Structural Engineering International

ISSN: 1016-8664 (Print) 1683-0350 (Online) Journal homepage: https://www.tandfonline.com/loi/tsei20

Once upon a Time in : The Tale of the Morandi Bridge

Gian Michele Calvi, Matteo Moratti, Gerard J. O'Reilly, Nicola Scattarreggia, Ricardo Monteiro, Daniele Malomo, Paolo Martino Calvi & Rui Pinho

To cite this article: Gian Michele Calvi, Matteo Moratti, Gerard J. O'Reilly, Nicola Scattarreggia, Ricardo Monteiro, Daniele Malomo, Paolo Martino Calvi & Rui Pinho (2019) Once upon a Time in Italy: The Tale of the Morandi Bridge, Structural Engineering International, 29:2, 198-217, DOI: 10.1080/10168664.2018.1558033 To link to this article: https://doi.org/10.1080/10168664.2018.1558033

Published online: 20 Dec 2018.

Submit your article to this journal

Article views: 1087

View related articles

View Crossmark data

Citing articles: 10 View citing articles

Full Terms & Conditions of access and use can be found at https://www.tandfonline.com/action/journalInformation?journalCode=tsei20 Once upon a Time in Italy: The Tale of the Morandi Bridge

Gian Michele Calvi , Eucentre Foundation, Pavia, Italy; Matteo Moratti, Studio Calvi Ltd, Pavia, Italy; Gerard J. O’Reilly , IUSS, Pavia, Italy; Nicola Scattarreggia , IUSS, Pavia, Italy; Ricardo Monteiro , IUSS, Pavia, Italy; Daniele Malomo , Mosayk Ltd, Pavia, Italy; Paolo Martino Calvi, University of Washington, Seattle, USA; Rui Pinho , University of Pavia, Italy. Contact: [email protected] DOI: 10.1080/10168664.2018.1558033

Abstract the 1960s that most European countries constructed the backbone of the modern On 14 August 2018 at 11:35 AM, a relevant portion (about 243 m) of the viaduct over freeway system. In Italy, the 760 km the Polcevera river in collapsed, killing43people.Thebridgewasdesignedin “Autostrada del Sole” freeway system the early 1960s by Riccardo Morandi, a well-known Italian engineer, and opened to linking Milan to and Naples was the public in 1967. The collapsed part of the bridge essentially comprised an designed and built between 1956 and individual self-standing structure spanning 171 m and two simply-supported 1964, coinciding with the American connecting Gerber beam systems, each spanning 36 m from the self-standing “greatest decade”. Regardless of its rela- structure to the adjacent portions of the bridge. This paper aims to reminisce the tively small length, this freeway rep- complete story of the bridge, from the Italian construction boom in the 1960s to resented an achievement of sorts, due some of the issues that soon arose thereafter: the strengthening intervention in the to the ingenuity required for its con- fi 1990s, the subsequent structural monitoring and, nally, the strengthening project ception and construction, as a result of never brought to fruition. Potential reasons for the collapse are discussed, together the challenging topography of the with some of the possible inadequacies of the bridge, its maintenance and loading Italian territory, particularly in the fl fi history based on critical re ection, comparison with speci c features of bridge mountainous area between Bologna construction practice today and results obtained using numerical models with and . Indeed, a total of 853 fi fi different levels of re nement. Since the entire matter (speci cally the debris) was bridges and viaducts (without consider- fi considered classi ed by the investigating magistrate in the immediate aftermath of ing 572 overpasses) and 38 tunnels the bridge collapse, this work is based entirely on publicly available material. needed to be constructed. Keywords: Morandi Bridge; structural collapse; forensic engineering; AEM There is little need for any detailed analy- modelling sis to state that a significant fraction of both North American and European infrastructure has reached or is reaching its nominal design life, requiring the allo- Introduction the programme and opposed the cation of relevant resources for their reduction of a temporary USD$0.01 assessment, repair and upgrading. In par- Construction of Road gas tax per gallon (≈ 3.8 litres) estab- ticular, entire inventories of structures Infrastructure in the 1960s lished to fund the Interstate Program. areinneedofassessmenttoallow By the end of 1966, some 29 000 km It is commonly stated, and easy to rational prioritisation in the allocation of highways had been completed, with verify, that the 1960s were an extraordi- of limited resources available. This a total cost of about $25 billion. Accord- nary time for the construction of free- implies the issue of rapid assessment ing to a document submitted to the US ways. The era of great bridge methods to perform an initial screening, Congress in 1965,3 the complete system construction had started much earlier, followed by the application of more (which required additional funding esti- fi as masterfully described by Petrovsky1 re ned approaches (with a proportional mated to be about $20 billion) would with main reference to the US, but acquisition of more data) to a limited have included “12,957 interchanges bridges were perceived as standalone number of cases and possibly the requiring 22,252 individual structures, masterpieces to cross rivers or straits, implementation of effective and well- as well as 20,748 other highway grade- fi rather than part of a roadway system. focused monitoring systems in speci c separation structures, 4,361 railroad It was on 29 June 1956 that President situations. The collapse of the bridge in grade separations, and 14,806 other Eisenhower signed the Federal-Aid Genoa has attracted media attention bridges and tunnels”. 4 Highway Act in the US, initiating the worldwide, with The New York Times “Greatest Decade”,2,3 or the construc- In Europe, the construction of freeways recently summarising that: tion of the Interstate System that orig- started much earlier with the initial but inally included around 60 000 km of limited experience in Italy in the 1920s . in France, the national non-conceded roadways. When the Eisenhower and a massive programme in Germany highway system comprising 12 000 administration ended in January 1961, in the late 1930s, where some 4000 km bridges is in a state of chronic under- about one fourth of this system had of roads were built between 1935 and investment, with 7% having damage been opened to traffic. With an 1940, which inspired President Eisen- that could eventually result in col- average construction of about 5000 km hower to say: “Germany had made me lapse if not addressed; per year, President Eisenhower’ssuc- see the wisdom of broader ribbons . in Germany, of the 39 621 bridges cessor, President Kennedy, refused to across the land”.2 However, it was after monitored by the Federal Govern- follow the suggestions of cutting back the Second World War and mainly in ment, 10.6% are in a condition that

This article has been republished with minor changes. These changes do not impact the academic content of the article.

198 Scientific Paper Structural Engineering International Nr. 2/2019 is not satisfactory and 1.8% are in problems related to creep, temperature raced across the Pacific Ocean “inadequate” condition; variations, strand relaxation, redistri- devastating Hawaii, Japan and the bution effects in statically indetermi- Philippines. to name but a few, with similar examples nate structures, nonlinear and . On 26 July 1963, a much smaller reported for other European countries. ultimate response, were only intui- earthquake (Mw = 6.1) hit Skopje in tively considered, and sometimes Macedonia inducing more than 1000 simply neglected. It was only years victims and leaving more than 200 Design of Bridges and Pre-stressed later, with the development of dedi- 000 people homeless. About 8.5% Structures in Italy in the 1960s cated software, that it became possible of the buildings were destroyed, to model most of such complex time- 34% had to be demolished and 36% In the aftermath of the Second World dependent effects. required major repairs,9 whereas the War, bridge design was essentially retra- 15th century stone bridge over the cing the experiences of the 1930s and In the context briefly depicted above, river Vardar was not reported to the dominant structural system was Riccardo Morandi was a very unique have suffered major damage. In the probably still the arch.5 Concrete arch individual and the bridge over the Pol- aftermath, a school building was bridges had recurrent spans in the cevera river in Genoa was a very base isolated using rubber bearings, range of 60–80 m and even in the con- unique design case, reflecting each for the first time in the world.10 struction of the Autostrada del Sole,a other. The bridge was cable-stayed . On 27 March 1964, a M = 9.2 (the span of 100 m was regarded as a limit with single post-tensioned concrete W largest ever recorded in North for standard practice. Trafficloads stays and spans exceeding 200 m. The America) earthquake occurred in were not significantly smaller with deck was temporarily pre-stressed Alaska, not far from Anchorage. Per- respect to those considered today, with during construction and locally post- manent ground displacements in the the exception of the maximum uni- tensioned in its final configuration. range of 9 m were recorded. The formly distributed load, which was set The connecting simply supported Seward Highway was devastated at 4 kN/m2, against the 6 kN/m2 typi- spans were made of 36 m precast pre- and severe damage to bridges was cally adopted nowadays. Cable-stayed stressed Gerber beams. It is evident reported, such as the span of the bridges were seldom considered, prob- that the definition of “cable-stayed Million Dollar Bridge that slipped ably because of the circular problem bridge” refers today to quite different off its pier due to soil liquefaction.11 of absorbing the axial force component structural configurations,8 with a large in the deck that tended to increase its number of stays used also as progress- cross-section size, which in turn would ive supports for the deck, to be kept In Italy, one of the most tragic cata- imply a higher self-weight, and hence as light as possible. The single concrete strophes was the Vajont dam disaster increase the axial force. The advent of stays used in the so-called “Morandi in 1963. A relevant portion of the light composite deck sections that bridge” did not garner much popular- mountain Toc (about 260 million would allow long span cable-stayed ity in the following decades, with essen- cubic metres) slid into the reservoir, bridges was still a few decades away. tially no followers. All in all, however, causing a flood wave that killed Pre-stressed concrete had been con- it was an absolute masterpiece but approximately 2000 people in the ceived and experimented starting in also a daring combination of advanced towns further down the valley. The the late twenties, mainly by Freyssinet, and relatively new technology 262 m tall concrete dam remained but even enthusiastic and brilliant assembled in a clear and relatively essentially undamaged, inspiring dis- researchers and engineers, like Torroja simple structural scheme. Neverthe- cussions about an engineering master- in Spain, Dischinger in Germany and less, as would be unveiled in the sub- piece built in the wrong place.12 Colonnetti in Italy, faced a then hardly sequent years, it possessed a high It is even more surprising to learn that surmountable obstacle in the difficulty potential for issues raising from some seismic assessment of most bridges of producing high strength steel, of the features mentioned above. built in absence of any seismic design proven to be a fundamental prerequi- code (in Italy the first one was site for its practical implementation. released in 2003, with the OPCM Earthquakes and Disasters in the After the Second World War, things 327413 has not been performed to 1960s changed at an impressive pace, particu- date, at least not in regions of rela- larly in Italy. While it is hard to find It may be surprising to realise that not tively modest seismic hazard. Like- pre-stressed concrete even mentioned much attention was paid to potential wise, such an assessment has also not in the standard text book used at the extreme actions on bridges, such as been carried out even in outstanding Politecnico di Milano to teach bridge those generated by seismic loading, cases like the one considered in this design in the 1950s,6 between 1945 even though the 1960s were not a study, which exposure, in terms of and 1960 a number of design manuals peaceful time from the point of view consequences of traffic interruption were published and a number of of ground shaking: or collapse, is extremely high. The patents were imported or deposited OPCM 327413 did indeed foresee a on elastic coaction, cable anchoring, . On 22 May 1960, the Great Chilean compulsory verification of the seismic etc. Pre-stressed concrete technology earthquake hit a region some safety of infrastructures which func- would have still been regarded as 570 km south of Santiago. With a tionality was fundamental for the being in its infancy but nevertheless, magnitude of 9.5, this is to date the purpose of civil protection or for gifted designers like Levi, Cestelli- largest energy ever released by a which collapse would imply relevant Guidi, Pizzetti, Oberti and Zorzi recorded event. The subsequent consequences, but in the (fifteen) immediately began applying it to rela- tsunami generated waves up to 25 years subsequent to its release, tively long span bridges.7 Solutions to metres on the Chilean coast and measures and provisions have not

Structural Engineering International Nr. 2/2019 Scientific Paper 199 been effective in providing specific villain was the Kármán vortices, Considering the above (i.e. that type b) time constraints for assessment and named after the investigator himself, collapses of reinforced concrete bridges possible strengthening. or the whirlpool of air that were shed have been predominant in the past, it is in the wake behind the moving model no surprise that in the case of the Bridge Collapses and thus buffeted it”. Morandi bridge, media and public opinion immediately focused on main- Recent reinforced concrete bridge col- Bridge collapses are reported all over tenance and deterioration. However, lapses seem to be often related to shear history and this is not the place for a it is important to also explore whether problems or loss of post-tensioning, recapitulation on the subject. It is, the bridge was flawed by some “original essentially ascribable to category b) however, of interest to note that fail- sin”, not with the aim of establishing above, and more rarely to c). Examples ures can be attributed to three main and assigning fault or blame, but of such cases include15: categories: rather to examine possible reasons for the Morandi bridge collapse using (a) Unexpected external actions, poss- (1) The Laval overpass in Quebec, robust engineering rationale. ibly due to both natural or anthropo- Canada, which failed in shear genic catastrophes (e.g. earthquakes, killing five in 2006.16 The Laval floods or sudden impacts); collapse led to an extensive review of 135 bridge structures in The Morandi Bridge: Design (b) Deterioration of mechanical prop- Quebec, resulting in 28 bridges and Construction erties, possibly due to corrosion of being demolished and further 25 17 Description of the Bridge the reinforcement or concrete car- being repaired immediately. bonation or fatigue, sometimes (2) A highway overpass that failed on The design and construction of the also in conjunction with increased 29 October 2016 in Lecco, Italy, bridge is described in detail by – traffic load (as in the case of killing one and injuring five.18 20 Morandi himself in a long paper pub- fatigue); The bridge was a concrete struc- lished in 196722 and in an unpublished (c) Inadequate original design or con- ture with a drop-in-span that, report.23 Whilst those documents struction, possibly related to from publicly available videos, refer to the entire bridge structure unknown structural effects, some- appears to have suffered a brittle (Fig. 1), the attention here will focus times related to dynamic actions. failure near the drop-in span on the three “balanced systems”, support ledge; the detailed investi- shown in Fig. 2, that constitute the A dramatic and well-known example of gation of the probable causes of large span portions of the viaduct. type c) is the Tacoma Narrows Bridge collapse is currently ongoing. Each of the 12 support points of the that collapsed soon after completion (3) A post-tensioned viaduct near bridge was numbered sequentially in 1940.1 The advisory engineer to the Fossano, Italy, on 19 April 2017. from the Savona side shown in Fig. 1, bond purchaser, Theodore Condron, Two police officers reported with piers 9, 10 and 11 comprising the expressed concerns about its horizontal having been underneath the aforementioned balanced systems. It slenderness (1:72, much smaller than bridge as it began to collapse. was pier number 9 that collapsed on those of all existing suspension The collapsing bridge span fell on 14 August 2018. Above the foundation, bridges; that of the conceptually and destroyed the officers’ which is not discussed here, each similar Golden Gate Bridge, completed vehicle; however, since the failure balanced system comprises the follow- three years earlier and featuring a much reportedly occurred over several ing main elements: longer central span (4200 versus seconds, the officers were able to 2800 ft), was about 1:47). The repu- escape unharmed.20 a) A pier with eight inclined struts (with tation and the self-confidence of the (4) The bridge “Santo Stefano”, near cross-section varying between 4.5 × main designer, Leon Moisseff, pre- Messina, which collapsed on 23 1.2–2.0 × 1.2 m) that props the deck vailed against what were then deemed April 1999. This case is less over a distance of about 42 m. unjustified qualms and the bridge col- known, possibly because no b) An antenna with two A-shaped lapsed four months after completion, casualties were involved, but structures (element cross-section when facing a wind of about 40 knots. deserves to be mentioned here varying between 4.5 × 0.9–2.0 × The explanation was found by von because it had been designed by 3.0 m) that converge about 45 m Kármán, who describes its experiment Morandi and the deck (with span above the deck level. with a van and a model of the bridge 78 m and a box section) was post- c) A main deck with a five-sector box in his magniloquent autobiography.14 tensioned with the same system section of depth variable between As reported by Petrovsky1 “The employed in the Genoa bridge.21 4.5 and 1.8 m, an upper and lower

Fig.1: Schematic of the piers and distances between each support of the Morandi Bridge, with the three balanced systems shown to pass over residential areas, numerous transportation lines and the Polcevera river (although not shown, the area between piers 1 and 8 is also heavily industrialised) (Units: m)

200 Scientific Paper Structural Engineering International Nr. 2/2019 slab 160 mm thick, and six deep parts of the bridge. Each Gerber portion of the structure by means of webs with thickness varying supported span was 36 m long post-tensioned cables laying on top of between 180 and 300 mm. In its and comprised six precast pre- the deck and slightly inclined by final configuration, the deck of the stressed beams, with a variable means of steel supports (2.1 m tall) balanced system 9 was 172 m long depth equal to 2.20 m at mid located in correspondence with the and supported at four locations: span, sitting on Gerber saddles inclined struts of the pier. two of these from underneath the protruding from the main deck. deck, provided by the pier inclined Following the progressive connection struts at the aforementioned and post-tensioning of the stays, the spacing of 42 m, and the other two Construction Process temporary cables were progressively from above, provided by the cable removed, finally obtaining a five-span Whilst the construction of the pier and stays at a distance of 152 m. There continuous deck compressed by the antenna is reported to have followed was therefore no connection horizontal component of the cable traditional methods (this is also between the deck and the stays’ force in the three central spans. evident from the photos taken during antenna. Two 10 m cantilevers According to the designer words “at the construction), the completion of completed the deck length. this stage the deck is essentially the main deck was inspired by a d) Four transverse link girders, con- lacking any longitudinal reinforce- rather original and ingenious expedi- necting stays and pier trusses to ment, with the exception of the end ent. Indeed, the deck was erected the deck. cantilever parts and of the areas next through a segmental construction e) Four cable stays, hanging from the to the intermediate supports”.22 The process departing from each side of antenna’s top and intersecting the conclusive construction operation was the antenna centreline, and each deck at an angle of about 30°. termed the “homogenisation of the segment (of a maximum length of f) Two simply-supported Gerber system” by Morandi and described as 5.5 m, which was the capacity of the beam spans connecting the the casting of concrete shells around launching girder) was temporarily con- balanced system to the adjacent the steel cable stays, their post- nected to the previously constructed

Fig. 2: Longitudinal and transversal section of one of the “balanced systems” that constituted the large span portions of the viaduct (Units: m)

Structural Engineering International Nr. 2/2019 Scientific Paper 201 Fig. 3: Details of the end part of a 16 ½ inch strands post-tensioning cable according to the Morandi System,24 re-drawn (Units: mm)

compression, while still not bonded to configuration upon mounting the 36 m responded elastically to any action the stays, and the final “usual injec- simply supported Gerber beams and (traffic, temperature and wind, tion” of all ducts with the definitive completing the dead weight on the whereas no mention is made to earth- connection between cable stays and entire deck. It is evident from descrip- quakes). In addition, the stays con- deck.22 tion and construction photos that this crete would have always been in phase followed the casting of the con- compression (therefore not suscep- The deck extremities were deformed crete stays and the cable injection. tible to cracking and consequent cor- upwards by appropriately tensioning According to Morandi, in its final con- rosion potential) and stiffer (thus less the cable stays, to obtain a straight figuration the bridge would have sensitive to fatigue problems and less

Fig. 4: Geometry and reinforcement (top) in addition to three cross-sections (bottom) of the main deck (Units: m)

202 Scientific Paper Structural Engineering International Nr. 2/2019 prone to deck rotation and to horizon- Pier and Antenna hanging out of the stays connection; tal displacement of the antenna tip). The high vertical force, combined these cables do not seem to be with the “balancing” of the system related to the aforementioned M5 and the relatively low live loads patented system, where only ½ inch strands made of seven Ø7 mm wires with respect to self-weight and dead load (less than 20%, with reference are described. In the central region, The Morandi Pre-compression =4849 mm2) to the deck only), rendered pier and 6 similar cables (total Asp System were located at the bottom slab of antenna elements as members essen- the box deck section, in correspon- Morandi developed and patented a pre- tially compressed. As a consequence, dence to each beam. The upper and compression system (M5, described in there was no need for reinforcement 24 lower pre-compression cables do not detail in based on seven-wire strands to absorb tensile forces and, consist- appear to be overlapping, but rather with the following characteristics: ently with his vision, Morandi used leaving limited portions of the deck a minimum reinforcement level. In reinforced by the ordinary reinforce- Nominal 12.7 mm (½ inch) general, this minimum seems to be diameter ment alone. setintherangeof0.3%ofthecon- Nominal section 92.90 mm2 crete section. For example, at the From a flexural point of view, base of the antenna, in a section of simple hand calculations indicate that Minimum 163 kN (i.e. 1758 4.5 × 0.9 m, the steel reinforcement the capacity of the deck section is ade- strength MPa) was 4 Ø30 mm plus 20 Ø24 mm quate, without much conservativism, in Minimum 3.5 % (measured on a bars, resulting in a geometrical per- the regions of maximum moments, elongation base of 610 mm) centage of longitudinal reinforce- both positive and negative. However, capacity ment of ρs=0.29%. A combination in the regions next to the point in of smooth (with minimum yield which a zero-moment value is pre- strength fy,min=270 MPa) and dicted (depending on the loading con- He recommended a working stress of deformed (fy,min=440 MPa) rebars dition), the capacity is largely 1000 MPa and an initial stress of was used. At casting interruption sec- dependent on the positive effect of 1200 or 1300 MPa. In all cases, the tions, the continuity of smooth bars the compression force originated by strands were coupled in groups of relied on standard hooks only. Hori- the stays’ force horizontal component, four; an example is the end part of zontal reinforcement was provided with limited ability of absorbing any M5/16, depicted in Fig. 3—each of by Ø10 mm stirrups at 250 mm moment inversion due to unexpected the four terminal ducts and plate spacing. Whilst this reinforcement actions. It is evident that the deck holes contains four strands. Compar- could give a significant contribution would not have been able to resist ing the Morandi M5 system to to shear strength, considering the even its own weight without the modern practice (e.g. Ref. [25]) it is large depth of the section, the con- restraining action provided by the striking how tight the strands were crete was essentially unconfined, cable; for this reason, during the con- located inside the duct. In modern according to modern concrete detail- struction phases when the stays were practice, the void section left for duct ing standards. not yet present or active, the presence injection is around 50% larger than of the temporary cables, later in the Morandi case, though varying Main Deck removed, was essential. with the number of strands. This is The geometry of the deck section has Shear reinforcement was provided in a even more surprising considering already been briefly described above, rather customary fashion, with varying that today’s injection materials are with its reinforcement being now numbers of stirrups and different far more fluid than those used some herein discussed. Morandi had noted diameters. In regions close to the sixty years ago, meaning their pen- that some parts of the deck were stay connections, it appears that etration into the duct is much easier. essentially lacking longitudinal Ø14 mm and Ø8 mm stirrups were It may be thus concluded that the reinforcement, a statement that provided at a spacing of 200 mm. In ducts used in the case of the Morandi seems indeed aligned with the actual other parts, two Ø12 mm stirrups bridge were essentially impossible to reinforcement quantities employed, were provided at a spacing of be injected and it is thus difficult to especially if compared to today’s 250 mm. Again, according to simple understand the meaning of “the standard practice. From the original hand calculations, the capacity seems usual injection of cement mortar will drawings, reproduced in Fig. 4,it to be in excess of the expected be executed”.23 appears that the continuous demand. From some descriptions and photos reinforcement provided at each reproduced in the report of the Com- beam was 4 Ø24 mm and 10 Ø8 mm Transverse Link Girders mission of the Ministry of Infrastruc- bars; on a standard concrete beam tures and Transportation,26 no cement section of 0.18 × 4.5 m, a reinforce- The transverse link girders that should mortar injection appears visible. This ment percentage of ρs=0.29% is thus transmit the load from deck to stays is consistent with the considerations obtained. Further, it appears that 8 andfromdecktopierstrutsarenot above and may shed some light on be- pre-compression cables with 21 described in any detail, but appear to 2 haviour potentially different from the Ø7 mm wires (total Asp=6465 mm ) be hollow sections with external one that was expected in terms of cor- were located on top of the beam in dimensions in the range of 4.5 × 2.0 m. rosion protection, effects of fatigue, a region of about 12 m on each side The thickness of their concrete shell structural stiffness and localised sec- of the connection with the pier looks like being in the range of 0.5 m, tional response. inclined strut and in the cantilever with the exception of the bottom of

Structural Engineering International Nr. 2/2019 Scientific Paper 203 the stays girder, which seems thicker, about the post-tensioning force and σ in the range of 1.0 m. Since these Phase Deck connection: 12 s352 = 367 the elastic moduli ratio; (ii) time measures have been visually deduced, 1 000 kN on 352 MPa dependent effects are not considered; they should be treated with some tendons and (iii) the previously noted unlikely caution. No detail is available about Phase Post-tensioning at σs112 = 900 injection of the cable ducts (which reinforcement and pre-stressing, if 2 900 MPa on 112 MPa would imply absence of bond any. As a consequence, a possible tendons between steel and concrete). Whilst a failure initiated in these transverse σ − certain conservativism in the design is girders cannot be excluded, but is c = 8.7 undoubtedly present against a stay col- MPa thus not addressed in the present lapse (the steel tendons alone would be work. Phase Addition of σs352 = 367 able to take the entire maximum load 3 supported span and +75 = 442 at a stress of about 600 MPa, with a DL: 10 500 kN MPa safety factor of around 2.8), potential concrete cracking is indicated by the Cable Stays (assuming a ratio σ = 900 s112 possible tensile stress up to 1.6 MPa. According to the original design between elastic MPa (Fig. 5), each cable stay contained a moduli Es/Ec = 10) As such, even though the abundance total of 464 strands with nominal diam- σ = −8.7 of the steel capacity has possibly eter of ½ inch, of which 352 were c played a role in avoiding premature fi +7.5 = located rst and connected to the −1.2 MPa problems, the potential for concrete deck to bear its dead weight. Then, cracking and the absence of grouting σ the concrete section was cast and the Phase At the extreme s352 = 436 may have induced relevant variations remaining 112 strands were used to 4 condition of +28 = 464 in the bridge stiffness, as well as in post-compress it. Finally, all ducts maximum LL: 4000 MPa periods and modes of vibration. The were stated to be injected and the kN absence of injection, in particular, cables connected to the deck. The may have resulted in the following σs112 ≈ 900 design hypothesis was that the stays + MPa effects: concrete would remain in compression σ − under the application of the full dead c = 1.2 and live load, simply supported spans +2.8 = 1.6 included. The calculations that follow, MPa (1) In the case of bonded tendons, the however, do not necessarily confirm concrete may be able to sustain a The concrete stress state estimate such design assumption. tensile stress of 1.6 MPa and, shown above for Phase 4 seems to even in case of cracking, tension Considering the geometry of the deck imply that there was a need for an stiffening would contribute to and some reasonable assumptions increase in the steel strands tensile reduce the cable elongation. In regarding material weights, the force stress to compensate for the concrete case of lack of bonding, instead, in each cable stay due to the self- contribution beyond decompression. the concrete part would not weight of the deck is estimated at However, the above calculations need absorb any tensile force and the around 12 000 kN, which increases to be considered with care, for several stiffness would reduce to that of up to a total of about 22 500 kN reasons: (i) the assumptions made the tendons alone; when considering the simply sup- ported Gerber spans and adding the dead loads (DL, about 10 500 kN added). A reasonable estimate of the maximum cable force due to live loads (LL, if one considers the specifications at the time of construc- tion; today the estimate will likely be 50% higher) is in the range of 4000 kN. The concrete (Ac)and steel (As) areas are about Ac = 1 073 776 mm2 (depurated of the ducts 2 area), As352 = 32 700 mm (consider- ing the 352 ½ inch tendons originally connected to the deck) and As112 =10 404 mm2 (considering the 112 tendons used to compress the con- crete). The post-tensioning stress adopted is not clearly stated and will be assumed as σp,s = 900 MPa after losses. In this case, assuming perfect bond, the expected stress values in concrete and steel in the subsequent phases of construction may be estimated as follows: Fig. 5: Geometry and reinforcement of the cable stays (Units: m)

204 Scientific Paper Structural Engineering International Nr. 2/2019 (2) Similarly, in the case of bonded Supported Spans systems were suffering from wide- cables, a local fracture of some spread general deterioration, as well The simply-supported spans were con- wires would possibly result in the as several instances of concentrated stituted by six precast pre-stressed local opening of a visible crack, degradation”.27,28 The papers describ- Gerber beams connected by an upper without losing the compression ing the situation are focused on the slab (Fig. 6). The maximum depth of force in the remaining part of the strengthening intervention on pier 11 each beam is 2.20 m, plus the cast in stay concrete. On the contrary, in (East bound, towards Genoa) and do place slab (depth of 0.16 m). From the case of unbonded cables, a not provide much detail about the find- the available drawings, it appears that progressive reduction of the ings on corrosion and cable injection. 10 cables were used, each made up of steel section, hence of the corre- Whilst the strengthening intervention 18 Ø7 mm wires. Consequently, the sponding post-tensioning force, is of limited interest herein (though it total pre-stressing force, after losses, would not necessarily induce may shed light on the state of can be estimated in the range of cracks, but rather a global damage), the fast pace at which 6235 kN, considering a 900 MPa elongation and loss of compression decisions were taken and strengthening average stress. In the central section in the entire stay concrete; (on pier 11) was implemented, is an this would result in a pre-stressing (3) The details of the strands at the indicator of the gravity of the state of bending moment in the range of top of the pier, simply curved on deterioration. The same worries can 6235 kNm, considering the beam a saddle, not connected to cable be inferred by the decision of transfer- depth only (i.e. without the collaborat- heads, had to sustain millions of ring the entire stays capacity to new ing slab). From hand calculations, this cycles with a small flexural com- external cables, maintaining the exist- bending moment value seems to be ponent, rather than tiny oscil- ing post-tensioned elements only for very similar to that originated at mid lations of the tensile force; convenience of the strengthening work- span by the total self-weight and dead (4) The sensitivity to the aggressive flow and to favour a more progressive loads acting on each beam. This marine environment and the con- transfer of the forces. seems quite consistent with the final sequent potential corrosion of the increased depth and strength and a A photograph taken at the top of the strands became much higher. maximum potential increment of the antenna after removing the concrete acting bending moment in the range cover seems to confirm a complete of 30% due to live loads. For what con- absence of any injection and some The combination of the presence of cerns shear reinforcement, Ø10, Ø8 corrosion. In the aforementioned tensile concrete stress and absence of and Ø6 mm diameter stirrups at publications, it is also described how bond, together with the possible 200 mm spacing were provided. the emission of high-frequency deterioration of the steel tendons and impulses at one end of a cable and the consequent reduction of the com- their recording at the same place fl pression force in the stay concrete, Structural Assessment, upon re ection can be used to may have resulted in increased Strengthening and acquire data about defects in the deformability and a consequent differ- Monitoring from the 1990s cable, as well as some measure of ent distribution of shear and bending Onwards the tensile force present in the moments, leading to a decreased force strand. However, the general in the cable and increased reactions at In the early 1990s, “during maintenance impression is that once it was the pier struts and in the corresponding and repair activities, it was discovered decided to essentially replace the deck shear. that the stays of the three balanced stays, there was no interest in

Fig. 6: Geometry and reinforcement of the supported span (Units: m)

Structural Engineering International Nr. 2/2019 Scientific Paper 205 gaining a deeper and detailed under- corrosion is also reported in most Whilst it can be considered unques- standing of the actual situation in of the examined reinforcement, tionable that a stay (and, as discussed which the bridge found itself in (for with estimated extension between above, most likely the S-W one) must instance, and as a minimum, it is 10 and 30% (2015); have released its retaining capacity at very likely that a dynamic identifi- (c) In some cases, an apparent loss of a certain point of the collapse cation of the system response would post-tensioning is reported, with sequence, the question remains on have indicated the presence, or some strands free to move (2011– what might have been the triggering absence, of bonding between cable 2015); cause and the ensuing progressive and concrete, but this does not seem (d) In two cables, made visible in a sequence of events. A few alternative to have been pursued). precast beam, “no fewer than four hypotheses have been considered in ” “ this study (based mainly, though not For pier 10 it was concluded that “the wires were fractured and all wires ” exclusively, on the observations criticalness was mainly concentrated were movable by hand having thus lost any tensioning (2011); described in the third section above): in the sections stretching to the cross- fi beam at the top of the tower, and (e) In a series of dynamic identi - cation tests, seemingly inconsistent hence the interventions were limited (a) A first possibility is the simple pro- ” responses of different stays have to these areas . For pier 9, now col- gressive deterioration of the stay “ been reported and ascribed to lapsed, it was concluded that the strands, with a progressive fast stays are in better condition due to differences in progression of cor- rosion and loss of post-tensioning. elongation or a fatigue collapse the more limited corrosion present in fi that induced a migration of the both primary and secondary cables. The identi cation of four natural vibration periods in the range of tensile force in the parallel stay on Hence no intervention of any type is the opposite side of the viaduct, ” 0.70–0.82 s is reported, but no cor- scheduled . As a safety measure, with a consequent in-plane “ relation with numerical models control over time of the state of con- rotation of the deck and a concur- servation of the pre-stressing cable is and related assumptions is described (2017). rent torsional effect. In this case, assured through the installation of a there is no solid estimate of the fl system of continuous re ectometer exact time of the event, apart ” The above observations are clearly control and the conclusion was that from the small load increment due “ indicative of a general state of deterio- from an estimate of the intervention to the passage of the heavy vehicle; time limit, and considering the empiri- ration of the entire structure, and whilst they do not allow one to pre- (b) Considering that the shear cal laws which govern the speed of fi strength of the deck in the inter- degradation, the limit condition is esti- cisely de ne deterioration on an ” element by element basis, they did mediate region was largely count- mated to be around 2030 . Leaving fi ing on the compression induced aside the somewhat intriguing nature assist in the identi cation of potential weak links and critical scenarios, evalu- by the stays, another possible of this conclusion (particularly for sequence of events is given by what concerns the definition of “the ated in the subsequent sections. ” the combination of elongation of limit condition ), it appears clear that: stay cables, shear/torsion collapse Estimation of Member in the deck and consequent com- . Pier 9 was not the object of any Demands and Assessment of plete failure in the stay. Note that strengthening measure, not even Capacity the mentioned unbonding of the locally; cables may have favoured this . The attention was focused on the stays, The “balanced system” under con- failure sequence, as pointed out not on the deck, transverse links, sup- sideration is nominally symmetrical earlier; ported spans or pier elements. around both the longitudinal and trans- (c) A third hypothesis is a possible verse axis, hence, from a numerical local failure in some part of the However, the report of the Commis- analysis point of view, it is irrelevant deck, which may have led to the sion of the Ministry of Infrastructures to discuss on which side of the system cable collapse. This may have 26 and Transportation describes an the collapse has been originated. been induced, for example, by a intense structural monitoring activity However, the position of the debris loss of post-tensioning in the term- in the years that followed the above (slightly located on the north side of inal cantilever element, with a con- fi retro tting, with inspections on all the viaduct on the western side) sequent shear collapse and loss of main elements of the bridge, along seems to indicate that the south-west support for the simply supported with further repair and strengthening side stay should have been released span. Other local failures could interventions. Whilst the discussion in first. Furthermore, the only heavy be those at the stay-deck or stay- such report about alleged delays in vehicle (a red truck transporting a antenna interfaces; intervening is not of interest here, the steel coil) travelling across the bridge (d) The collapse could also have been following aspects may instead carry at the time of the event was driving originated by shear failure of the some relevance to the current work, on the south lane, towards Genoa, simply supported span, in the with reference to the implications on and the driver reported that he per- region next to the Gerber saddle, the structural modelling: ceived an initial collapse behind him. triggered by a local impulsive One can thus assume, while keeping load such as the tumble of a steel (a) The absence of any injection the aforementioned considerations on coil from a heavy vehicle’s trailer mortar is repeatedly evident both double symmetry, that the first stay to (though no evidence of this event in the stays and the deck cables; release its restraining capacity to the has so far been publicly reported). (b) Presence of oxidation and deck was the south-west one. In such a case, the resulting

206 Scientific Paper Structural Engineering International Nr. 2/2019 sudden release of the applied force ½ inch tendons. The stays are mod- At this stage, which is the final on the main deck and correspond- elled with cable elements and pro- one excluding live loads, the com- ing stay, together with a migration vided with an initial tension strain pression force in the deck, due to of the compression force to the corresponding to about 140 mm the stays action only, varies adjacent beams and the conse- total shortening, so as to obtain between 21 000 and 28 500 kN. quent torsional effects, could an approximately zero vertical dis- The addition of the live loads have caused failure in the main placement once the supported does not induce major changes, deck and in the stays. spans and the dead load are with maximum action increments added. This is the same procedure in the range of 5–20%, depending All the hypothesised mechanisms described by Morandi during the on the element considered. The above have been examined and their bridge’s actual construction. At element where the variation is relative likelihood of occurrence has this stage, the resulting axial force more relevant is the post-com- been assessed, developing first a BIM in each stay is about 12 300 kN pressed concrete part of the stay, model to ensure a common interpret- and the total vertical reaction at with an increase in the stress ation of geometry and reinforcement the pier base is about 170 MN. from 9000–12 000 kN. The corre- in the development of the different The vertical deck displacements sponding deck displacement is structural models that are described vary between 96 mm (upwards, at between 5 and 6 mm, making it below. the cantilever tip) and 120 mm essentially irrelevant. The final (downwards, approximately at tensile stress in the 352 original Estimation of Member Demands mid span); tendons varies between 650 and (3) The post-tensioned cables and their 750 MPa; Elastic Static Analysis concrete casing are added to the (5) The S-W stay is removed (without The structural analysis software stays, in parallel to those considered considering any applied live load), SAP200029 was used to develop a in the previous step. The total verti- to check if the structure would linear-elastic model that would prop- cal reaction is about 177.6 MN, con- have been able to find equilibrium erly reproduce the sequence of con- sistent with the total weight in this situation (it is noted that the struction and applied loads. Five assessed by the BIM model; removal of the S-W stay implies stages were considered, as described (4) The supported spans and the also the immediately subsequent 2 in what follows and illustrated in Fig. 7: dead loads (2.4 kN/m and loss of its S-E counterpart, given 18 kN/m for the three longitudi- that, as discussed before, the (1) The pier, the antenna and the nal lines of New Jersey barriers) cable is a continuous element central span of the deck are mod- are added. The axial force in passing over a saddle at the top elled by frame and shell elements, each stay increases to a total of of the antenna, without any local which are under their self-weight 22 600 kN (13 600 kN taken by restraint). The resulting bending only, as applied load; the original 352 tendons and moment diagrams showed, as (2) The complete deck is added 9000 kN by the post-compressed expected, a sign reversal in the together with the four stays with concrete element). The total ver- external deck ribs on the side of an area corresponding to the 352 tical reaction is about 212.4 MN. the missing stays, incompatible with the beams capacity and a high bending in the horizontal plane, acting mostly on the stays diaphragm. The axial force in the remaining stays increases to about 39 000 kN, still compatible with the cable capacity, but the vertical displacement exceeds 400 mm on one side and 1 m on the other, which is incompatible with the presence of the supported beam. At the base of the balanced system, the overturning moment in the transversal direction is about 918 MNm, with an equivalent eccentricity of 4.5 m. Consistently with the deck vertical displace- ment, a temporary situation, in which one side of the supported span collapses, has been con- sidered. In such case, a longitudi- nal base overturning moment of 307 MNm is observed, with an equivalent eccentricity of 1.55 m. Both eccentricities do not appear Fig. 7: Illustration of the four construction stages adopted for the modelling (aligned with the to induce relevant tensile forces bridge’s actual construction sequence), plus the case where the S-W stay is removed in any element.

Structural Engineering International Nr. 2/2019 Scientific Paper 207 Whilst a comparison of demand and moment demands of 88 590 and −110 response of reinforced cross-sections capacity at the main critical sections 696 kNm, respectively. subjected to shear, moment, and axial will be addressed later on, it is antici- load, was employed. Considering the Subsequently, starting with the Open- pated that no significant problem double symmetry of the deck’s cross- Sees model described above with seems to occur in any as-designed section and taking into account the each of the elements modelled as element of the bridge, considering also restrictions of the software, a simplified linear-elastic but with the possibility the described sequence of construction strategy was adopted, analysing one of defining inelastic elements if/where and loading and consequent internal single I-beam portion of the complete elastic capacity would have been restraints. Time dependent effects section, formed by a five-sector box. reached, the increment in forces for have not been considered; however, The total bending moment capacity of dynamic loading due to the appli- even if they might have induced the deck cross-section was therefore cations of a series of ground motions increased deck displacements in the considered to be six times the capacity was analysed. The mass of the bridge stays connection regions, migration of of the analysed single I-beam. elements was modelled as continuous shear forces towards the pier struts or mass and the static loading scenarios Three critical locations were con- irregularities in the road level, it is unli- included as constant gravity loads. sidered for the verification of the kely that these aspects have played an The seismic action was that stipulated bending moment capacity of the main important role in the observed collapse by the 2008 Italian design code,31 with deck cross-section, namely: (1) at the of the bridge, as further discussed in reference return periods of 120, 201, support provided by the main pier; subsequent sections of this paper. 1808 and 2475 years being adopted (2) at the connection with the cable- for the operational, damage limitation, stay; and 3) approximately at mid- Time-history Dynamic Analysis life safety and collapse prevention limit span between the pier-deck connection states, respectively. These assume a and cable-stay-deck connection (sec- Further investigation of the elastic be- bridge structure with a 100-year tions 1 and 2). Using moment-curva- haviour, considering also dynamic nominal life and the maximum impor- ture analysis, both positive and analysis, was carried out using a tance class (IV) that amplifies negative bending moment capacities model developed in OpenSees.30 As nominal life by 2.0 to a reference were estimated for the deck cross- in the case of the SAP2000 model, period to 200 years, since the bridge section at the aforementioned this was initially modelled using can be classed as being of public func- locations, to consider eventual elastic elements with the aim of analys- tion and strategic importance. Soil moment demand reversal caused by ing the sectional demands along the type C, corresponding to a soil shear the different described collapse mech- bridge deck at the various construction wave velocity between 180–360 m/s, anism possibilities. The moment-curva- stages and for potential loading situ- was also considered. For the collapse ture responses of the three identified ations. The basic assumptions regard- prevention limit state, this resulted in critical sections are presented in Fig. 8. ing the geometry, material properties a peak ground acceleration of 0.184 g. and construction sequence were identi- The comparison between the bending cal to that of the SAP2000 model. Each Whilst all intensity levels were ana- moment demand and capacity in the of the four stages of construction out- lysed using a set of ten spectrum-com- main deck is illustrated in Fig. 9, from lined above was followed and similar patible accelerograms at each which it can be seen how the capacity results between the two models were intensity, only the collapse prevention foreseen by the original design is well observed, which was reassuring. For limit state results are discussed herein enough to cover the demand stemming example, at stage 2, when the 352 since it is the maximum potential from all the loading stages, as well as cables are added and tensioned to force increase that is of principal inter- the increase originated by the consider- bring the deck to a horizontal position est here. The fluctuations in cable axial ation of the seismic loading. Even in under its self-weight, an initial axial force, deck shear and moment were the case of a stay removal on one side deformation of 0.145 m was required. recorded and are discussed with of the balanced system, it is quite In fact, it was observed that had these respect to the section capacities com- clear that no flexural problem would stays not been added and the deck puted in the following sections. arise. In addition to the verification of left to work as a simply supported can- the main deck, the corresponding tilever, an end displacement of over Verification of Critical Sections bending moment profile was also esti- 1.2 m would have resulted, highlighting mated along the simply-supported Flexure Capacity of Main Deck and the need of constructing the deck seg- Supported Spans mentally with temporary restraints as described. Once the 352 cables were Following the calculation of the added and tensioned, their axial force bending moment and shear force was reported as 12 575 kN, similar to demands along the deck and simply- what had been obtained also with the supported span for the different con- SAP2000 model. Following this, the struction stages, moment and shear rest of the stay elements were included, capacity analyses at the most relevant the Gerber beams added and the live locations were carried out. Such critical loads considered. The final axial force sections were identified in correspon- observed in each cable was 22 359 kN, dence to peak demands or discontinu- which again is similar to the SAP2000 ity points (pier support or cable-stay modelling case, in addition to a connections) and the sectional analysis maximum shear demand of 12 574 kN program Response-2000,32 which cal- Fig. 8: Moment-curvature response of the and sagging and hogging bending culates the full load-deformation considered deck cross-sections

208 Scientific Paper Structural Engineering International Nr. 2/2019 Fig. 9: Comparison of the bending moment demand along the deck with respect to the section capacities computed from moment curvature analysis

Gerber beam. Fig. 10 shows this demandforthevariousloadcasesin be seen that there is a significant demand, together with the capacity, Fig. 11. As in the case of flexure, the reserve of capacity in each loading and it can again be seen that there is shear capacity of the main deck is well scenario investigated. a good degree of reserve capacity in above the maximum anticipated each loading scenario investigated. demands from both the static loading For what concerns the possible acci- and also the seismic loading, indicating dental point load, mentioned at the that it had sufficient reserve capacity. beginning of this Section, a preliminary Shear Capacity of Main Deck and assessment, described in the paragraph Supported Spans Furthermore, for the case of removal of the stay on the left-hand side, it can below, has been carried out, leaving a The shear capacity estimates are based be seen that the shear demand in the more refined analysis to future detailed on a simplified version of the Modified deck over the left pier does not exceed studies, given that not only is an impact Compression Field Theory (MCFT), the capacity, indicating that should a analysis of such a complex system and 33 formulated by Vecchio and Collins. stay be removed from the system, no the evidence very complex and time- The MCFT represents a generalised problems would be expected as a consuming, but also because, as dis- approach for modelling the behaviour result of the increased shear demand. cussed later, even a complete collapse of reinforced concrete elements sub- of the supported span would hardly With respect to the simply-supported jected to multi-axial loading conditions. result in the global collapse of the It consists of a smeared, rotating crack Gerber beam, the corresponding fi system model that treats stresses and strains shear force capacity pro le was also in a localised average sense, and estimated for different cross sections, Assuming a weight W tumbling on the allows their reorientation as a result of located at 2, 9 and 18 m distance deck from a height h, the equivalent changing load and/or material response. from the support ledge, connecting force can be estimated equating the Similar to the case of the flexural the main deck and the simply-sup- potential energy at the beginning of capacity, the shear resistance of the ported Gerber beam. The shear the event and at maximum displace- main deck was computed at various capacity is depicted, together with ment (d) of the impacted section, points and is compared with the the demand, in Fig. 12. Again, it can obtaining, assuming perfectly elastic

Fig. 10: Comparison of the flexural capacity of the simply-supported deck computed from moment-curvature analysis with the static loading and seismic demands

Structural Engineering International Nr. 2/2019 Scientific Paper 209 Fig. 11: Comparison of the shear demand along the deck with respect to the section capacities computed using modified compression field theory

response, Pe = W·h/d. In case of an capacity (in the range of hundreds of shear capacity of the ledge estimated as elastic response and of a falling height mm). The only possible events are a 3805 kN, which corresponds to a safety of the order of 1 m, this equivalent complete punching of the upper and factor of 4.6, when considering the force can be in the order of ten times lower slabs, of little interest here, and shear demand due permanent loads the tumbling weight and should be a collapse of the cantilever part of the (820 kN), but may again imply a local further amplified to consider the deck, in proximity of the Gerber failure when considering the aforemen- dynamic response of the structure, as saddle, which will produce similar tioned hypothetical accidental point a function of the ratio of the duration effects to the saddle collapse itself, dis- load acting, in addition to the permanent of the impulse and the structure cussed later. It is noted, however, that loads, at the most unfavourable location. proper period; as well known, the if such impulsive action would maximum amplification factor is equal damage the transverse link, it could Torsion Capacity of Main Deck to 2. It is noted, however, that consider- have an impact in the stay-deck con- Considering the large in-plane bending ing the shear and flexural capacities nection, something that would instead and torsion resulting from a potential depicted above (Figs. 10 and 12), it be of relevance, as shown later in this stay release, the strength of the deck results evident that a nonlinear paper. was also evaluated considering the sim- response is to be expected, implying fi A shear veri cation of the support ledge, ultaneous presence of torsional, shear damage, larger displacements and connecting the main deck and the and flexural actions. To this end, the added energy dissipation, implying simply-supported Gerber beam, was “Variable-Angle Truss Model” pro- that a correct estimate of the impulsive also carried out. The shear strength of posed by Rabbat and Collins34 load cannot be obtained by the simple the ledge was estimated using two implemented in the CSA A23.3-1435 use of the equation described above, approaches: i) strut-and-tie method, to was employed. In this model, the cross valid only in the elastic domain. ensure that no crushing of the diagonal section is idealised using four parallel A collapse induced by an impulsive struts or failure of the ties (in this case longitudinal chords, made of longitudi- load on the main deck has to be represented by the cables) would occur; nal pre-stressing steel, reinforcing bars excluded, considering its large shear and ii) interface shear transfer equations, and concrete. The chords are connected strength (in the range of 50 MN, Fig. to ensure that no interface failure would by four “walls”, consisting of diagonally 11) and considerable displacement occur. These calculations resulted in a cracked concrete and transverse

Fig. 12: Comparison of the shear capacity of the simply-supported deck computed using MCFT with the static loading and seismic demands

210 Scientific Paper Structural Engineering International Nr. 2/2019 Fig. 13: Torsional capacity, computed both under normal and post-stay removal loading conditions, at various points of the main deck reinforcement. Moment and axial forces compression forces resulting from the maximum value of ν=0.11. These acting on the cross section are resisted by in-plane bending moment. Therefore, values suggest that for both normal axial stresses that arise in the chords, a torsional failure of the deck following and seismic loading conditions, while shears and torsions acting on the the rupture of the stays on one side rep- neither the pier nor the antenna cross section are resisted by shear flows resents a plausible sequence of events. exhibit any cases of relatively large that develop in the walls. In performing loading nor would they have presented the calculations, a simplifying assump- Seismic Capacity of Pier and any alarming results had a seismic ver- fi tion was made that is the shear flow gen- Antenna i cation been examined. erated by an applied torque was In addition to the deck elements, the Summary and Preliminary assumed to distribute only along the per- forces acting in the antenna were also Conclusions imeter of the cross section, thus weight- checked to ensure that it too possessed ing only on the flanges and on the two sufficient reserve capacity for the situ- An examination of the outcomes of the most outer webs. ations examined here. For the static analyses and verifications described loading with the self-weight and above already allows one to derive a As before, the torsional capacity was live loads, the vertical force acting number of preliminary considerations. computed for a number of sections through each antenna leg is found to In general, the “balanced system”,as along the main deck, as shown in be 27 100 kN, which when considering conceived and designed, appears to Fig. 13. For what concerns the torsional the cross section to be 4.5 × 0.9 m, have had significant capacity reserves, demand, Section 4.1.1 reported that gives an axial load ratio of ν≈0.18, com- as demonstrated by the large force/ upon the removal of one of the stays, putedastheaxialloadnormalisedby moment capacity-demand ratios in the forces increased from a balanced the product of the gross cross-sectional flexure, shear and torsion mechanisms. 22 600 kN in each stay to about area and concrete compressive strength, 39 000 kN in a single stay. Considering Indeed, and more specifically, it seems taken as 37 MPa. For the case when the that the upward force in the remaining that the complete loss of a stay could whole stay is removed on the south side, stay is now unbalanced, this would be have resulted in the type of complete the forces acting through on the anticipated to translate as a torsional collapse that was observed, given that: antenna legs on the same side reduce, force in the main deck. Taking the verti- with the opposite legs being compressed cal component of this 39 000 kN axial . The flexural and shear capacities of further. The maximum compressive force as half, given the stays are inclined the deck are in the range of two or load through the antenna legs on the at approximately 30°, and multiplying more times the demand under north side increases to 32 300 kN, by the lever arm of half the deck normal loading conditions and may which gives ν=0.22 and the maximum width, taken as 18 m, the estimated even sustain the impact of a stay load on the side with the stay removed torsion induced along the main deck is removal; reduces to 17 400 kN, giving ν=0.12, about 175.5 MNm, which was reported . However, a stay removal will induce which confirms that the legs of the by the numerical model and shown in a bending moment in the plane of antenna would be expected to remain Fig. 13. In addition, the eccentricity of the deck and a torque that will be in compression despite losing a stay. the axial force in the plane of the deck above the capacity (Fig. 13); will produce an in plane bending In the case of the seismic loading, the . The live loads are only a small frac- moment, which can be estimated in axial load ratio in each antenna leg tion of the permanent loads and the range of 250 MNm, with simple for the collapse prevention limit state cannot change significantly the equilibrium consideration similar to intensity increased from the ν=0.18 stress and strain demand. those applied for the torque. The tor- reported above to a value of ν=0.21. sional demand alone, following the For the case of the pier legs, under However, it cannot be excluded that an stay removal, is far in excess of the tor- normal loading conditions the axial impact on the deck induces local sional capacity computed under normal load ratio is computed as ν=0.05 in damage and possibly attains the loading conditions (Fig. 13), and will be each leg, but when examined under capacity of one or more of the beams worsen by the additional tensile and seismic loading increases to a of the simply supported span, though

Structural Engineering International Nr. 2/2019 Scientific Paper 211 not necessarily implying a global col- lapse. Local failures may have been favoured by the combination of signifi- cant deterioration of some tendons (i.e. a significant reduction of their cross- section), combined with exceptional point loads. These considerations, and others of the same nature, guided the progressive collapse analyses presented and dis- cussed in the next section.

Fig. 14: Screenshot of the AEM model (320 000 degrees-of-freedom) Assessment and Explicit Modelling of Possible explicitly reproduced, as depicted in with a view to simplify the analyses, Collapse Mechanisms Fig. 14, including both active and no live loads have been considered at passive reinforcement. The stays this stage. The explicit representation of com- were modelled as an assembly of plete structural collapse, and corre- two different elements working in sponding formation of debris, is still parallel; beam elements to represent Scenario 1—Progressive an open challenge in numerical mod- the post-compressed concrete com- Deterioration of the Reinforcement elling. However, recent appli- 36–39 ponents, and nonlinear links to rep- in the Stays cations have shown that the resent the pre-tensioned tendons Progressively reducing the cross- Applied Element Method (AEM) (making sure that zero vertical dis- section area, as potentially induced by does appear to be able to capture ade- placements were obtained after the corrosion, of the 112 tendons providing quately the progressive failure of both addition of the supported spans and the post-compression on the S-W con- masonry, steel and RC structures. the dead load). The complete model Originally developed by Meguro and crete stay implies an equally progress- – featured 320 000 degrees-of-freedom. Tagel-Din40 42 to simulate controlled ive decrease of the stay’s stiffness and structural demolition and the impact Before the undertaking of the collapse hence progressive elongation. As dis- of blast events, it is based on the analyses, a consistency check of the cussed previously, the latter could mechanical interaction between rigid AEM model was carried out by com- induce a migration of shear and a tor- bodies connected to each other by paring the internal forces and defor- sional action in the deck that could zero-thickness interface spring layers, mations produced by the latter during potentially lead to the failure of the in which the material properties of the application of the static loads S-W stay and the consequent collapse the system are lumped. A disconti- against their SAP2000 and OpenSees of the bridge. However, this is some- nuum-based formulation therefore, counterparts. For instance, the initial thing that was not observed numeri- that renders this approach naturally tension strain to obtain zero vertical cally. Indeed, reducing the cross- suitable for representing contact, displacement after both the application section area of the post-compression impact and collision phenomena. In of dead load and the construction of tendons of the S-W stay all the way this work, the AEM-based software the supported spans predicted by ELS up to unrealistically low values, thus tool Extreme Loading for Structures43 was 148 mm (SAP2000: 140 mm, inducing significant changes on the has thus been employed with a view OpenSees: 145 mm), the recorded ver- axial stress of the 352 pre-tensioned to numerically investigate potential tical reaction at the pier base was 165 cables, did not lead to a collapse of failure mechanisms and triggering MN (SAP2000: 170 MN, OpenSees: the bridge. For instance, considering a factors that might have contributed 167.7 MN), the axial force in the 352 50% tendons cross-section area to the observed collapse of the tendons and the post-compressed con- reduction leads to only a −19 mm Morandi bridge. To this end, the influ- crete element was 20 800 kN additional vertical displacement at the ence of several parameters, including (SAP2000: 22 600 kN, OpenSees: connection between the S-W stay and corrosion-induced deterioration of 22 359 kN). The relatively minor differ- the deck (and naturally even smaller reinforcement in different locations ences reported above were expected, vertical displacements on the N-W, S- of the bridge, have been assessed given that, in addition to the conspicu- E and N-E stays-deck connections). numerically through a sensitivity ously diverse underlying numerical for- Considering instead a reduction of study. mulation, in the ELS model the 70% (see Fig. 15) leads to a mechanical interaction between RC maximum displacement of −45 mm The AEM model was assembled con- beams and pre-stressed reinforcement on the S-W side (and −17 mm N-W, sidering analogous assumptions to is explicitly accounted for (whilst it −10 mm S-E, +6 mm N-E), which is a those adopted for the elastic models had been neglected in the other two condition still far from inducing developed in SAP2000 and Open- models). collapse. Sees (including the fact that only one of the three balanced systems In what follows, a number of modelling Given that the reduction of cross- that constitutes the bridge, the one scenarios are presented and discussed section area of the 112 post-com- that collapsed, was modelled). As with a view to explore what possible pression tendons of the S-W stay for the previous models, therefore, causes could be behind the observed alone did not lead to collapse, a each structural component was collapse of the bridge, noting that, number of additional cases have been

212 Scientific Paper Structural Engineering International Nr. 2/2019 Fig. 15: Deformation induced by a 70% reduction of the cross-section area of the 112 post-compression tendons in the S-W stay modelled assuming cross-section area structural distress would have had to It is therefore concluded that no reduction also for the 352 pre-tensioned appear well in advance. reasonable level of impulsive loading cables, both in the S-W stay alone, as could cause the collapse of the bridge, before, as well as in the other three unless in combination with other pro- stays. Even if significant vertical displa- Scenario 2—Collapse Induced by blems, for example, a concurrent loss cements (up to −800 mm) were an Impulsive Load Acting on in the stay capacity. obtained (which would have induced Critical Sections noticeable progressive structural This modelling scenario explores the damage), in most of the cases the Scenario 3—Failure of the Deck- possibility of a collapse induced by bridge seems to be able to cope with the previously introduced hypothetical stay or Antenna-stay Connections them, thanks to its good capacity of case of an impulsive load acting on As depicted in Fig. 18, two scenarios are accommodating relative displacements critical sections, possibly weakened by herein considered; either a failure at the and to find different equilibrium con- some loss of post-tensioning. The interface between the S-W stay and the figurations through the exploitation of aforementioned local impulsive load antenna (possibly related to fatigue in the large over-strength present in was thus considered acting in the vicin- the tendons), or the sudden loss of con- many elements and sections, discussed ity of the support ledge, with an ulti- nection between the same S-W stay and in the previous section of this paper. mate capacity of 3400 kN being the main deck (as previously discussed, As an example, considering a cross- obtained, when local shear failure of the limited knowledge about the trans- section area reduction of 50% of both the occurs, albeit not leading to the col- verse link details cannot exclude this the S-W and S-E steel cables, in lapse of the entire supported span, as possibility). The collapse sequence (as addition to a deterioration of 70% of illustrated in Fig. 16. induced by the antenna-to-stay inter- the cross-section of the S-W 112 post- face failure) is depicted in Fig. 19; (i) a Finally, and although the simply-sup- compression tendons, leads to a S-W torsional collapse of the deck in a ported Gerber span appears to vertical displacement of −480 mm, section next to the west side of the possess sufficient strength to withstand whilst on the S-E, N-W and N-E, pier strut and the subsequent falling to the considered hypothetical accidental −240, −250 and −140 mm were the ground of the west supported impulsive load, the effect of its poten- respectively predicted. Although this span, (ii) the consequent release of the tial failure on the global dynamic case was specifically selected for maxi- S-W stay and flat collapse to the response of the bridge was nonetheless mising the deck torsional response and ground of the west deck and supported investigated, through the sudden a considerable relative vertical displa- span, (iii) the collapse of the south removal (after the application of the cement between the S-W and the N- antenna, followed by the north one, static loads) of one, and then two of W side of the bridge was observed, (iv) the collapse of the central span its six constitutive Gerber beams. In no collapse occurred. Indeed, in order when hit by the falling antenna debris. the first case, no explicit collapse of to be able to obtain an explicit collapse A very similar collapse sequence was the supported span was obtained. On of the structure, an area reduction in obtained for the case of deck-to-stay the contrary, the simultaneous the range of 60–70% of both the 112 interface failure. removal of two of the Gerber beams post-compression tendons (S-W stay) did lead to a collapse of the supported The progressive collapse sequence and the 352 pre-tensioned cables (S- span, which induced on the main described above seems to be remark- W and S-E stays) would need to be bridge system a flexural deformation ably consistent with the actual evi- introduced. It is thus concluded that producing vertical displacements at dence, as may be gathered also from whilst a progressive reduction of the connection between the S-W stay Fig. 20, where observed and predicted tendons cross-section area and related and the deck of + 160 and + 170 mm debris are compared. Such a good post-tensioning force might have been towards N-W and S-W respectively, agreement seems to lend further a con-cause of the observed collapse, whilst on the N-E and S-E sides −135 weight to the possibility that the col- it could not by itself alone be the and −145 mm. As also gathered from lapse of the bridge was indeed trig- cause of the collapse of the bridge, Fig. 17, however, such scenario does gered by a failure of the deck/antenna since conspicuous signs of significant not lead to the collapse of the bridge. interfaces of the S-W stay.

Structural Engineering International Nr. 2/2019 Scientific Paper 213 Fig. 16: Potential local shear failure in the supported span caused by an accidental impulsive loading

Fig. 17: Bridge response when one of the simply-supported Gerber spans is taken to collapse

Fig. 18: Failure of the S-W stay at the interface with antenna (left) and deck (right)

Fig. 19: Predicted collapse mechanism associated to a sudden failure of the connection between antenna and S-W stay

214 Scientific Paper Structural Engineering International Nr. 2/2019 Fig. 20: Actual vs. predicted debris extent and configuration (identical colours are used to outline corresponding observed-modelled col- lapsed segments of the bride)

Concluding Remarks: What term effects of time dependent safety margins of different sections Might Have Happened phenomena (such as creep and relax- and elements. From these verifications, ation) and the actual injection of it can be concluded that: tendons ducts and the potential conse- The introductory part of this paper . tries to outline the exciting time of quences in terms of corrosion; Morandi All elements, with no exception, had freeways booming construction in the himself raised these issues in the year ample margins of safety towards 1950s and 1960s, with the rapid intro- following the construction of the failure, considering the structure as bridge.44 With specific reference to described at the time of construction. duction of new, advanced construction . the case under scrutiny, it appears The addition of variable live loads technology and namely of large span fl pre-stressed bridge structures. In this that the tendon ducts were certainly seems to have little in uence on the daring context, the Morandi Bridge poorly injected and possibly not assessed demands, thus being an stands out as one of the most original injected at all in most cases, however, unlikely trigger of failure. . “ ” and well-devised structures. However, this regrettable situation does not An exceptional point load, acting it appears that some relevant aspects appear to have had a serious impact on a critical section of the had not been properly considered, on the collapse, unless in favouring supported span may induce local because of an insufficient level of the progression of corrosion. element collapse, particularly in pres- knowledge, or because they were ence of relevant progress of corrosion. In the main part of this study, well- . A local failure such as those men- deliberately considered minor and not known structural analysis codes were relevant, or simply because they were tioned in the previous point will not employed to model the structural extend into a global collapse (e.g. a overlooked and taken for granted. system and equally well-established Two typical examples are the long- shear failure of at least two Gerber theories were used to calculate the beams of the supported span may

Structural Engineering International Nr. 2/2019 Scientific Paper 215 induce the collapse of the span, but References [17] CBC News. Quebec to Destroy 28 Bridges, not bridge collapse). Repair Others. CBC News Montreal, 2008 Apr . The progressive reduction of the [1] Petrovsky, H. Engineers of Dreams: Great 9. Available from: http://www.cbc.ca/news/ canada/montreal/quebec-to-destroy-28-bridges- steel section and the connected con- Bridge Builders and the Spanning of America. New York: Alfred A. Knopf, Inc, 1995. repair-others-1.710558. crete compression in the stays would result in increased local displace- [2] Weingroff, RF. The greatest decade 1956– [18] BBC News. Italian bridge collapses on busy – ments and significant deck level irre- 1966 Part 1: Essential to the national interest. road in Lecco. BBC News Europe, 2016 Oct 29. Public Roads (FHWA), March/April 2006. Available from: http://www.bbc.com/news/ gularities, long before a near world-europe-37810808. collapse would have been reached. [3] Weingroff, RF. The greatest decade 1956– – . The structural over-strength is so rel- 1966 Part 2: The Battle of its life. Public [19] BBC News. Italy: Two die as motorway bridge Roads (FHWA), May/June 2006. collapses near Ancona. BBC News Europe, 2017 evant that the bridge is generally March 9. Available from: http://www.bbc.com/ compliant with a seismic verification [4] Pérez-Peña, R. After Italy collapse, Europe news/world-europe-39218467. applying the actions considered asks: how safe are our bridges? The New York Times, 2018 Aug. 21. [20] Fulloni, A. Crolla un viadotto a Fossano today. Schiacciata auto dei carabinieri. Corriere della . fi The loss of a stay seems to be the nal [5] Santarella, L, Miozzi E.PontiItalianiin Sera, 2017 April 18. Available from: http://www. cause of collapse, regardless of the Cemento Armato (3rd ed.). Milan: Hoepli, corriere.it/cronache/17_aprile_18/crolla-viadotto initiation of the collapse sequence. 1948. -fossano-schiacciata-auto-carabinieri-ae2f0300-2 . The possible initial cause of the col- [6] Stabilini, L. Ponti (III ed.). Libreria Editrice 43f-11e7-9ccc-1412672da04e.shtml. Politecnica Cesare Tamburini, Milan, 1950. lapse may thus be related to fatigue [21] Gazzetta del Sud. Edition of April 24, 1999. problems in the tendons near the [7] Iori, T, Poretti S. The golden age of “Italian [22] Morandi, R. Il viadotto sul Polcevera per style” engineering. Proceedings of the 3rd tip of the antenna, or by deterio- l’autostrada Genova-Savona. L’Industria International Congress on Construction ration of the connection between Italiana del Cemento. 1967a; XXXVII:849–872. stay and transverse link. These local History, Cottbus, 2009. [23] Morandi, R. Il viadotto del Polcevera phenomena have not been explored, [8] Calvi, GM, Sullivan TJ, Villani A. Conceptual dell’Autostrada Genova – Savona. Roma, 1967b. because of lack of detailed data. seismic design of cable-stayed bridges, J, Earthquake Eng. 2010; 14(8):1139–1171. [24] Morandi, B. La precompressione sistema Morandi M5. CESAP S.p.A, 1970. As previously stated, all these con- [9] Sozen, MA. Structural Damage Caused by the siderations are based on elaborations Skopje Earthquake of 1963: A Report to the [25] DSI. DYWIDAG bonded post-tensioning of data publicly available; potential Committee on Masonry and Reinforced systems using strands. Dywidag Systems future cross-correlation with evidence Concrete of the American Society of Civil International. Available from: https://www. fi of actual phenomena, such as corrosion Engineers and the American Concrete Institute. dywit.it/ leadmin/downloads/dywidag-emea/dsi- Urbana: University of Illinois, 1964. dywidag-bonded-post-tensioning-systems-using- or wires fracture, will allow a more strands-eu.pdf. thorough discussion of what really [10] Gjorgjiev, I, Garevski M. Replacement of fi happened. the old rubber bearings of the rst isolated build- [26] Mortellaro, AP, Ievolella G, Lombardo F, ing in the world. Proceedings of the 15 WCEE, Nuti C, Vanzi I. Relazione della Commissione Lisbon, 2012. Ispettiva Ministeriale crollo Viadotto Polcevera. Roma: Ministero delle Infrastrutture e dei [11] Brocher, TM, Filson, JR, Fuis, GS, Acknowledgements Trasporti, 2018. Haeussler, PJ, Holzer, TL, Plafker, G, Blair, JL. The authors would like to acknowledge the The 1964 Great Alaska Earthquake and [27] Martinez y Cabrera, F, Camomilla G, Pisani valuable assistance of Maria Grazia Tsunamis – A Modern Perspective and F, Marioni A. Il risanamento degli stralli del via- Accurso-Tagano in the preparation of all Enduring Legacies. U.S. Geological Survey Fact dotto Polcevera. Atti delle Giornate AICAP, the bridge drawings included in this paper. Sheet 2014–3018, 2014. Available from: https:// 581–590, 1993. In addition, the authors are also grateful dx.doi.org/10.3133/fs20143018. to Luis Alvarez, for his assistance in the [28] Martinez y Cabrera, F, Camomilla G, [12] Corriere delle Alpi. Capolavoro di ingeg- moment-curvature analysis of the bridge Donferri Mitelli M, Pisani F. Rehabilitation of neria nel posto sbagliato, 2013. Available from: deck. Finally, the assistance and collabor- the stays of the Polcevera Viaduct. http://temi.repubblica.it/corrierealpi-diga-del-va ation of the technical support staff from International Symposium on Cable Stayed jont-1963-2013-il-cinquantenario/la-costruzione – Applied Science International LLC (ASI), Bridges, Shanghai, 640 665, 1994. -della-diga/. on the use of the employed AEM soft- [29] CSI. SAP2000 Integrated Software for ware—Extreme Loading for Structures, is [13] OPCM 3274. Primi elementi in materia di Structural Analysis and Design, 2016. Available also greatly appreciated. criteri generali per la classificazione sismica del from: https://www.csiamerica.com/products/ territorio nazionale e di normative tecniche per sap2000. le costruzioni in zona sismica. G.U. of the Italian Republic, N. 108, 2003 May 8. [30] McKenna, F, Scott MH, Fenves GL. ORCID Nonlinear finite-element analysis software archi- [14] Von Kármán, T, Edson L. The Wind and tecture using object composition. J. Comput. Civ. Gian Michele Calvi http://orcid.org/ Beyond: Theodore von Kármán, Pioneer in Eng. 2010; 24(1):95–107. 0000-0002-0998-8882 Aviation and Pathfinder in Space. Boston, MA; Gerard J. O’Reilly http://orcid.org/ Little Brown, 1967. [31] NTC. Norme Tecniche per le costruzioni. Rome, 2008. 0000-0001-5497-030X [15] Calvi, PM, Proestos GT Ruggiero DM. Nicola Scattarreggia http://orcid.org/ Towards the development of direct crack-based [32] Bentz, EC. Response-2000, 2017. Available 0000-0002-3297-8068 assessment of structures. ACI SP. 2018; 328 from: http://www.ecf.utoronto.ca/~bentz/r2k.htm. – Ricardo Monteiro http://orcid.org/ (9):1 20. [33] Vecchio, FJ, Collins MP. The modified com- fi 0000-0002-2505-2996 [16] Johnson, P M, Couture A Nicolet R. Report pression eld theory for reinforced concrete Daniele Malomo http://orcid.org/ of the Commission of Inquiry into the Collapse of elements subjected to shear. ACI Journal. 1986; – 0000-0002-4518-5841 a Portion of the de la Concorde Overpass. 83(2):219 231. Rui Pinho http://orcid.org/0000- Gouvernement du Québec, 2007. ISBN 978-2- [34] Rabbat, B, Collins, MP. A variable angle 0001-6767-9036 550-50961-5. space truss model for structural concrete

216 Scientific Paper Structural Engineering International Nr. 2/2019 members subjected to complex loading. ACI SP. [38] Malomo, D, Pinho, R, Penna, A. Using the [41] Meguro, K, Tagel-Din, H. Applied 1978; 55: 547–587. applied element method to simulate the element simulation of RC structures under dynamic response of full-scale URM cyclic loading. ASCE J. Struct. Eng.. 2001; 127 [35] CSA. Design of Concrete Structures.A23.3- houses tested to collapse or near-collapse con- (11):1295–1305. 14. Mississauga: Canadian Standards Association, ditions. 16th European Conference on 2014. [42] Meguro, K, Tagel-Din, H. Applied element Earthquake Engineering, Thessaloniki, , method used for large displacement structural [36] Elshaer, A, Mostafa, H, Salem, H. 2018. analysis. J. Nat. Dis. Sci. 2002; 24(1):25–34. Progressive collapse assessment of multistory [39] Salem, H, Mohssen, S, Nishikiori, Y, reinforced concrete structures subjected to [43] ASI (Applied Science International LLC). Hosoda, A. Numerical collapse analysis of seismic actions. KSCE J. Civ. Eng. 2017; 21 Extreme Loading for Structures. Durham (NC), Tsuyagawa Bridge Damaged by Tohoku (1):184–194. 2018. tsunami. J. Perform. Const. Facil. 2016; 30 [37] Karbassi, A, Nollet, M-J. Performance- (6):04016065. [44] Morandi, R. The long-term behaviour based seismic vulnerability evaluation of of viaducts subjected to heavy traffic and [40] Meguro, K, Tagel-Din, H. Applied element masonry buildings using applied situated in an aggressive environment: The method for structural analysis: Theory and appli- element method in a nonlinear dynamic-based viaduct on the Polcevera in Genoa. IABSE cation for linear materials. Struct. Eng. Earthq. analytical procedure. Earthq. Spectra. 2013; 29 Reports of the Working Commissions, 032, Eng. 2000; 17(1):21s–35s. (2):399–426. 170–180, 1979.

Structural Engineering International Nr. 2/2019 Scientific Paper 217