Georgia Institute of Technology
RiG-X: Retrieval of Spatio-Temporal Knowledge from Cyber-Physical Building Systems over ifcLD Graphs with eXtensible Semantic Overlays
Abstract
dc:description.abstractToday’s buildings are complex and elaborate Cyber-Physical Systems (CPS) that generate a lot of data streams from a myriad of sensors and devices that communicate with each other. However, a lack of data interoperability, and a resulting inability for decision-making and generalized intelligence over building data, leads to major inefficiencies affecting the triple bottom line of the built environment; current Building Information Models (BIM) are a repository for design/construction data but their potential for building operations is largely restricted to static data. Conversely, Internet of Things (IoT) technologies provide live time-series data about buildings but often lack the spatial context and linkage architecture to BIM. There have been recent advances in associating BIM to Linked Data (LD) principles, Knowledge Graph Question Answering (KGQA), Large Language Models (LLMs), graph-retrieval, and limited time-series data; yet, there is a lack in semantic normalization of building data and extensible frameworks that unify spatial structure, ontologies, and time-series telemetry into a queryable format. This is crucial to enabling cross-disciplinary reasoning, such as enabling an agent to reliably correlate BIM entities with operational sensor streams to support adaptive decision-making in facility management. In order to address the inefficiency of traditional BIM formats on the web and semantics, this research couples the IFC-LD ontology, which is a recently developed RDF-encoded version of the IFC standard for BIM data exchange, with metadata models, which are further overlaid with semantic-temporal enrichments. The data model is used to represent static building data in a SHACL-validated graph. The graphs are overlaid with the ASHRAE 223P and Brick ontologies that embed types for sensors, data streams, and units. Time-series data is read, stored separately, and accessed via external references in the graph from a PostgreSQL database. Experiments demonstrate that the semantic graph structure can be queried using natural language, which is converted to SPARQL and SQL via a GraphRAG pipeline, where an LLM grounds answers through graph retrieval that assembles structured evidence for downstream analysis. This architecture for a semantically normalized data model enables future LLM-based agents to build upon this knowledge hub, improving intent interpretation, and cross-domain reasoning over inconsistently named building points, supporting agentic proptech applications built on traceable, ontology-grounded evidence.
Degree
thesis:*- Name thesis:degree_name
- Architecture, MS
- Department thesis:degree_department
- Architecture
- Grantor
- Georgia Institute of Technology
- Year dc:date.issued
- 2026
Author and committee
dc:creator, dc:contributor.*- Author dc:creator
-
- Aerathu, Joseph Mathew
- Advisors dc:contributor.advisor
-
- Kastner, Patrick
- Sheward, Hugo
- Committee members dc:contributor.committeemember
-
- Shelden, Dennis R
- Sacks, Rafael
- Panagoulia, Eleanna
- ORCID person:identifier.orcid
Rights
- Language dc:language.iso
- English
Identifiers
dc:identifier.*- Handle dc:identifier.uri
- https://hdl.handle.net/1853/81610