Virtual Vaults: Using AI to process, analyse and supplement digital data

In recent years the study of architecture has been transformed by the use of 3-D modelling and digital analysis, through which a growing corpus of 3-D data has been generated. Whilst such data is relatively easy to capture, existing methods of processing and analysis are labour-intensive and costly, limiting the capacity for exploiting the data. AI (artificial intelligence) has the potential to speed and scale up these developments. This project has three aims: (1) To investigate the potential of AI for tracing key elements of medieval vaults in 3-D data. The aim would be to automate workflows developed by the AHRC-funded ‘Tracing the Past’ (TTP) project to make data processing quicker and cheaper, enabling more sites to be incorporated into comparative models, making hypotheses more testable and conclusions more robust. No one has previously developed such a tool, although comparable tools exist in other contexts; its feasibility is likely but its potential is untested. If successful, the project could be extended to other architectural elements, e.g. domes or tracery. Objectives: (a) To investigate the potential of AI for tracing vault elements to create wireframe models and recording key data for analysis, i.e. impost level (height at which the rib deviates from the vertical, not necessarily at abacus level); apex (the height at which ribs would intersect - a notional point as the junction is usually masked by a boss); rib radius or radii (based on 'best fit' curves); location of centre of rib curvature in relation to impost (at impost level or distance above or below). (b) To test AI against human-processed data derived from TTP. (c) To develop a standardised and generally applicable approach to AI data processing. The benefit would be the enhanced capacity to develop links with other researchers to develop larger international comparative projects. (2) To investigate the use of AI for analysing 2-D data to identify design methods. Objectives: (a) To use the wireframe models generated by (1) to create 2-D plans of the vaults, whose designs would be compared by AI against geometrical and proportional design models identified by TTP and those proposed by other researchers. (b) To use AI to identify other design parameters and shape grammars. (3) To investigate the use of generative AI for reconstructing missing elements from scans of ruined vaults. This would enable the addition of hypothetical vaults to the growing dataset of 3-D scans of extant vaults and offer the potential to test human assumptions about their 3-D design. No one has previously developed such a tool; its feasibility is likely but its viability is unknown. Objectives: (a) To use the geometrical rules developed by TTP and any additional rules identified by (2) to propose hypothetical forms for unbuilt vaults (b) To assess these against expert opinion on likely forms If successful, this tool would provide proof-of-concept for future development of further tools for AI analysis, which would again test existing (human intelligence informed) models of design analysis and categorisation, potentially finding new patterns and invoking new hypotheses.