Abstract

Industry Foundation Classes (IFC) enables the open exchange of Building Information Models, yet effective validation of real-world IFC deliveries remains challenging. Project models frequently contain incomplete, inconsistent, or incorrectly exported metadata, and explicit IFC relationships are often missing or degraded during conversion from proprietary formats. Many validation questions are inherently spatial, including connectivity, containment, proximity, and clashes, and therefore cannot be answered reliably from properties alone. As a result, federated model validation is often slow, fragmented across tools, and dependent on specialized IFC expertise.

This challenge is increasingly important for coordination, quality assurance, and information management across disciplines and lifecycle stages. At the same time, the complexity of the IFC schema, its class hierarchy, relationship structures, and property set conventions, presents a major barrier to adoption. Even BIM-proficient practitioners and project stakeholders often struggle to express validation intent in IFC terms, limiting the accessibility and reuse of advanced openBIM validation workflows.

This thesis addresses two gaps: the need for robust validation when IFC semantics are incomplete, and the need for schema-agnostic access to technical validation capabilities. The objective is to combine deterministic semantic querying and filtering with derived geometric and topological reasoning, and to provide a tool-augmented multimodal LLM interface that enables natural-language and vision-assisted IFC validation while maintaining traceable, evidence-based outputs. To achieve this, the Intelligent Node-based System for IFC Processing and Evaluation, with Conversational Tool Support (INSPECT) was developed as an open-source node-based framework for federated IFC processing and evaluation.

INSPECT provides a modular node library for model loading and federation, schema exploration, element-level data retrieval, and chainable filtering over IFC attributes, properties, quantity sets, associations, spatial containment, and explicit relationships. To compensate for missing or unreliable relationships, derived spatial queries are implemented using a Bounding Volume Hierarchy (BVH) to support efficient geometric reasoning, including proximity, touching, intersection, containment, and geometric connectivity. The same BVH supports a two-phase clash and clearance validation pipeline, using broad-phase candidate pruning followed by mesh-based narrow-phase evaluation with a tolerance parameter for both hard clashes and clearance violations. A separate interactive 3D viewer supports visualization of federated models and presentation of results. A conversational multimodal LLM agent is integrated through tool calls to validation functions and viewer controls, leveraging vision-based reasoning over rendered model views and web-retrieval support to incorporate external domain context when required.

This work demonstrates the feasibility of a two-layer validation approach: technical users can author precise, repeatable node-based validation workflows, while non-expert users can request technically detailed outcomes in natural language without requiring IFC schema literacy. The system shows how geometry-derived reasoning can support validation tasks when explicit IFC relationships are absent, and how vision-based reasoning can assist when element meaning or design intent is not reliably represented in IFC metadata.

School

School of Sciences and Engineering

Department

Construction Engineering Department

Degree Name

MS in Construction Engineering

Graduation Date

Fall 2-15-2026

Submission Date

1-26-2026

First Advisor

Khaled Nassar

Second Advisor

Ossama Hosny

Committee Member 1

Ibrahim Abotaleb

Committee Member 2

Volker Krieger

Extent

162 p.

Document Type

Master's Thesis

Institutional Review Board (IRB) Approval

Not necessary for this item

Disclosure of AI Use

Thesis editing and/or reviewing; Code/algorithm generation and/or validation

Download

Share

COinS