How to leverage Building Information Modeling (BIM) for historic buildings

Michael Carrancho, Deputy Director Office of Planning Design and Construction at Smithsonian Institution

Through this article, Michael Carrancho explores the innovative use of Building Information Modeling (BIM) for historic buildings, a practice dubbed "Heritage BIM." Carrancho highlights the potential for BIM to preserve and analyze historical structures, despite limitations in current modeling software. The project also revealed the possibility of leveraging artificial intelligence (AI) for future restoration efforts by categorizing material deterioration and creating a standardized glossary of deterioration mechanisms.

For over ten years, the Smithsonian Institution has been collaborating with the Georgia Institute of Technology through Georgia Tech’s Digital Building Laboratory on a pioneering research project to leverage Building Information Modeling (BIM) for historic buildings, which we call "Heritage BIM." Initially, the project's goals were unclear and undefined; however, we were confident that specific objectives would emerge through engagement with historic buildings.

Two Smithsonian buildings were selected as pilot projects to explore the potential of BIM technology for documenting, analyzing, interacting with and improving historic structures. The first building was a wooden tenant house located at the Smithsonian Environmental Research Center in Edgewater, Maryland. This structure, believed to be a pre-civil war building, was thought to have been assembled from two cottages initially built to house enslaved people. The second building was the Freer Gallery of Art, an Italian Renaissance-style structure that opened in 1923 on the National Mall in Washington DC.

The gallery, clad in granite, marble and limestone, provided a contrasting architectural style and material composition to the tenant house, allowing us to evaluate our BIM application across diverse scenarios.

We employed an array of scanning technologies to document the existing conditions of these buildings. Utilizing the combined technical and personnel resources of the Smithsonian and Georgia Tech, we employed a range of tools, including a tripod-mounted Leica BLK360 and FARO Focus S-350 laser scanners, handheld scanners, Matterport Pro2 and Pro3 cameras and DJI Mavic Pro-2, Pro3, 3 Enterprise with RTK and Skydio 2 drones. These were used to photo the roof and provide topography of the site, enabling us to capture interior, exterior and roof information in various formats and resolutions.

We processed this raw information into Revit BIM models upon gathering the scanned data. At this stage, we encountered a significant limitation of Revit and similar modeling technologies. These tools are primarily designed for new construction and assume that all dimensions are standard and orthogonal. However, historic structures rarely conform to standard dimensions and are rarely orthogonal. We accurately represent the existing conditions of a building over a hundred years old—one that has undergone wood creep, settlement, shifting, poor original construction and numerous other changes—proved impossible. Consequently, we abandoned the goal of perfect accuracy and instead used the data to create an idealized approximation of existing conditions.

During our investigation, we also identified several types of material and structural deterioration. One potential research goal was to explore leveraging artificial intelligence (AI) to analyze historic buildings. We anticipated that photographing and categorizing different material deterioration mechanisms could provide training data for a machine learning AI system, potentially allowing it to assess future buildings without needing a trained professional. However, this line of inquiry revealed another significant limitation: the lack of a unified glossary and definitions for wood and stone deterioration. In response, the team began developing material deterioration glossaries, prioritizing the wood glossary.

Before advancing with machine learning, the wood and stone deterioration glossaries must be reviewed and agreed upon. Additionally, an extensive collection of photographs must be assembled, with each deterioration mechanism identified and assigned to the corresponding photo. The AI would then process this data to learn as many examples of material deterioration as possible. Once trained, the AI could review new photographs and apply the appropriate deterioration mechanism to each image based on what it has learned.

“Heritage BIM allows us to honor the past while using cutting-edge technology to document, analyze and preserve historic structures for future generations.”

Although we were not able to advance the machine-learning aspects of the project, we were able to provide 3D models of the building and its structure with historical attributes associated with colored model parts. It was confirmed that the tenant house was assembled from different structures and was from three different eras. The interior wall framing was circa 1800 and was colored blue in the model, the rafters of the roof structure were circa 1870-1880 colored green, and the exterior cladding was circa 1940-1960 colored brown. The photos, models and colored graphics will be incorporated into a future educational plaque to be placed at the site to inform visitors of the history of the building.

The application of artificial intelligence to the documentation and conservation of historic buildings represents a significant advancement. It would allow unskilled facility managers to use simple techniques such as photography and drone surveys to document existing conditions, enabling AI to assess and diagnose the state of the buildings. With this diagnosis, repair, restoration and rehabilitation could be conducted without requiring extensive technical knowledge.