Reviving the American Guastavino Vault

To view the figures and tables associated with this article, please refer to the flipbook above.

The monumental waiting room ceiling at Michigan Central Station was designed and built by the R. Guastavino Company in 1912, as part of the construction of the train station in Detroit. After the station closed permanently in 1988, the vaulted ceiling, along with the rest of the building, fell into disrepair and was nearly demolished decades later. Now recently restored as part of Ford Motor Company’s new downtown Detroit office campus, this Guastavino ceiling serves as an example of how a combination of iterative site documentation and nuanced structural analysis, using historic methods with modern computational tools, can allow for a repair strategy that minimizes impact and maximizes preservation of original structure.

The structural tile assembly that makes up the ceiling is derived from traditional timbrel or Catalan vaults and was brought to the United States and patented by Spanish builders Rafael Guastavino I Moreno and his son, Rafael Guastavino I Exposito. This type of vaulting is made up of layers of thin structural clay tiles set in overlapping courses with a Portland cement mortar and allows construction of a wide range of structures varying in both size and complexity. The design for Michigan Central Station consists of four arches spanning north to south and three sail vaults, located between the arches. The vaults are built of several layers of structural terra cotta tiles and a single layer of glazed ceramic tiles; the arches are built of many layers of structural tile and finished with ornamental plaster. While it appears that the sail vaults are supported on the arches, the vaults are, in fact, partially supported by steel trusses. The presence of a secondary load path creates structural ambiguity and complexity in the system.

TYLin’s work on this project, led by Quinn Evans, touched every piece of the building to assess, repair, and restore the original structure. Significant structural efforts would be required to complete the restoration, first just to ensure that the building could be saved. After sitting empty for decades with damaged or missing roofs and windows, the structure was rapidly deteriorating. The very permeable envelope permitted a near constant flow of water into the building, allowing plants to grow out of the concrete floor slabs in the tower and corroding steel framing throughout. In some areas, steel was so badly corroded that it disintegrated when touched, and concerns were raised that the building may no longer be salvageable. The TYLin engineering team began work with extensive structural evaluation to better understand the feasibility of bringing this building back to life.

The conditions of the waiting room ceiling, at the start of the project, reflected that of the full building. Much of the decorative plaster had been lost, exposing the structural terra cotta clay tiles at the arch above. The soffit tiles at the vaults, while still mainly intact, showed areas of cracks, lost or loose tiles, and staining (Fig. 1). Above the ceiling, the conditions were similarly poor. The walls at the upper monitor roof were missing and roofing and flashing were gone, allowing water to enter the interstitial space (attic) above the vaults and come into contact with the steel roof trusses and tile. Broken and loose tiles and other debris were scattered across the tops of the vaults (Fig. 2). Initial site visits focused on global evaluation of the structure to ensure the team could safely access the space to begin the survey work.

The roof structure is made up of concrete slabs spanning to steel purlins that frame to small secondary (east-west) trusses. These secondary trusses sit above the vaults and are supported by the primary trusses. The primary (north-south) trusses are much deeper and clear span from the north to south walls of the waiting room. They are aligned with the main arches of the Guastavino ceiling. The primary trusses are a full story tall, extending the full height of the attic interstitial space. The bottom chord of the primary trusses is partially embedded in the Guastavino arches. The gusset plates extend below the bottom chord with additional clip angles that appear to be designed to better engage the tile at the embedded bottom chords.

The design team had access to a limited number of original design drawings. The earliest architectural drawings identified the ceiling as “imitation stone,” which we assume to indicate plaster scored to mimic stone based on similar finishes used elsewhere in the station, while later historic drawings show the change to Guastavino tile. From the available documents, it is unclear if the design of the trusses had been adjusted to reflect the change in the ceiling system, moving from a fully hung plaster ceiling to a system of tile vaults that were supported partially on tile arches and steel trusses. Original structural design drawings were not found, and available drawings were limited to steel shop drawings and a single sheet from the original Guastavino drawings.

From this limited documentation, it was unclear whether the vaults were designed to span to the arches, with all load landing at the spring points, or to span to the roof trusses. Given the multiple load paths, it was assumed that some combination of these approaches was used. Without the other Guastavino Company drawings or steel design drawings, these questions had to be answered through additional investigation and analysis, which would be crucial to ensure appropriate repairs could be executed safely.

The next phase of work involved significant time on-site to accurately document the geometry and conditions of the vaulted ceiling both for the soffit tile at the intrados and the structural tile and steel trusses in the interstitial space above at the extrados. Documentation included probes and local cores to measure tile thickness at various areas across the ceiling. Existing conditions and deficiencies were documented, including loose tile, missing plaster, and cracks at the soffit tile.

The entirety of the ceiling was sounded at the extrados (upper surface) using an acoustical sounding tool that allowed rapid assessment of the arch and vault assemblies (Fig. 3). Sounding allowed the team to identify tiles that might look to be in good or fair condition but had delaminated from the lower layers. Generally, the delaminated tiles were concentrated near the steeper portions of the vaults at the pendentives, as well as at areas where the tiles were “smooth” instead of ridged from the extrusion process. The conditions at the pendentives frequently corresponded to soffit tile distress, as these areas had seen more consistent exposure to water and, in freezing conditions typical of Michigan winters, formation of ice.
The condition of the steel was an added challenge and drove the need for the substantial documentation and analysis that would be completed on this project. Because many steel members and connections were embedded, much of the repair work would not only impact the steel frame but also require local removals of tile. In many areas, the structure needed full member replacements. Areas closer to the base of the arches were often the worst, where steel corrosion had reached a point of near complete deterioration and section loss (Fig. 4).

Many Guastavino structures are based on simple forms of masonry construction such as an arch, a dome, or a vault. Graphic statics is an analytical tool well-suited for these simple forms. Using this technique, the 3D structure can be reasonably represented by a 2D curve or, where required, multiple curves. Many structures built by the R. Guastavino Company were designed using graphic statics, and the analysis is occasionally included on the Guastavino Company’s drawings. The process of using graphic statics is described in handbooks and engineering textbooks from the period when the R. Guastavino Company was active. It remains an effective tool for analysis of vaulted structures today. However, for some of the Guastavino Company’s structures--generally those vaults in buildings where steel is the primary structure and the mass of masonry bearing walls is not present to resist gravity and lateral loads from the thrust reactions of the vaults--the complexity of the vaults and supporting structure are not a good fit for graphical analysis methods.

Guastavino structures were originally marketed primarily as fireproof floor construction. Later, following the Guastavino Company’s work at the Boston Public Library in the 1890s, the arches, domes, and vaults became more prominent architectural features and were even proposed as alternates to concrete and steel. But, in the early twentieth century, Guastavino began incorporating steel into larger monumental ceiling vaults, allowing the tile vaults to span even larger distances with less weight. The addition of the integrated steel frame adds a second structural system that shares load with the vaults, and the assumptions and simplifications required to use a 2D graphic statics analysis may no longer accurately represent the behavior of the combined structure due to the more nuanced boundary conditions. A graphics statics analysis, following historical methods, does not easily account for partial support offered by alternate load paths such as an integrated steel frame or allow estimation of load sharing based on stiffness compatibility.

However, the simplified approach of hand methods and graphical analysis remains an important first step in masonry vault evaluation. Following the initial evaluation, additional modern analysis methods, including computational design tools such as Rhino and Grasshopper and finite element modeling, provided additional clarity on structural behavior and informed repairs, based on the complex interaction between the masonry and steel structures.

Initial analysis focused on simplified 2D finite element models of the trusses alone, with a conservative loading assumption that the trusses were supporting the full weight of the vaults. These initial checks showed deficiencies in this load path as the sole means of support because the trusses did not have sufficient capacity for these imposed demands. Additionally, hand methods were used to check capacity of the structure if the vault was assumed to be self-supported and spanned to the steel building frame at the spring points. This simplified approach showed that this load path alone was not viable, as the cantilevered steel beams and connections supporting the arch springing points did not have adequate capacity to support the full gravity and thrust reactions of the vaults.

The results of this preliminary analysis confirmed the indeterminate structure was making use of multiple load paths and sharing load between the structural tile and steel roof trusses. Based on the level of repairs required, it was determined a finite element model should be built to more accurately evaluate the load sharing and allow the design of efficient repairs for both deteriorated steel and tile. A full model of the waiting room structure was built using Grasshopper and Rhino computational tools and imported to SAP2000 for finite element analysis. Steel members, including columns, beams, bracing, and truss members were modeled as frame elements, concrete slabs as shell elements, and the tile vault as meshed triangular shell elements. The vaults were modeled with accurate thickness based on hand-documentation of the existing geometry and material properties based on physical testing.

The model was developed through an iterative documentation process, involving multiple return site visits to confirm and clarify geometry. While initial documentation focused on primary geometry and tile thickness, later observations more thoroughly documented small “tertiary” tile bracing elements, such as horizontal bracing arches, tile buttresses, and single-wythe brick walls constructed at steeper portions of the vaults and arches at areas of potential arch instability or weakness (Fig. 5). The inclusion of the bracing elements in the structural model significantly improved both stiffness and behavior.

Nonlinear staged construction analysis further improved the level of accuracy of the model. This analysis allows the application of load to mimic the original construction sequence of the waiting room: first building and loading the steel framing and trusses, followed by the tile at the arches, then the vault bases and pendentives, and finally the center of the vaults (Fig. 6). This significantly improved some of the model deficiencies such as excessive steel deformation from columns pushing outwards due to thrust of the vaults.

Additional iterations included adjustments to material properties, stiffnesses, and connectivity between trusses and vaults using link and spring elements. The results from this envelope of cases were evaluated and compared to establish a smaller subset of baseline “existing condition” models that most closely represented the expected behavior of the vaults and were then used to evaluate proposed repairs (Fig. 7).

Results from the finite element model were post-processed and compared with simplified hand methods, a critical piece of the analysis process. The direct model results are tied to inputs and assumptions made, and by changing inputs, the engineer can greatly alter the takeaways from the analysis. Engineering judgement and critical comparisons with hand calculations are necessary to interpret the results from envelope of iterations.

One area that required a wide envelope of results and methods was the end of the primary arches, which land on cantilevered steel beams that had seen significant deterioration and were supported by emergency shoring installed at the beginning of the project. Given the deteriorated condition of the steel, it was clear there was a redundant load path for the vertical reaction and the thrust of the vault. TYLin established three load paths: support from the trusses, cantilevered framing at the spring points, and friction with masonry strut walls at the pendentives. The team assembled results from each of these individual load paths along with combinations of the three.

These results were compared to steel member capacities to estimate the percentage of load shared through each load path. From this, bounded member demands were determined, allowing an appropriate design for the cantilever repairs that did not require added reinforcement beyond the original structural configuration. The repair approach was to only introduce reinforcing and repairs where necessary. Refining the analysis allowed a more judicious repair approach, retaining as much of the original historic fabric as possible in keeping with preservation goals of the project.

A major consideration for the repair design was to ensure construction could proceed on both the soffit tile from the underside (Fig. 8) and the structural tile from the top side. To meet the construction schedule, these activities were slated to happen in parallel, which meant the approach for structural repairs could not rely on extensive shoring of the vaults from below. Instead, analysis would need to justify stability of the vaults during construction. The finite element model was used to evaluate appropriate construction live loading limits for workers and materials.

The structural finite element model was further developed during construction to evaluate temporary conditions as repairs were executed. One of the most complex repairs occurred at multiple locations at the truss nodes at the bottom chord intersection with horizontal and vertical bracing members, where the connection was partially embedded in the tile (Fig. 9). The repairs required tile removal along with steel replacement at severely corroded members. Multiple steel members would be temporarily disconnected to complete repairs, significantly altering the load sharing between the two systems during the repair process.

The engineering team worked closely with the contractors to develop an initial proposed sequence of member repairs that limited removals to one member at a time. Non-linear staged analysis evaluated the redistribution of stresses through steel and tile elements during each construction stage. Based on the analysis, repairs were completed with adequate access using only very limited shoring; a steel HSS strut was welded across the arch between trusses to resist the unbalanced thrust from adjacent vaults while the original brace was disconnected (Fig. 10), the element repaired, and the original load path restored (Fig. 11).

Using this iterative and collaborative approach, repair continued throughout the steel and tile structures at the interstitial space to restore the deteriorated structure back to its original condition. Based on the structural analysis, the team determined acceptable levels of corrosion at steel elements and only called for repair of elements that crossed this threshold, limiting unnecessary tile removals and additional steel reinforcing. The repairs were executed in-kind with a high level of care and craftsmanship (Fig. 12).

The number of masons with experience working with Guastavino structures is limited, so the design drawings and specifications called for a high level of training and mock-ups (Fig. 13) for the contractors to control the quality of work. Though building Guastavino vaults does require unique materials, skills, and techniques, training programs are available to teach these skills.
Two teams of masons worked to restore the vaults. One was a partnership between a local company (Grunwell-Cashero) and a masonry contractor with extensive experience with Guastavino vaults (Graciano); the other was a local firm whose masons attended a three-day training course (Leidal & Hart). Identifying skilled contractors and specifying mockups allowed for a successful restoration process using traditional craftsmanship.

The Michigan Central Station reopened to the public in 2024. While most of the structural work to repair the steel trusses and structural tiles above the ceiling is not visible to visitors, that work enabled the transformation of the waiting room back to the level of finish it exhibited in 1913 when it was first constructed.

The restoration highlights the importance of nuanced structural analysis for preservation of monumental buildings constructed with archaic materials and technologies. The structural work focused first on a thorough understanding of the complex and ambiguous load paths. Modern analysis tools, including finite element modeling, were critical in more closely understanding the load sharing, but the tools were not relied upon alone. Engineering judgment and comparison with original design methods such as graphic statics and simplified analyses were used to validate results, develop repair strategies, and facilitate construction. Additional time spent documenting and developing a creative approach towards analysis reduced the duration and complexity of construction due to a more accurate understanding of the structural behavior. The team was able to complete repairs with fewer interventions, aligning with the project’s goals to restore and preserve the monumental historic structure. ■

About the Author

Margaret Cowie is a licensed structural engineer working in Washington, DC at T.Y.Lin. Her project work encompasses a range of building types and systems, with a focus on historic preservation and adaptive reuse. Much of Cowie’s work has focused on analysis of unreinforced masonry structures, including restoration of Guastavino structures at Michigan Central Station, the Smithsonian Natural History Museum, Buffalo Central Terminal, and several others throughout the United States.