As part of the firm’s robust internship program, our summer interns each complete a research project that they present to the firm. In this multi-part series, our 2023 cohort shares their findings; see all the internship posts here.
Artificial intelligence has already started to reshape the relationship between design professionals and their work. These technologies have also sparked fear and concern around issues of labor replacement, lack of human connection, and data management. Historic preservation is likewise affected. Recent advancements in machine learning are potentially transformative as tools to streamline documentation and data recall.
My research reviews how preservationists document historic resources and how nascent technology may affect these workflows. Using time capsules as a framing device for the storage and discovery of valuable information, it conceptualizes emerging relationships between historic resources and their metadata. My hypothesis is that digital documentation processes can act as a new kind of time capsule: one that can be updated over a building’s lifetime, centralized, and made accessible to interested parties for future study.
Diagrams illustrating interpretation, analysis, and intervention in historic preservation workflows.
Defining Current Methods
I began by distilling preservation deliverables into three thematic types: interpretation, analysis, and intervention. The types are not exhaustive but rather meant to better illustrate current data/resource relationships. In all three cases, documentation exists separately from the historic resource. The usefulness and longevity of these documents relies on several factors, including the interest of the client and future accessibility. In other words, where documentation is stored and who can access it impacts the understanding of a historic resource.
HBIM
Historic Building Information Modeling (HBIM) has enriched and complicated documentation processes. Hennebery Eddy’s recent inventory of the Montana State Capitol included an HBIM model as a deliverable. Capitol facilities staff reference and update it as maintenance work is completed on the building. The model represented historic data through several methods, including filled regions and model lines to signify damage and custom door and window parameters. These methods of documentation are not native to Revit; however, an accurate accounting requires painstaking manual labor and customized workarounds. Still, HBIM introduces a fourth-dimensional aspect to preservation documentation, making it easier to update and remain in communication with the resource.
A.I.
Machine learning can strengthen and automate communication between historic resources and their documentation. Deep Learning (DL) models that use Convolutional Neural Networks (CNNs) have been applied in recent preservation studies. A recent experiment used a CNN model to identify and classify damage on the exterior of a historic church in Aveiro, Portugal. A typical LiDAR-to-HIM process served as the control. The research team also took manual photographs of damage conditions on the building’s façade and trained an A.I. model to identify and classify the damage. The A.I. algorithm was then plugged into the BIM space and asked to identify and classify the same types of damage on the digital proxy. The intent is for future applications to rely on established A.I. models to be able to identify damage in a digital space, entirely bypassing the need to manually document damage on site.
A second study used hygrothermal sensors inside a historic museum to create a Decision Support System (DSS). An A.I. model received real-time hygrothermal data and relayed its conclusions to an HBIM proxy via a Grasshopper script. The script then made occupancy and ventilation recommendations based on the determinations of A.I. Documentation related to the preservation of the building was not only fully integrated with the building itself but also entirely automated, creating a “sentient” feedback loop that could continue long after the conclusion of the study.
Recent studies used machine learning to automate damage detection on a church façade (left) and to create a hygrothermal feedback system for a historic museum (right).
Summary
Digital processes are already helping resolve issues of time, accuracy, and maintenance in historic preservation. Our processes may soon exist in an entirely digital space. Establishing effective and meaningful methods to document historic resources — and making them accessible to future interested parties — is critical. The future will be looking to digital spaces for compelling arguments as to why they should care about our cultural heritage. If preservationists can leave behind a truthful and rich trail of information that is closely tied to the resource itself, historic buildings and sites may continue to have a cherished place in a more digital world.
Sources
- Hamlin, Carole. Informational Interview Regarding Hennebery Eddy’s 2019 Montana State Capitol Building Inventory, July 24, 2023.
- La Russa, Federico Mario, and Cettina Santagati. “An AI-Based DSS for Preventive Conservation of Museum Collections in Historic Buildings.” Journal of Archaeological Science, Reports 35 (2021): 102735.
- Rodrigues, Fernanda, Victoria Cotella, Hugo Rodrigues, Eugénio Rocha, Felipe Freitas, and Raquel Matos. “Application of Deep Learning Approach for the Classification of Buildings’ Degradation State in a BIM Methodology.” Applied Sciences 12, no. 15 (2022): 7403.
- Smith, Jason. Informational Interview Regarding Hennebery Eddy’s 2019 Montana State Capitol Building Inventory, July 6, 2023.