the result that bows developed. Inelastic bowing further reduced the panel strengths. INTRODUCTION There are a number of prominent buildings built during the 1960's and 1970's where deterioration of exterior marble facings reflects a lack of architectural and engineering understanding of the mechanical properties of rocks. These non-structural claddings were designed to minimize the weight on the load-bearing framework, utilizing thin panels--often 32 to 38 mm (1.25 to 1.5 inches) thick and more than 1.4 m (4.6 feet) on a side. Panels were independently hung so as not to be in contact with adjacent ones. Unfortunately these architectural and engineering decisions were often made without adequate investigation of the properties of the material, with the result that most of those exteriors have bowed panels. Marble panels are often at an incipient stage of failure, and owners are faced with repairs in some cases costing tens of millions of dollars. This paper presents a case study of one building owned by Amoco Corporation, Chicago in which the marble cladding was replaced. Amoco initiated, funded, assisted and permitted investigation into processes operative in the marble and the conditions of the building affecting deterioration of the exterior. This is the most comprehensive study to date to the knowledge of the authors, with the results relevant to a wide number of similar situations. t Center for Tectonophysics, Texas A&M University, College Station, TX 77843 and J. M. Logan Assoc., Rt 1, Box 1393A, Bandon, OR 97411 1: Ocean Drilling Program, Texas A&M University, College Station, TX 77843 ~fl Amoco Production Co., Houston, TX 1531
Construction on this building was carried out between 1971 and 1973. It is over 83 stories and about 346 m (1136 feet) tall making it presently the fifth tallest, and formerly the tallest marble-clad, building in the world. It is roughly square in plan view with re-entrants at the corners, and the sides containing vertical triangular columns separated by glass windows. The panels are .83 to 1.11 m (33 to 44 inches) wide depending on location and about 1.32 m (52 inches) long and 32 mm (1.25 inches) thick, except above the 42d floor in the re-entrants where they are 38 mm (1.5 inches) in thickness. The panels were hung by anchors inserted into continuous kerfs at the top and bottom of each panel. Although designed to provide freedom for thermal elongation and slight elastic bowing, this condition was compromised during construction when a two-part caulk was introduced into the upper kerf of each panel to inhibit pooling of water. Unfortunately incorrect mixing of the caulk resulted in an elastic compliance of it equal to that of the marble. By 1980 bowing was apparent, and by 1987 bowing of the exterior marble panels was sufficiently severe (Figures 1 and 2) to warrant investigation. A working hypothesis containing two elements was advanced to explain the bowing and associated strength loss. This was that (1) thermal cycling due to exposure to sunlight produced a loss of strength in the marble and (2) this in turn allowed the release of residual strain resulting in elongation of the panel that lead to bowing, and further loss of strength along the hinge line of the bow. This hypothesis developed from a number of independent observations. First, Lord Rayleigh [1] did some experimental studies relating bowing of marble to heating. He noted
1532
ROCK MECHANICS IN THE 1990s
Figure 1. Photograph of a re.entrant corner and chevron of Amoco building, Chicago, 1988. Panels are 1.32 m (52 inches) on a side. Note pervasive bowing of panels. 9-91°/~
t
Northwest [ Reentrant 35.65% 24.58%
N
"07% 39.64%
I Northeast Reentrant
18.75%
I 30.94~
~
30.92%
60.10% 26.67%
.87%
54.44%
I Southeast Southwest 139.98% 57.14%1 Reentrant Reentrant I 12.17%V 2 2 . 5 5 %
Figure 2. Schematic diagram (not to scale) of plan of Amoco building, Chicago showing generalized arc of sun's travel. Percentage of panels bowed 1.13 cm (0.5 inches) or greater at the vertical midline are shown. Values are for the total number of panels for a given side of the building or reentrant. A single chevron is shown to characterize all 14 chevrons on a given side. that after heating, the marble lengthened, became more porous, and its deflection measured in a vertical, cantilever-beam device increased. In addition to producing load-related elastic deflection of a beam, he found larger inelastic deflection occurred with increasing temperature. He attributed this to the thermal anisotropy of calcite. Later measurements by others showed that upon heating from 20°C to 100°C calcite expands 0.189% parallel to the c-axis and contracts 0.042% perpendicular to it [2]. Rayleigh postulated that the individual grains change
shape straining the original grain mosaic, eventually rupturing grain contacts and resulting in an increase in internal friction opposing recovery to the initial grain configuration. Second, observations were made of panels stored outside, which had been exposed to thermal cycling but had never been on the building. Measurements of these panels showed at their mid-point some were bowed in excess of 2 cm (0.8 inches). All panels were measured mechanically with a feeler gage that gives the height of the panel surface normal to an imaginary chord connecting their edges. It was concluded that bowing had no relation to building conditions. Third, when preparing marble specimens for microscopic thin-sections, thin wafers were observed to curl, suggesting strain relaxation. A program designed to explore this hypothesis involved: (1) visits to the source of the marble--Carrara, Italy to determine the geologic setting of the quarries; (2) hydraulically loading full-size panels to failure on the building, by Wiss, Janney, and Elstner Associates, Inc., Chicago, IL (herein refereed to as Test panels); (3) tensile strength measurements on panels that had been in controlled temperature storage since the beginning of the building (Virgin panels) and on panels cut from the building to provide access for the hydraulic testing (Access panels); (4) thermal cycling pieces of panels in the laboratory and testing their tensile strengths at various cycles; (5) overcore measurements of residual strain prior to and during thermal cycling; (6) petrographic observations of the marble; (7) temperature and pressure measurements on the building for one year; (8) compilation of data on historical temperature and wind histories for the area.
MARBLE AFTER 17 YEARS ON BUILDING Measurements of the amount of bowing along a vertical centerline of each panel were completed with the general results shown in Figure 2 where average percentages of bowed panels with amplitudes greater than 1.13 cm (0.5 inches) are given. There are 14 chevrons on each side of the building and the values are averaged for each side. The north side has the least number of bowed panels where only about 7% of the panels are bowed more than 1.13 cm. The east side by contrast has an average of 44% panels bowed more than this amount, with the south and west having intermediate values. The distribution of bowed panels in the reentrants also varies with the orientation of the panel face, with 61% bowed this amount in the southeast reentrant and both the northeast and northwest having about 30%. Even smaller domains of inhomogeneity can be seen with differences in sides of the chevrons and sides of the reentrants. Generally the larger percentages of bowed panels, and the higher bow amplitudes are found on faces receiving the greatest heating from the sun. The influence of panel thickness is also reflected in these data by comparing the measurements above and below floor 42. In the northwest reentrant, for example,
ROCK MECHANICS IN THE 1990s there are one third to one fourth the numbered of panels bowed panels over 1.13 cm above floor 42 as compared to those below. Other reentrants show comparable differences. The panels on the building exterior are nominally "Carrara marble", a term that encompasses about fifty different varieties of varying properties quarried in the Apuane Alps region of Italy [3]. X-ray and petrofabric analyses show that the marble used is nearly 100% pure calcite with traces of quartz, dolomite and fluorite. The average grain size is approximately 220 microns with little preferred orientation of c-axes. As the grains are roughly equant in shape the grain boundaries are also random. Some grains in Virgin panels show minor twinning and a few microfractures; none show preferred orientation. Grain boundaries in the Virgin material are closed and in some cases difficult to recognize. The Virgin panels have an average porosity of about 0.81% and a permeability of about 0.019 md. In contrast, in panels from the building, grain boundaries are sharply defined and in some cases noticeably separated. Average porosity in Access and Test panels has increased to 1.18% and permeability is 0.064 md. To determine current strength of the panels on the building and amount of loss in the 17 years since installation 132 panels were tested. It was assumed that 19 Virgin panels stored inside under controlled temperature conditions closely approximated the strength of the material as it was installed (Figure 3). In addition 55 Access panels and 48 Test panels had tensile strength-tests performed. These panels came from representative locations on the building and had all magnitudes of bows. Tensile strength was measured by point-load tests using 2.5 cm diameter specimens [4], with supplemental uniaxial, Brazilian, and flexural strength tests (ASTM C880). Each given strength value is the average of 12 to 18 individual measurements made on adjacent specimens. Thus the data represent over 1600 measurements. ~
• 1800 MEASUREMENTS
"6,0
"S.O
4.0
"3.0
~a 4 0 0 ' Ira
"2.0 A NO OF pANELS T]KVI~D:
- LO ll
$$
o . . . . . . . . Vifitli~. . . . . . Ac~S . . . . . . TVhT. . . . . . . .
0
Figure 3. Average tensile strength versus panel type, which in turn reflect the panels history. Access panels were cut from the building, Test panels were taken to failure on the building. All panels were then tested in the laboratory for tensile strength. Each data point is an average of at least 12 tests. Mean values and one standard deviation are shown. Percentage of average decrease in strength from that of the Virgin panels is shown.
1533
The Access and Test panels clearly show a strength loss reflecting the building's history when compared with Virgin panels, with an average decrease of 26% for the Access and 41% for Test panels or 34% if the groups are combined. They are divided in all subsequent data as measurements of all properties show a separation of resuits, which is ascribed to microfracture damage inflicted during the testing on the building prior to the tensile strength measurements. We believe that the 26% strength loss is the appropriate figure for 17 years between panel installation and testing. No differences in tensile strength were found between architectural grades on the building (that is varying color and texture), nor were there any discernible variations with height on the building. LABORATORY T H E R M A L CYCLING EXPERIMENTS Thermal cycling experiments were carried out in the laboratory to (1) quantitatively determine strength loss as a function of temperature cycle and number of cycles and (2) establish a base for extrapolating the future condition of the panels on the building. Three temperature cycles above room conditions of 75°F were investigated: 54 °, 150 °, and 225°F (Figure 4). Measurements on linear thermal expansion of marble [5,6] show that it is reasonably linear over the temperature range of interest so that the specific absolute temperatures chosen are not critical to establishing the strength-time curves. A cycle consisted of placing the panel in an oven at 75°F, raising the temperature for 12 hours, removing the panel and allowing it to cool to room temperature for 12 hours. Temperatures were calibrated with a thermister implanted in the middle of a panel with mid-panel temperatures and oven-air temperatures becoming equal after about 60 minutes of heating and within 10% of each other after about 4 hours of cooling. Pieces were cored and tensile strengths measured at various times during the cycling history. Six Virgin panels were used, 12 Access and 8 Test panels. For any given cycle the data were calculated from tests on 2 to 15 panels where the number of measurements ranged from 24 to 180 per panel; in all over 5000 strength measurements were made. Initial pieces of marble were about 45 cm on a side, but became progressively smaller as they were cored for testing. The differences in size produced no noticeable differences in strengths. Some pieces have now been run for over 600 cycles but the bulk of our data is limited to about 200 cycles. Pertinent results of these experiments are: (1) Generally all groups show a relatively rapid initial decrease of strength within the first 15-30 cycles, which is followed by a more gradual loss of strength with further cycling (Figure 4). (2) While this second stage is one of gradual strength decreases, there are intervals when the average strength increases and then declines again; these oscillatory changes are present not only from averages for all panels within a group, but are typical for individual pan-
1534
ROCK MECHANICS IN THE 1990s
els. (3) The strength loss within the first 15-30 cycles is largest (20.40%) for Virgin panels, intermediate (1834%) for Access panels and smallest (16-30%) for Test panels. (4) The strength decrease in all panels is directly proportional to the size of the temperatare cycle; with the smallest decrease found at the lowest temperature cycle of 540F in all cases and generally increasing as the magnitude of the cycle increased. 9~ $4* F
~s%
2 ~'~-~
I-I VIRGIN
F6"0
•
/
~
b.
40@ 3O@ 95 CYCLES
it.
~o oF
700"
1;$
0
VIRGIN ACCESS TEST
~6N. Ba
"5.0
4.0
v
22% 3.0
~I 4@0" ,-t
2.0 2OO
o
9'$ CYCLES
b
195
SU" 2230 F
I"1 VIRGIN • ACCESS O TEST
6N
'$.0
"4.0
z "3.0 A
~a ,=I 41%
z 22%
~
'2.0
% 1.0
c.
0 9'5 CYCLES
195
Figure 4. Average tensile strength as a function of number of thermal cycles at cycle temperatures shown. Percent decreases f~om original strength areshown. Data points are averages of 6 to 12 panels with at least 12 tests for each.
The differences in strength loss with panel type, with the Virgin panels showing the greatest strength and the Test the least, reflects their prior history. Test and Access panels have been subjected to thermal cycling and have already undergone grain-boundary damage and separation. They would be expected to show less response to cycling than the Virgin panels. The fact that all of the curves have the same characteristics regardless of their prior history was initially puzzling as it was expected that the Access and Test panels would have already gone through the initial rapid strength loss during the 17 years on the building. Additional testing was done to see if differences existed between the concave and convex sides of the bows as it was observed that textural differences existed between these two sites. Thermal cycling by itself produces grain-boundary damage and separation. Subsequent bowing exacerbates this condition on the convex side of the bow resulting in the largest grain-boundary separations. Additionally, bowing tends to compact the grains on the concave side, even though they were previously separated due to thermal cycling. Tensile strength measurements on specimens from these two sites show that the initial loss of strength in the Access and Test panels is primarily due to losses in the convex portion of the bow. After 15-45 cycles, both areas react similarly, with a slower rate of strength loss. In Virgin panels, both sides of the panel react uniformly. The transient increase in strength was unexpected. They are increases from a base strength, and they return to about that value with further cycling. Not all of the panels showed this phenomenon, but it is widespread enough to be characteristic. Petrographic observations suggest that this phenomenon is related to grain-boundary microfractures that are in turn opened, closed and reopened. Given that the grain boundaries are randomly oriented, of different lengths, and that the calcite is thermally expanding and contracting in perpendicular directions, the nature of the grain-boundaries will be expected to change as cycling proceeds; mismatches at boundaries will increase, decrease and generally be dynamic with cycling. Large numbers of open grain boundaries can close and lock, yielding higher strengths while with continued cycling these boundaries may open again yielding values slightly lower than initial base strengths. Although these periodic increases in strength are common, lasting from ten to over two hundred cycles, the strengths eventually return to a "base" value that continues to decrease although at a slower rate. Studies of mechanical and thermal fatigue processes in metals show exponential decreases of strengths as a rule [7]. Supporting the argument for similar behavior in marble is our observation that larger strength losses are found as the magnitude of thermal cycle is increased. This is in agreement with other findings that fewer higher temperature cycles produce similar effects as more lower temperature cycles [7, 8]. Finally, the strength losses found in thermal cycling, as noted, are a function of panel history, with Test panels that have already undergone the most grain-
ROCK MECHANICS
boundary damage showing the smallest losses. This supports an interpretation that the largest decreases occur during the first cycles, and once the grain-boundaries achieve a critical separation further strength losses continue at a slower rate. These arguments have led us to fit our data with an exponential curve (Figure 5) in order to assess the long-term changes, although it is recognized that other interpretations are viable. As the shape of the curves is similar over the temperature range investigated, one can interpolate to specific temperatures within the dam set. 850" • O
~750"
ISOOF 2zs*v • 5.0
[.,.
~ 650" III
20%
t...J
t..a
4.0
.~
5~.
z 450"
3.0 45%
350
51%
•
95
195
a.
CYCLES 900
I
'
i
, --
t
I
t
VIRGIN PANELS
[] •
800"~
6.0
54°F 150°F
~
IN T H E 1 9 9 0 s
1535
panels were derived field observations were made in Italy. Because of the large number of panels involved, the material probably came from a number of quarries, of which many are situated in the cores of large folds. Elongated and strained inclusions testify to the insitu strain present in many of the quarries. Additionally, conversations with quarry operators indicate that anecdotal knowledge of the presence of insitu strain is widespread and has existed in a qualitative form for many years so that quarries are identified as to its presence or absence. Unfortunately to our knowledge no insitu measurements have been made. Residual strain measurements were made in each panel by overcoring a strain gage rosette, and monitoring the response until the gage stabilized, generally about 24 hours. This was done in the laboratory with a 5 cm core so air and cooling water temperatures could be controlled. The release of inelastic residual strain was measured in all of the panels (Figure 6) with the average differential strain (algebraic maximum minus minimum, expansion taken as positive) increasing from 20 microstrains (me) in Virgin to 50 in Access and to 85 in Test panels. Maximum differential strains ranged up to 958 microstrains. Although the magnitude of the strain varied widely this consistent separation between the panels based upon their histories suggested the importance of grain-boundary damage. Thus it was postulated that some damage, either as microfracturing or grainboundary separation, through thermal cycling or mechanical stress facilitates the release of the stored strain.
5.0 y = 794.53
x^-5.633Se-2
z
R^2 = 0.844
120 ¸
< sl
~ ~' 26%
~ 28%
<
~._~ 400
lOO
4.0
[]
~
3.0
8o
,.z ~
60
~
40
t
45% y = 706.17 * x^-0.10238
R^2 = 0.812
47%
300 b.
r~
100
200
300
400
500
CYCLES
Figure 5. a) Tensile strength versus thermal cycles for laboratory tests on Virgin panels. Average tensile strength for three temperatures as a function of number of thermal cycles. Percent decreases from original strength are shown. Data points are averages of tests on 6 panels with at least 12 tests for each panel. b) The experimental data points fit with the regression curves. R E L E A S E OF RESIDUAL STRAIN "Carrara" marble is quarried from the hanging wall of a large-scale, low-angle, crustal shear zone. Frictional heating has been postulated to produce the greenschist facies metamorphism observed in the region [9]. In an attempt to determine the geologic setting from which the
c,-
N
t
2o
o
VIRGIN
ACCESS
TEST
Figure 6. Differential (maximum minus minimum; expansion taken as positive) residual strain measured upon overcoring of panels versus panel type. Mean values and one standard deviation are shown. Measurements came from 19 Virgin, 55 Access and 48 Test panels with some panels measured more than once so that a total of about 155 measurements are recorded. To evaluate this, a number of panels were thermally cycled to 90 cycles and overcored at intervals in the cycling process. Although many of the panels, especially the Access and Test, showed a continual release of strain throughout the cycling history, the magnitudes did
1536
ROCK
MECHANICS
not change significantly. There were a number of panels, specifically Virgin and some Access, that released more strain after a period of cycling (Figure 7), although the magnitude appeared to stabilize as cycling continued. In all eases this increased release was correlated with a decrease in the tensile strength (Figure 7). There is, however, not a strong correlation between magnitude of strain release and magnitude of tensile strength for all of the panels tested.
0 ,d me
-200
0
20
a.
40
60
80
I00
NUMBER OF CYCLES
u~m
IN THE
1990s
is often less than 10° F and never over 20 ° F, (4) The temperalare in the middle of a panel usually responds within an hour to changes in the outside air; that is, the panels appear to be close to thermal equilibrium most of the time. (5) The difference between panel and air temperatures varies, with the north side being the closest, the coolest and showing little radiant heating, while the south-facing panels on the east side had the greatest difference and were the warmest with most radiant heating. Although no measurements were made in the reentrants because of accessibility, it is anticipated that the highest thermal heating occurred in the southeast and southwest reentrants. (6) The largest temperature cycle was annual as expected with superimposed major cycles of about 78°F occurring about every 23 days and minor cycles of about 20°F developing about every 37 hours. Deviations from these averages occurred depending upon building conditions and orientation of panel faces but generally were lowest on the north and largest on the east and south. (7) The direct correlation with magnitude of thermal cycles and bowing (Figure 2) is noticeable. 70
SOUTH SIDE 60
0
20
40
60
80
100
NUMBER OFCYCLES
Figure 7. a) Tensile strength difference and b) residual strain difference released versus number of thermal cycles at 540F for panel V. 57. Difference values are cycled value minus uncyeled value. Tensile strength values are the mean of 12 or more tests at a given cycle. Strain values are for one overeore at each cycle.
~'so o
• x 0
OUTSIDE AIR OUTER MIDDLE
A
INSIDE AIR
+ I~NEI P'a ,++aaa
G4' "~" .
~aVq*'A AL.X
•. l ' ' k _ ~ ,a+
i oo 10 ~
IWXmNXSI~N
oN,oN
N~ON TIME
NATURAL T H E R M A L CYCLES While the process of thermal cycling and subsequent loss of strength leading to a release of residual strain are supported by the preceding laboratory data, it remained unclear whether the magnitude of natural thermal cycles was sufficient to produce the same process. To address this question temperature measurements were made for one year (March 1, 1986 through February, 1987) by Amoco Corporation. Eight monitoring stations were located on the 35th floor, two on each side of the building and on opposite sides of the column faces. Thermocoupies were placed to measure air temperatures outside and behind the panels, temperatures at the outside and inside surfaces and in the middle of the panels. Typical dam are shown in Figure 8 from which the following observations are made: (I) Panel temperatures are within 10*F of air temperatures at night but during the day rose by radiant heating. (2) Temperatures in the middle of panels are generally higher than the outside air and cooler than air ~ lhe panels, (3) Although there is a dlffermace in temperatures measured on the outer and inner surfaces of the panel, this difference
Figure 8. Temperatures measured for panels on the south side of the building on the 35th floor. Measurements are for one side of a chevron which faced east. Measurements were made January 24-25. 1987. The measurements were for the outside air, on the outer face of the panel, in the middle of the panel, on the inner face. and the air temperature behind the panel. Finite element modeling by N. Higgs of Amoco Production Co. of thermal expansion in the panels over these temperatures suggests they could result in bows of less than 0.3 mm (0.124 inches) indicating that some additional process must be involved. Some workers have suggested that thermal differences across the panel thickness would produce sufficient thermal elongation of the panel to produce bowing. To demonstrate this large difference, e.g., freezing on one face and steam on the other, have been utilized. But our field data show that differences between temperatures on the inside and outside surfaces of these panels are less than 20 ° F and generally less than 10° F. Our analytical models simulating various steps in thermal heating show that when the temperature steps are about 55 ° F or less, the mid-panel temperatures will come to within 5 ° F of equilibrium in less than one hour and
ROCK MECHANICS IN THE 1990s the inside face in about 3.5 hours. Both results are similar to those measurements for our thermal cycling experiments in laboratory ovens. Additionally, study of the historical records for the 17 years of the building life shows natural temperatures to change gradually without abrupt steps of this magnitude. Thus, it is hard to envision thermal conditions existing with this system of panel installation where large thermal differences would persist for sufficient time to produce the differential thermal expansion resulting in bows of the magnitude found.
1537
that the weak strength of calcite permits insitu and residual strain to be released so that little attention has been given to this lithology. This is clearly false in this case, but to the authors' knowledge systematic studies of other marbles are needed to assess the extent of the phenomena.
Acknowledgment-The authors wish to acknowledge Amoco DISCUSSION AND SUMMARY The field and laboratory data corroborate our initial hypothesis that the bowing is a step-wise process. Thermal cycling and the accompanying anisotropic dimensional changes in the calcite fatigue the grain boundaries eventually causing some to fail. This reduction in tensile strength allows the residual strain within the grains to be released. Both processes contribute to elongation of the panels. There may be sufficient heterogeneities in both processes to produce bowing of panels even when their ends are unconstrained. More commonly, the process requires some end constraints in the form of panel anchors which inhibit the elongation. The bowing is inelastic with the convex outer surface of the bow suffering more grain-boundary damage and separation. All of these processes reduce the panel strength. Although thermal cycling initially produces a rapid loss of strength, this rate decreases as cycling continues. As the amount of strength loss and thus the release of residual strain are functions of the magnitude of thermal cycles, the conditions on any building will be heterogeneous depending upon the sun's travel path and the amount of shading and radiation produced by surrounding buildings. The fundamental character of this process is the loss of strength produced by the anisotropic character of calcite when heated. This is exacerbated in marble by the low initial porosity. In limestones, although of similar composition, usually higher porosity provides accommodation of changes in grain geometry and thus dampens the amount of grain-boundary damage, and subsequent strength loss. Other minerals such as quartz and feldspars, the dominant components of granite, do not possess the unique thermal characteristics of calcite, and thus are not prone to such strength loss. Mitigation of the problem in marble can the achieved by using thicker panels, or making the panels smaller. Another option is to employ marble that is not 100% calcite, but has a matrix of other composition. Such lithologies are found in quarries as breccias. Here the matrix has significant amounts of clay which acts to dampen the thermal expansion. Such material is on the exterior of one building to the authors' knowledge and only shows incipient bowing after 25 years. The second component in the process, residual strain, needs more investigation to determine how common it is in marbles. The assumption has generally been made
Corporation for funding and support of this project. Discussions and assistance from J. Spencer, S. Pierce, S. Holdaway, B. Simmons, J. Dombrowski, N. Higgs, and M. Traugott of Amoco were especially profitable. D. W. Coyne of Amoco directed the temperature measurements on the Amoco building. Amoco Corporation demonstrated that industry can take the "high road" to solving immediate problems, through the integration of practical questions and thorough research. Discussions with Ian Chin and Jack Stecich of Wiss, Janney, and Elstner, Inc. expanded our knowledge of the problem. Provocative reviews by Gene Simmons and an anonymous reviewer significantly improved this paper.
REFERENCES 1. Rayleigh, L., The bending of marble, Proc. R. Soc. A, 144, 266-279 (1934). 2. Clark, Jr., S. P. Ed. Handbook of Physical Constants, 587 p., Geol. Soc. Am., New York (1966). 3. ERTAG, I Marmi Apuani, 55 p., Nuova Grafica Fiorentina, Firenze (1980). 4. ISRM, Suggested method for determining point load strength, Int. J. Rock Mech. Min Sci. & Geomech. Abstr. 22, 53-60 (1985) 5. Johnson, W. H., Parsons, W. H., Thermal expansion of concrete aggregate materials, Nat. Bureau Stds. Research Paper RP 1578 (1944). 6. Rosenholtz, J. L. and Smith, D. T., Linear thermal expansion of calcite, var. Iceland spar and Yule marble, Am. Mineralogist 134, 846-854 (1949). 7. Collins, J. A., Failure of Materials in Mechanical Design, John Wiley & Sons, New York (1981). 8. Rosengren, K. J. and Jaeger, J. C., The mechanical properties of an interlocked low-porosity aggregate, Geotechnique, 18, 317-326 (1968). 9. Carmignani, L., Giglia, G. and Klingfield, R., Structural evolution of the Apuane Alps: an example of continental margin deformation in the Northern Apennines, Italy, J. Geol., 86, 487504 (1978).