erched on a hillside in southwestern Pennsylva- nia, about 72 miles from Pittsburgh, is one of the world’s most famous houses. Fallingwater, the stunning creation of architect Frank Lloyd Wright, has been an American icon since its con- A firsthand account struction in 1937. MoreP than two million tourists have visited the site and stared in by renowned engineer awe at the building’s concrete terraces hanging over a clear, swift-running stream. Architecture critics have extolled Fallingwater as Wright’s greatest achievement. In Robert Silman fact, in 1991 the American Institute of Architects voted it the best work ever pro- duced by an American architect. Yet this incomparable structure has a critical flaw. Wright’s design did not pro- vide enough support for the portion of the house that hangs over the stream. As a result, Fallingwater’s famed terraces began to droop as soon as they were built, causing large cracks to appear in the concrete. What is more, the sagging gradual- ly increased over the next six decades. In 1995 the Western Pennsylvania Conser- vancy, which owns Fallingwater, was concerned enough to hire our engineering firm, Robert Silman Associates in New York City, to examine the house’s structur- al problems. The results of our investigation indicated that the beams supporting the house were continuing to bend and that the building would eventually collapse into the stream below if nothing was done. In 1996 the conservancy prudently decided to shore up Fallingwater with tem- porary steel beams and columns. At the same time, our office began to draw up a plan to permanently repair the house. We had previously worked on two other buildings designed by Wright—the Darwin D. Martin House in Buffalo, N.Y., and Wingspread in Racine, Wis.—but Fallingwater posed a unique challenge. To deter- mine how to relieve the stresses that were threatening the house, our engineers probed the building with radar and ultrasonic pulses, then performed a rigorous structural analysis. Along the way we also tried to retrace the thinking of Wright and his apprentices. We now have a plausible theory to explain how the design of Fallingwater went awry. The story of Fallingwater begins with Edgar Kaufmann, Sr., who owned a suc- cessful department store in Pittsburgh in the 1930s. His son, Edgar Kaufmann, jr. (he always spelled “junior” with a lowercase “j”), spent a short time as an ap- prentice in Wright’s studio at Taliesin, the architect’s estate in Spring Green, Wis. Kaufmann, jr., convinced his father to retain Wright to do some work at the store and later to design a weekend house for the family on a site that had formerly been The Plan to Save Fallingwater This breathtaking house designed by Frank Lloyd Wright was in danger of collapse until an engineering firm found a way to stop it from falling down ROBERT P.RUSCHAK/COURTESY OF WESTERN PENNSYLVANIA CONSERVANCY CONSERVANCY ROBERT P.RUSCHAK/COURTESYWESTERN PENNSYLVANIA OF

Copyright 2000 Scientific American, Inc. Copyright 2000 Scientific American, Inc. a summer recreation camp for the store’s the house so that the section over Bear is a concrete slab that serves as the fin- employees. Run acts as a cantilever. Like a diving ished underside of the structure. Wright The wooded property features a small board, it has a fixed end and a free end. chose this design to give the house’s ex- stream known as Bear Run that cas- The fixed end consists of four large bol- terior a monolithic look, but it also had cades over a series of rocky ledges. The sters, three of reinforced concrete (that a structural purpose. In engineering Kaufmanns had always assumed that is, concrete with steel bars embedded in terms, a cantilever has a negative bend- their house would be located down- it) and one of stone masonry. These ing moment—the load at the free end of stream from the ledges, at a point bolsters rise from the sandstone ledge the horizontal beam is resisted by ten- where the waterfalls could be viewed to the building’s first floor [see illustra- sion in the beam’s upper side and by from below. But it was Wright’s genius tion on pages 92 and 93]. Each one compression in the lower side. (In con- to site the house above the falls, on top supports a horizontal reinforced-con- trast, a bookshelf has a positive bend- of a large sandstone ledge that over- crete beam that extends some 4.42 me- ing moment—the weight of the books is looks the stream. The building was de- ters (14.5 feet) beyond the bolster, jut- resisted by compression in the shelf’s signed in 1935, and construction start- ting southward over the stream. The upper side and by tension in the lower ed in 1936. The design work was con- beams are connected to one another by side.) Wright’s decision to put the con- ducted at the Taliesin studio, with concrete joists, each 100 millimeters crete slab under the cantilever beams Wright’s apprentices Bob Mosher and (four inches) wide. Together the beams turned them into inverted tee beams— Edgar Tafel participating significantly. and joists create a rectilinear grid. each shaped like an upside-down T— The structural calculations for Falling- Above this grid are wooden two-by- thereby raising their resistance to com- water were done in the same studio by fours and planking, which support the pression and enabling them to support engineers Mendel Glickman and Wil- stone floor of the house’s living room a greater load. liam Wesley Peters. and the first-floor terraces. Fallingwater has more than one can- Wright and his apprentices designed Beneath the joists and cantilever beams tilever, though. Terraces extend from

90 Scientific American September 2000 The Plan to Save Fallingwater Copyright 2000 Scientific American, Inc. INTERIOR VIEW of Fallingwater’s living room ment was not enough, as the builders shows the stone floor that rests on the discovered during Fallingwater’s con- house’s cantilever beams. The windows at struction. When workers removed the the south end of the room are divided by wooden formwork from beneath the four steel mullions that help support the concrete of the first floor, they recorded weight of the second floor. an instantaneous downward movement of 44.5 millimeters. It is not unusual for out than the first floor does, extending a small amount of deflection to occur an additional 1.83 meters (six feet) when the scaffolding is removed from a southward [see left illustration below]. concrete structure, but in this case the Four T-shaped window mullions rise bending was especially pronounced. from the south edge of the living room Mosher, the apprentice on site, tele- to the terrace above. At first glance phoned Glickman at the studio in Tal- these steel mullions appear to be merely iesin. After a quick check of his calcula- )

decorative, but we would eventually tions Glickman is reported to have ex- ight

learn that they, too, play a key role in claimed, “Oh my God, I forgot the om r ott b Fallingwater’s structure. negative reinforcement!” (

Concerns about the soundness of Glickman was referring to the rein- ANCY V Wright’s design arose even before con- forcement needed to balance the nega- struction started. Metzger-Richardson, tive bending moment, which causes ONSER ANIA C

the Pittsburgh engineering firm that compression in the lower part of each V supplied the steel bars for the rein- cantilever beam and tension in the up- forced concrete, insisted that there were per part. In any beam made of rein- ENNSYL not enough bars in the cantilever beams forced concrete, the concrete resists the

below the living room. To make the compression on the beam and the steel WESTERN P beams strong enough to resist bending bars in the concrete resist the tension. under their load, the firm doubled the Fallingwater’s cantilever beams could TESY OF OUR C );

number of one-inch-square bars in each handle the compression caused by the t lef m

beam from eight to 16. Wright was fu- negative moment, but there were not o ott rious when he learned about the change. enough steel bars in the upper parts of b

He believed that the additional steel the beams to balance the tension. and op bars would increase the weight of the The problem became even more ap- t ANCY (

beams too much and thus weaken the parent after the completion of the sec- V structure. In an angry letter to Kauf- ond floor. Soon after workers removed ONSER mann, Sr., he wrote: “I have put so the formwork from the concrete of the ANIA C much more into this house than you or master bedroom terrace, two cracks ap- V any other client has a right to expect, peared in the terrace’s parapets. In 1937 ENNSYL that if I don’t have your confidence—to Metzger-Richardson conducted load hell with the whole thing.” tests of the structure and calculated that Kaufmann, Sr., appeased his architect the stresses in the cantilever beams were WESTERN P

the east and west sides of the first floor, by asserting his confidence in him. But near or even exceeded the margins of TESY OF

supported by concrete joists under their Wright was clearly wrong about the safety. The engineering firm recom- OUR C floors and by edge beams in their para- cantilever beams: if Metzger-Richard- mended placing permanent props in the K/

pets. And on the building’s second son had not slipped in the extra steel streambed to support the first floor and USCHA R .

floor, directly above the living room, bars, the beams surely would have failed. thus reduce the length of the can- T P

the master bedroom terrace juts farther Even the greater amount of reinforce- tilevers. But Wright stubbornly defend- OBER R

CRACKS IN THE PARAPETS of the second- floor terrace appeared as a result of the stresses in the concrete structure (left).Elec- tronic monitors attached to the parapets measured changes in the width of the cracks over 18 months and confirmed that they were growing (below).

The Plan to Save Fallingwater Copyright 2000 Scientific American, Inc. WEST TERRACE

MASTER BEDROOM TERRACE

EAST TERRACE

ed his design. Once again he forced low, because the two floors are struc- Kaufmann, Sr., to choose between him turally interdependent. and Metzger-Richardson. Kaufmann, Our first question was, “Have the de- Sr., decided to go ahead with the house flections stopped, or are they still grow- as originally planned. ing?” Using an instrument called a wa- Still, the house’s owner remained ter level, we took height readings at concerned about the tilting of the ter- more than 30 locations and attempted races, so he commissioned a surveyor to to relate them to the survey readings measure the deflections on a regular ba- done earlier. Our measurements showed sis by recording the elevations of the that the edge of the west terrace had tops of the parapet walls. This was sagged by as much as 146 millimeters done from 1941 until 1955, when Kauf- and the edge of the east terrace by as mann, Sr., died. In 1963 Kaufmann, jr., much as 184 millimeters. The deflec- presented the house to the Western Penn- tion of the south end of the master bed- sylvania Conservancy. Between 1955 room terrace was about 114 millime- and the time our firm was retained in ters. We then installed electronic moni- 1995, only one or two random mea- tors to measure very small movements surements of the terraces’ deflections of the terraces and changes in the width were recorded. of the cracks in the terrace’s parapets. The results over more than one and a Engineers as Detectives half years, corrected for daily and sea- sonal temperature variations, confirmed floors and parapets. The tests also pro- he conservancy initially asked our that the cracks were still growing and vided data on the quality of the house’s Toffice to evaluate the structural ade- the terraces sagging ever lower. concrete. The work was performed by quacy of the master bedroom terrace, The next step was to examine the GB Geotechnics of Cambridge, Eng- the part of the house that historically structure’s as-built condition to see how land. To investigate the main cantilever had the most severe visible cracks. closely it conformed to Wright’s plans. beams, the technicians had to remove Work was ongoing to repair Fallingwa- In particular, we needed to verify the several paving stones from the living ter’s facade, including the terrace’s actual number, size and location of the room floor so that they could gain ac- cracks, and the conservancy wished to reinforcing bars in the cantilever beams cess to the hollow space below. know whether it was wise to continue and other structural elements. We or- Our engineers then conducted an inde- repairing these cracks cosmetically with- ganized a program of nondestructive pendent structural analysis of the house. out first performing a structural review evaluation, employing instruments that Metzger-Richardson had done such an and, if necessary, repairs. We soon real- used impulse radar, ultrasonic pulses analysis in 1936 and 1937, but we OSS R ized that we had to broaden our inves- and high-resolution magnetic detection wanted to make our own determina- Y ARR tigation to include the living room be- to plumb the interiors of the beams, tion of how the structure functioned. B

92 Scientific American September 2000 The Plan to Save Fallingwater Copyright 2000 Scientific American, Inc. FRANK LLOYD WRIGHT’S DESIGN for Fallingwater combined beams from the ground and support the horizontal cantilever beams (red). and parapets made of reinforced concrete with floors and walls made The beams are connected to one another by concrete joists (purple). of sandstone (opposite page). A cutaway of the portion of the house The steel window mullions (blue) embedded in the first-floor parapet overhanging Bear Run (below) shows the concrete bolsters that rise help support the concrete joists (yellow) of the second-floor terrace.

WINDOW MULLIONS

MAIN CANTILEVER BEAMS

Using a computer model of Fallingwa- If our computer model predicted more than 283 megapascals (41,000 ter, we tested three hypotheses: that the stresses that were significantly higher pounds per square inch). master bedroom terrace can support it- than the yield strength of the steel or First, we tested the hypothesis that self through cantilever action; that the concrete, we knew that some of our as- the master bedroom terrace could sup- living room is a self-supporting can- sumptions had to be incorrect, because port itself through cantilever action. If tilever; and that the living room sup- such overstressing would have resulted this were the case, our calculations re- ports both itself and the master bed- in the immediate collapse of Fallingwa- vealed that the stress in the reinforcing room terrace. For each scenario we cal- ter. Tests of the house’s concrete indi- bars in the terrace’s parapet would be culated the bending moments that cated an in situ strength of about 34 1,195 megapascals, or more than four would be caused by the dead load of megapascals (5,000 pounds per square times the steel’s yield strength. This sce- the house. Then we calculated the re- inch). We also recovered a small piece nario is therefore not possible. Next, sulting stresses in the steel and concrete of reinforcing steel from the building we examined whether the living room of the supporting beams, as well as the and sent it to a metallurgical laborato- is a self-supporting cantilever. Our amount of deflection that these loads ry; the results of the mechanical analy- analysis indicated that the weight of the would induce. sis showed a yield strength of slightly living room alone would induce tolera-

The Plan to Save Fallingwater Scientific American September 2000 93 Copyright 2000 Scientific American, Inc. TEMPORARY SHORING installed in 1997 ensures that Fallingwater will not collapse before permanent repairs can be made to the structure. A line of steel columns and girders rising from the streambed of Bear Run supports the concrete underside of the house’s cantilevered first floor.

Thus, in 1997 workers installed a rel- atively unobtrusive line of steel columns and girders rising from the streambed of Bear Run to the underside of the first floor [see illustration at left]. In addi- tion, they also shored a portion of the streambed itself, the jutting sandstone ledge over which Bear Run cascades. The ledge was braced with pipe struts in a cave behind the waterfall. The tem- porary shoring, which ensures the safe- ty of the tourists who continue to visit the house, will remain in place until the ble stresses: a maximum of 152 mega- the maximum stress induced in them permanent repairs are completed. pascals in the steel of the main can- would be 64 megapascals, which is well From the analysis of existing stresses, tilever beams and 16 megapascals in below their allowable strength of 112 we determined that three of the four the concrete. We knew, however, from megapascals (assuming that they are cantilever beams below the living room the failure of the first hypothesis, that braced by the concrete of the living need reinforcing. (The fourth beam, the the living room does not stand alone— room parapet). What is more, it is this easternmost one, does not require inter- it has to support the master bedroom extra weight supported by the mullions vention because it is already propped up terrace as well. If we assume that the that has raised the stresses on the main by a steel strut that is part of the railing living room is propping up the terrace cantilever beams to critical levels. Al- for the stairway that goes down to the by means of the T-shaped window mul- though we cannot know for certain stream.) Practically, there is only one lions at its south end, the calculations what led to this design flaw, the struc- method that can provide sufficient rein- predict stresses of 288 megapascals in tural evidence suggests a possible chain forcement without altering Fallingwater’s the steel of the main cantilever beams of events. According to this scenario, outward appearance. This method in- and 30 megapascals in the concrete. when Wright’s engineers realized that volves post-tensioning the main beams— These stresses are at critical levels—they the master bedroom terrace could not that is, connecting them to steel cables are just about equal to the yield strengths support itself, they redesigned the win- and using the tension in the cables to re- of the materials. dow mullions to carry some of the lieve the stress in the beams. Furthermore, the deflections that load. The engineers failed, however, to The repair scheme calls for the stone would be caused by these stresses close- redesign the main cantilever beams to floor of the living room to be removed ly match the observed tilting of Falling- support the extra weight. temporarily. This will allow access to the water’s terraces. Our computer results three main cantilever beams from above. yielded only the initial deflections and Fixing Fallingwater At the south end of each beam—the end did not allow for the subsequent shrink- jutting over Bear Run—we will attach age and creep of the concrete in the hen the conservancy’s trustees re- concrete blocks to both sides of the main cantilever beams. Shrinkage occurs Wceived the results of our analysis beam [see left illustration on opposite as concrete hardens; creep is the continu- in May 1996, they were naturally con- page]. Into each block we will insert a ing contraction that takes place as con- cerned. Our study indicated that the hollow duct with an inside diameter of crete is subjected to a constant compres- stresses in Fallingwater’s main canti- 6.35 millimeters. The ducts will run sion load over time. The amounts of lever beams were great enough to raise alongside the beams, angling upward shrinkage and creep depend on many questions about the house’s safety. The and extending through holes drilled in factors, including the quantity of the re- trustees decided to commence the de- the concrete joists. We will also drill inforcing steel and the quality of the sign of permanent repairs. We advised holes in the exterior of the south para- concrete. When we added these factors them that during the construction pet so that high-strength post-tensioning into our calculations, the expected de- phase it would be necessary to shore the cable can be threaded through the ducts. flections of the east and west terraces ends of the main beams while repairs The cables will be anchored at the turned out to be surprisingly close to were under way. Because the house north end of each beam. At the south the actual conditions. would ultimately have to be shored, the end, we will tighten the cables from the The logical conclusion is that the T- trustees wisely chose to do it immedi- outside using a hydraulic jack. The shaped window mullions do, indeed, ately and thereby eliminate the fear that tightened cables will be rigged in such a support the weight of the master bed- the building might collapse or that way that they exert a positive bending room terrace. Our engineers confirmed some structural element might fail be- moment on each cantilever beam. This that the mullions could handle the load: fore repairs could be made. positive moment will essentially coun-

94 Scientific American September 2000 The Plan to Save Fallingwater Copyright 2000 Scientific American, Inc. PLANNED REPAIRS involve relieving the stresses in the cantilever beams through the creative use of post-tensioning. Steel cables will be rigged on both sides of each beam, anchored in concrete blocks at- tached to the beam’s ends (left).The cables will then be tightened from the outside us- ing a hydraulic jack.The tension in the ca- bles will exert a positive bending moment on the beam, counteracting the negative moment caused by cantilever action.A sec- tion of one cantilever beam beneath the POSITIVE living room floor (below) has already been BENDING exposed to allow engineers to inspect it. MOMENT

CANTILEVER BEAM

POST-TENSIONING CABLE

ANCHORAGE BLOCK

teract the negative moment caused by We anticipate that the structure will also includes the waterproofing of the cantilever action, lowering the tension lift slightly off the temporary shoring entire house. This work will be super- in the upper part of the beam and the when the post-tensioning forces are ap- vised by Wank Adams Slavin Architects compression in the lower part. We will plied, but we do not intend to restore of New York City. The conservancy is also connect post-tensioning cables to the cantilever beams to their original also upgrading the water supply and the parapet edge beams in the east and horizontal level. We will fill the cracks sanitary facilities at the property. west terraces to relieve the stresses in in the tops of the beams prior to jack- The strengthening of Fallingwater’s

) those beams. On the second floor, we ing to limit the amount of upward cantilever beams will guarantee the ation plan to reinforce the overstressed con- movement. When the repairs are com- structural stability of the house for illustr crete joist just above the steel window pleted, the terraces will still be tilted, years to come. Moreover, the plan sta- OSS (

R mullions, either by bolting steel chan- but they will not sag any further. The bilizes the house without the need for Y nels to each side of the joist or by bond- deflected structure will illustrate the his- permanent props rising from Bear Run. ARR B

); ing carbon-fiber plates to it. At the con- tory of the building and the problems Thanks to state-of-the-art technology, aphs

gr clusion of the project, we will patch that it has encountered over its lifetime. we can preserve the most striking archi- o and paint the holes in the parapet, re- The repairs are scheduled to take tectural element of Fallingwater, its can- phot place the stone floor and remove the place during the winter of 2001–02 as tilevered terraces stretching gracefully ANCY ( V temporary shoring. part of a larger restoration project that over the rushing stream. SA ONSER ANIA C V The Author Further Information ENNSYL ROBERT SILMAN is president of Robert Silman Associates, P.C., Consult- Fallingwater: A Frank Lloyd Wright Country House. ing Engineers, a structural engineering firm with offices in New York City and Edgar Kaufmann, jr. Abbeville Press, 1986. WESTERN P Washington, D.C. Since its founding in 1966, the firm has worked on more Frank Lloyd Wright’s Fallingwater: The House than 7,300 projects, including the renovation of 300 designated landmark and its History. Second edition. Donald Hoffmann. TESY OF buildings. Among these are the U.S. Supreme Court Building, Carnegie Hall, Dover Publications, 1993. OUR C George Washington’s mansion at Mount Vernon and six other Frank Lloyd Models of Fallingwater are available at www.greatbuildings. K/ Wright buildings. Silman received a bachelor’s degree in government from Cor- com/buildings/Fallingwater.html on the World Wide Web. USCHA

R nell University and bachelor’s and master’s degrees in civil engineering from Information on visiting Fallingwater is available at www. .

T P New York University. He is a fellow of the American Society of Civil Engineers. paconserve.org on the World Wide Web. OBER R

The Plan to Save Fallingwater Scientific American September 2000 95 Copyright 2000 Scientific American, Inc.