Incremental IFC updates with ‘why’ and ‘how’: A case of IFC extension for LLM agent-enhanced simulation-based optimization for building energy efficiency Lingming Kong, Qianyun Zhou, Fan Xue Department of Real Estate and Construction, University of Hong Kong, Hong Kong SAR, China u3009509@connect.hku.hk qianyun8@connect.hku.hk xuef@hku.hk Abstract Simulation-based optimization (SBO) for energyefficient building design exemplifies a typical designchange process that remains disconnected from open BIM-enabled collaboration. The gap stems from IFC’s limited capacity to represent the information generated by SBO. To overcome the limitations, this study proposes a novel IFC schema extension for LLM Agent-enhanced SBO, named IfcSboExtension, that enriches incremental IFC updates with contextual metadata capturing not only what changed, but also why the change occurred and how it was derived. The proposed extension model enables exchangeable and traceable SBO iterations by embedding decision rationale and methodological provenance directly into incremental IFC updates. The feasibility and value of the extension are demonstrated through a case study on building energy efficiency optimization, in which an LLM agent enhances the interpretability of design changes. Keywords – Building information modelling (BIM), Industry Foundation Classes (IFC), Incremental IFC updates, Simulation-based optimization (SBO), Large language model (LLM) 1 Introduction Building information modelling (BIM) is recognized as a digital integration of information throughout a building lifecycle, and increasingly mandated in the AEC (Architecture, Engineering, and Construction) industry [1]. BIM-based collaboration has progressively grown to level 2, namely “open BIM”, which signifies a fully integrated, collaborative environment for project stakeholders through federated discipline models and open exchange standards [2]. Industry foundation classes (IFC), and ISO 16739-1:2018 standard, are per se open BIM that provides a vendor-neutral data format for 2086 seamless data exchange between various BIM applications [3]. However, the scalability and adoption of IFC are constrained by the monolithic, file-based exchange paradigm. In practice, BIM stakeholders receive the modified model as a new monolithic file from collaborators and check manually to identify the potential impact on their own disciplines, which is intensive in labor and prone to errors. IFC 5, the upcoming modular version, plans to implement incremental IFC updates to address these issues [4]. From a theoretical perspective, the transition toward incremental IFC updates aligns with the processcentric knowledge philosophy [5], which emphasizes that construction knowledge is dynamic, iterative, and context-sensitive regarding knowledge dissemination among project partners. Consequently, the modular architecture of IFC 5 is expected to better support both operational efficiency and effective knowledge management in construction projects. Simulation-based optimization (SBO) is a typical design-change process in which design parameters are iteratively adjusted, leading to modifications to model design. Recent studies have leveraged large language models (LLM) to promote the interpretability of traditional black-box SBO workflows [6]. However, the information generated by the LLM-enhanced SBO remains neither exchangeable nor traceable within the context of sustainable, collaborative design based on open BIM. To address the gap between LLM-enhanced SBO and open BIM, this paper proposes an IFC extension for LLM-enhanced SBO, which explicitly answers ‘why’ and ‘how’ in addition to ‘what’ to enrich such incremental IFC updates. The remainder of this paper is organized as follows: Section 2 reviews the state-of-the-art research on incremental IFC updates and simulation-based optimization. Section 3 introduces the proposed IFC schema extension for incremental update. Section 4 demonstrates the implementation of the extension through a case study employing large language model (LLM) agents for building energy efficiency optimization. Section 5 provides the conclusion. The primary contributions of this study are twofold. First, the proposed IFC schema extension emphasizes that the ‘why’ and ‘how’ dimensions are essential to the ‘what’ of incremental IFC updates. Second, the feasibility of the proposed extension was validated through a real-world project. 2 Literature Review 2.1 IFC and schemas for changes IFC was originally conceived as a monolithic, objectoriented schema based on the EXPRESS data modelling language [7]. As an object-oriented data schema, the IFC schema contains several classes, namely “entities”, and serves as “templates” to describe building information. The semantic meaning and the implementation of entities are generalized through inheritance mechanisms. Each entity declares a set of local attributes, in addition to the global attributes inherited from its supertype. There are two types of IFC entities, namely rooted and non-rooted entities, which contain or do not contain a globally unique identifier (GUID). In practice, the instantiated IFC entities are transformed into ASCII text rows in the text-based IFC-SPF (STEP Physical File) formatted file [8]. An IFC-SPF file is exchanged in a snapshot-based exchange paradigm, which means each IFC-SPF file represents a complete state of a model at a given time, regardless of actual changes. This exchange paradigm leads to inefficiencies in data transmission, version control, and change impact analysis for decision making in the collaborative, multi-disciplinary environment [9]. To overcome the limitations of holistic IFC file exchange, recent research and standardization efforts have shifted toward change-based IFC exchange approaches. In the literature, three categories of difference (or delta) computation methods have been developed to capture changes between two IFC models, which are (1) GUID-based; (2) flattening-based; and (3) graph-based approaches. The GUID-based approach recognizes two IFC instances as identical if they share the same GUID, which is widely adopted by commercial BIM platforms such as Autodesk Revit, Navisworks and Graphisoft ArchiCAD. However, a significant limitation of this approach is that GUIDs may not remain consistent across different BIM authoring tools, rendering GUID-based comparison unreliable. The flattening-based approach eliminates randomly referenced identifier numbers (i.e., the “#”-IDs) in IFC files. The “flattening” process is to replace all the “#”-IDs with specific instant IDs, such as content-based 2087 hash ID [10][11]. Therefore, flattening-based approaches enable deterministic textual comparison. Both the GUID-based and flattening-based methods are fundamentally text-oriented. In contrast, the graphbased approach leverages graph-based algorithms to compute semantic differences between two graphrepresented IFC models. There are two main steps involved in the graph-based approach. First, the graph representation of IFC models is developed, such as RDF (resource description framework) graph and directed property graphs. Then, graph-based algorithms, such as graph transformation [12], are implemented to capture the changed part of two IFC models. 2.2 Knowledge engineering applications of changes in construction The dynamic theory of knowledge creation developed by Nonaka and Takeuchi [13], namely the SECI model, posits that knowledge is created, transferred, and applied through the creative tension between tacit and explicit know-how. The SECI model reveals a significant source of knowledge, that knowledge is generated from continuously combined, refined, and reconfigured explicit assets. This definition aligns with the iterative development processes of construction projects [1]. Accordingly, changes in a construction project, such as design iterations, site conditions, and stakeholder feedback, are recognized as a knowledge source which has great influence on successful project delivery [14]. Knowledge engineering (KE) is the process of designing, building, and maintaining knowledge-based systems by identifying relevant domain knowledge, representing it in formal structures (such as ontologies, or knowledge graphs), and implementing reasoning mechanisms that can use the knowledge to make inferences, decisions, or recommendations [15]. A core KE approach is the development of domain ontologies that explicitly define change types, actors, causes, and impacts. Early work by Beetz et al. [16] transforms the EXPRESS schema of IFC into an OWL ontology, namely IfcOWL, and paves the way to change detection as ontology alignment. The following work of Oraskari and Törmä [17] introduces a signature algorithm for RDF diff computation. Furthermore, a domain ontology for pumped storage hydropower system was developed by Zhang et al. [18] to support design changes management. IFC also can be serialized into various data structures, such as ifcJSON and labelled property graph, to support algorithmic reasoning for change detection. For example, A semantic differential transaction (SDT) approach and the following traceable SDT approach were developed based on the JSON-format IFC to capture the semantic changes through structure matching [10][11]. 2.3 Simulation-based optimization as a typical change Simulation-based optimization (SBO) is widely applied to building design by iteratively tuning design parameters towards sustainable objectives [19]. The generic SBO workflow comprises iterative simulations that are evaluated through objective functions defined in optimization algorithms. SBO constitutes a typical design-change process in which the design parameters vary across iterations, producing corresponding modifications in design. For example, energy performance optimization is a key task in SBO for early stage building design, aiming to reduce the environmental impact of future buildings. SBO involves tuning design parameters such as window-to-wall ratio and building materials. In each iteration, these parameters are adjusted to gradually improve energy performance. Traditional SBO is challenged by high computational complexity and low interpretability. A recent study by Zhou and Xue [19] introduces an Automatic Information Gain-guided Convergence (AIGGC) method to refine critical design parameters through interpretable information gain analysis. Subsequent research integrated the large language model (LLM) to further enhance the interpretability and efficiency of complex SBO [6]. However, the information produced by LLM-Agent enhanced SBO is neither exchangeable nor traceable within the context of sustainable, collaborative design based on open BIM. Existing literature characterizes BIM-supported SBO as a one-way workflow that portions of the required parameter data are extracted from BIM [20], yet the SBO iteration results are not embedded back into BIM for collaborative use. The limitation arises for two reasons. First, contemporary SBO practices are predominantly implemented on visual programming platforms such as Grasshopper, which do not natively support IFC. Secondly, the IFC schema lacks the capacity to represent the information generated by LLM-Agent enhanced SBO, such as design parameters, constraints, and convergence analysis. The proposed IfcSboExtension uniquely encode the decision rationale, optimization provenance, and LLM interaction context to support change-based IFC exchange for traceable and interpretable collaborative design. 3 IFC extension for LLM agent-enhanced SBO To extend IFC’s capability of representing information of LLM Agent-enhanced SBO, an IFC extension model named ‘IfcSboExtension’ was proposed. 2088 The extended schema illustrates ‘what’ data should be involved, ‘why’ the IFC schema should be extended, and ‘how’ to capture the incremental IFC changes within the LLM Agent-enhanced SBO. A general IFC extension workflow is illustrated in Figure 1, which constitutes three steps: (1) Information requirements in LLM Agentenhanced SBO, (2) IfcSboExtension model development, and (3) verification and implementation for incremental IFC updates. In ormati on re uirements IFC ased tension model Identifying through literature review Sets of SB related M gent related information Comparison with existing IFC entities imitations of the existing IFC schema on M gent enhanced SB Mapping information with the IFC data schema xtended entities and attributes Implementing extended entities n IfcSbo xtension model illustrates the relationship between extended entities odel Synthetic semantic eri ication chec s implementation model view definition file (.exp Case study Figure 1. IFC extension workflow The first step answers ‘what’ and ‘why’ the IFC schema should be extended to LLM Agent-enhanced SBO. Table 1lists the crucial data of two categories: (1) SBO-related data such as design parameters and optimization constraints, as well as (2) LLM Agentrelated data such as prompts and structured responses. A meticulous comparison with the existing entities within IFC4 ADD3 revealed that both SBO-related and LLM Agent-related information required extension to enable seamless data exchange. For example, IfcProcess is defined in existing IFC schema to represent one individual activity that has sequence relationships with other processes. Three subtypes of IfcProcess, namely IfcTask, IfcEvent, and IfcProcedure do not contain the semantics such as parameters, constraints, and objective functions. One compromise approach supported by IFC schema is identifying specific property sets, such as Pset_SBO. However, this approach converts SBO semantics to unstructured strings and leads to semantic ambiguity and semantic loss. 3.1 The IfcSboExtension model According to the required data listed in Table 1, three rooted entities and two non-rooted entities were extended align with the IFC4 ADD3 schema. As Figure 2 shows, the extended entity IfcSboWorkflow and IfcAiAgentInteraction are two subtypes of the abstract IfcControl, and IfcOptimizationIteration is a subtype of the abstract IfcProcess. The reason lies in the semantic definition of IfcControl and IfcProcess. In IFC4 ADD3, IfcControl represents all concepts that control or constrain the utilization of products, processes, or resource, whereas IfcProcess is defined as one individual activity or event. According to the definition, IfcControl and its subtypes are suitable for representing a workflow, whereas IfcProcess and its subtype focus on one specific process. In LLM Agent-enhanced SBO, SBO is a wor flow that contains a set of ‘simulation-optimization’ iterations, and the interaction between stakeholders and LLM Agent is also recognized as a set of communications rather than single process. One optimization iteration aligns with the definition of IfcProcess which transforms input in output. IfcObjectiveFunction and IfcSboOptimizationDirectionEnum are two non-rooted entities extended at the resource layer. These two non-rooted entities cannot exist independently but can only exist if referenced by the rooted entity IfcSboWorkflow. Due to page limits, only the extended IfcSboWorkflow entity and relevant attributes is present in Figure 3, which illustrates the EXPRESS representation of IfcSboWorkflow to ensure both machine and human readability. Table 1. Crucial data in LLM Agent-enhanced SBO SBOrelated LLM Agentrelated Crucial Data Data type (in IFC) Example SBO parameters SET [1: n] of REAL (0.45,0.45,1.1,2,5.0733) SBO constraints STRING |𝐹𝐴𝑅𝑜𝑟𝑖 − 𝐹𝐴𝑅𝑜𝑝𝑡 | ≤ 0.05 (FAR for floor area ratio) SBO objectives STRING EUI = 112.684 𝑘𝑊ℎ/𝑚2 ∙ 𝑦𝑟 SBO results STRING or BINARY JSON or CSV LLM prompts STRING “Please identify which design parameters most influence the f(S)?” LLM responses STRING or BINARY Response text or Log file language grammar. The exported use-case model view definition, designated as IfcSboExtension, incorporates 19 existing IFC entities and 5 newly defined IFC entities. In the second stage, semantic verification was performed by compiling the extended IfcSboExtension model into Java classes using the Apstex IFC Framework. Additionally, accessor methods of explicit, derived, and inverse attributes of the extended IFC entities are developed. Critically, domain rule enforcement methods were developed, and corresponding unit tests were designed to verify the IfcSboExtension model based on the created Java classes. Figure 2. Extract of the IFC LLM Agent-enhanced SBO extension with new entities 3.2 Model verification For verification, the official IFC documentation toolkit IfcDoc and a test software platform APSTEX were employed. The verification procedure involves two steps: syntactic verification and semantic verification. As summarized in Table 2, syntactic verification ensures compliance with the formal grammar of the EXPRESS language. The extended IFC entities were documented using IfcDoc to verify their compliance with EXPRESS Figure 3. The extended IfcSboWorkflow entity and relevant attributes Table 2. Model verification procedure, criteria, and tools Verification procedure Criteria Tools Syntactic checks Formal compliance with the grammar defined in ISO 10303-11(EXPRESS): IfcDoc 2089 • Notation conventions EXPRESS-compliant declaration of • ntities (“entity data types” and types (“defined data types” , • numerations (“enumeration data types” , • Subtype and supertype constraints, • Explicit, derived, and inverse attributes of entities Semantic checks Compilation of the LLM Agent-enhanced SBO schema into Java classes for semantic checking: • Java classes for entity, type, enumeration considering the inheritance hierarchy • accessor methods for explicit, inverse, and derived attributes • methods to check domain rules and uniqueness rules Apstex IFC Framework Figure 4. Examples of domain rule checking method and unit test for "HasObjectives" of the IfcSboWorkflow entity Figure 4 presents an example of a domain rule checking method and its associated unit test. For efficient compilation, so-called “lambda expressions” were used to implement the domain rule checks. The example ‘whereRuleChec ’ is defined to chec the applicable objectives of the extended entity IfcSboWorkflow. Corresponding unit test is shown in Figure 4(b). “Has bjectives” of the IfcSboWorkflow entity is shown in Figure 3, which represents a set of IfcObjectiveFunction instances are assigned to one IfcSboWorkflow instance. In the unit test, the process of adding a module group to IfcSboWorkflow results in a ‘False’, which indicates an incorrect state of the IFC test model. 4 4.1 Case settings In this case, the optimization objective was to minimize the energy use intensity (EUI), defined as the total annual energy consumption divided by the gross floor area (𝑘𝑊ℎ/𝑚2 ∙ 𝑦𝑟). As shown in Figure 5, the project features a symmetrical layout with fixed modules and movable modules. The optimization process focuses on three types of movable modules which vary in length, namely 𝑀𝐴 , 𝑀𝐵 , and 𝑀𝐶 . Thus, module layout, window size and material properties are involved as design parameters shown in Table 3. A case study of LLM agent-enhanced SBO To illustrate ‘why’ and ‘how’ in addition to ‘what’ of IFC incremental updates, an energy consumption optimization scenario is selected as a representative design-change case study. The design optimization of a modular integrated construction (MiC) project for a university student hostel located in Hong Kong is selected as a case. The developed LLM Agent for efficient and interpretable SBO is presented in the published work of Zhou and Xue (2025). 2090 . m ass e e s Figure 5. MiC case layout and schematic of module types (adopted from Zhou & Xue (2025)) 4.2 4.2.1 Results Schemas and sample changes The implementation of the IfcSboExtension model using IfcDoc is presented in Figure 6 as an EXPRESS-G diagram. Black rectangles represent rooted entities, whereas grey rectangles denote non-rooted entities. ntity’s attributes – comprising own and inherited attributes - are separated by an ellipsis and categorized as either explicit or derived attributes (in black font) or inverse attributes (in light grey font). Relationships between the extended entities are connected through inverse attributes. For example, an IfcOptimizationIteration is referenced by an IfcSboWorkflow through its inverse attributes ‘ asIte at s’. Figure 6. EXPRESS-G IfcSboExtension model representation reducing Energy Use Intensity. This transparent exposure of the 'why' (energy optimization) and 'how' (sensitivity analysis) enables the owner to grasp the technical rationale instantly, facilitating an informed approval decision. 4.2.2 Sensitivity analysis The text-based IFC-SPF file exhibits acute sensitivity to: (1) excessive line lengths, (2) special characters such as unescaped Unicode or JSON brackets, and (3) exponential file growth. Specifically, the extended IfcAiAgentInteraction records prompt and structured responses, which may comprise multi-megabyte strings, that can significantly decrease data exchange efficiency. In this study, two mitigation strategies were evaluated: (1) compressing the structured response using the Gzip algorithm and encoding the compressed result as Base64 strings; and (2) storing LLM Agent prompts and structured responses in external files referenced via IfcURIReference. Gzip and Base64 are open standards for data compression and encoding, respectively. Metadata describing the compression and encoding algorithms are embedded in IfcAiAgentInteraction to ensure self-descriptiveness. URIs for external files should conform to standard URI schemes (e.g., HTTP and HTTPS), and the corresponding MIME type should be explicitly specified. The evaluation was performed over 30 iterations of LLM Agent-enhanced SBO, with results presented in Table 4. The result indicates that the ‘Gzip + Base64’ strategy increased the IFC file size by a maximum of 2.47% and an average of 1.91%. The IfcURIReference strategy effectively eliminates the impact of LLM Agent responses on IFC file size. of The embedded LLM Agent-enhanced SBO semantics is illustrated in Figure 7. IfcSboWorkflow is assigned to a module through IfcRelAssignsToGroup, and IfcAiAgentInteraction is assigned to IfcSboWorkflow through IfcRelAssignsToControl. To illustrate the change of parameters during SBO iterations, parameters listed in Table 3 are selected. For instance, when an iteration increases the north window area from 4.33 m² to 4.49 m², embedded convergence semantics allow the owner to immediately identify this change as the primary driver for Table 3. Design parameters for MiC parametric design Design parameters Parameter type Unit Parameter domain intervals [𝑀𝐴 , 𝑀𝐵 , 𝑀𝐶 ] Precision MiC Module Type Categorial - Window-to-Wall Ratio (𝑊𝑊𝑅𝑁 , 𝑊𝑊𝑅𝑆 ) Window Height (𝑊𝐻𝑁 , 𝑊𝐻𝑆 ) Continuous - 0.10 ≤ 𝑊𝑊𝑅𝑁 , 𝑊𝑊𝑅𝑆 ≤ 0.60 Discrete M 1.0 ≤ 𝑊𝐻𝑁 , 𝑊𝐻𝑆 ≤ 2.5 0.1 Window U-value (𝑈𝑤 ) Continuous 𝑊/(𝑚2 ∙ 𝑘) 0.50 ≤ 𝑈𝑤 ≤ 5.78 0.01 Window Solar Heat Gain Coefficient (g) Window Visible Transmittance (𝜏𝑣 ) Continuous - 0.100 ≤ 𝑔 ≤ 0.775 0.001 Continuous - 0.300 ≤ 𝜏𝑣 ≤ 0.881 0.001 4.2.3 User acceptance and comments The extended IfcSboExtension model was implemented on a real MiC project for a university student hostel located in Hong Kong. Stakeholders of the MiC project gave positive comments. For example, the owner of the project claimed that “The M gent 2091 0.01 platform using extended IFC has transformed our design optimization. The black box SBO is interpretable, traceable and exchangeable through open BIM. This enables non-professional stakeholders to understand the design intentions and make decisions. Technical users emphasized that “Through the adoption of open BIM, we can work smoothly between different software and disciplines with an open standard, which helps us streamline and automate many manual processes in LLM BIM version (input in Revit Agent and SBO. This has greatly increased our productivity by eliminating misunderstanding of design changes that happened in the black box SBO process. BIM version ( utput in Revit Figure 7. Embedded LLM Agent-enhanced SBO semantics in IFC Table 4. File size comparison of two strategies Strategy Minimal Maximum Average 5 Original (KB) 1939 1939 1939 Gzip & Base64 (KB) Δ (%) 1946 +0.36 1987 +2.48 1976 +1.91 External Ref. (KB) Δ (%) 1940 +0.05 1941 +0.10 1941 +0.10 Discussion and Conclusion The snapshot-based exchange paradigm of IFC introduces inefficiencies in data transmission, version control, and change impact analysis for making decisions in the collaborative, multi-disciplinary environment. To address these limitations, a modular version of IFC 5 is under development to support incremental IFC updates. Simulation-based optimization (SBO) for energyefficient building design exemplifies a typical designchange process which remains disconnected from open BIM-supported collaboration. The gap stems from IFC’s poor capability of representing information generated by SBO. To bridge the gap and illustrate the full provenance of such design changes, including the ‘why’, ‘how’, and ‘what’ in incremental IFC updates, this study proposes a novel IFC schema extension for the LLM Agentenhanced SBO. The contributions of this study are twofold: (1) An IfcSboExtension model was proposed and verified to enhance the IFC’s capacity to represent the LLM Agent-enhanced SBO process. 2092 (2) The feasibility and practical value of the proposed extension were validated through a MiC case study. The case study yields two key findings: (1) All information generated by the LLM Agentenhanced SBO is embedded within the extended IFC entities, involving three rooted IFC entities: IfcSboWorkflow, IfcAiAgentInteraction, and IfcOptimizationIteration; and two non-rooted IFC entities: IfcObjectiveFunction and IfcSboOptimizationDirectionEnum. (2) The changes of design parameters throughout the LLM Agent-enhanced SBO process can be captured as incremental IFC updates. Nevertheless, several limitations remain that require further investigation: (1) The extended IfcAiAgentInteraction faces a trade-off between semantic integrity and storage efficiency. If LLM Agent prompts and responses are fully encoded as structured data within IfcAiAgentInteraction, individual lines in the IFC file may exceed parser limits, resulting in increased file size and reduced exchange efficiency. Conversely, compressing the data (e.g., via Gzip) or referencing external files avoids file bloat, but incurs semantic opacity, requires stakeholders to employ additional tools to interpret the compressed string or external files. (2) Semantic changes of the extended IFC entities are not supported by the mainstream commercial BIM platforms (e.g., Revit) or open-source IFC parsing tools (e.g., IfcOpenshell). This limitation arises because the proposed IFC extension has not yet been submitted to buildingSMART for formal standardization – a necessary step for broader adoption. Acknowledgement The work presented in this paper was supported by the Hong Kong Research Grants Council (Nos. C708022GF and 17200221). References [1] C. Eastman. BIM handbook: A Guide to Building Information Modeling for Owners, Managers, Designers, Engineers and Contractors. John Wiley & Sons, 2011. [2] M. C. Georgiadou. An overview of benefits and challenges of building information modelling (BIM) adoption in UK residential projects. Construction Innovation, 19(3):298-320, 2019. [3] Y. Yu, S. Kim, H. Jeon, and B. Koo. A Systematic Review of the Trends and Advances in IFC Schema Extensions for BIM Interoperability. Applied Sciences, 13(23):12560, 2023. [4] buildingSMART International. IFC 5 Core Consultation. Online: https://www.buildingsmart.org/ifc-5-coreconsultation/, Accessed: 01/25/2026. [5] G. Mentzas, D. Apostolou, R. Young, and A. Abecker. Knowledge networking: a holistic solution for leveraging corporate knowledge. Journal of Knowledge Management, 5(1):94-107, 2001. [6] Q. Zhou and F. Xue. L-SCOPE: An LLM-assisted interactive platform for efficient building design optimization. Proceedings of the International Conference on Smart and Sustainable Built Environment (SASBE). 96-105. https://isasbec.abc2.net/index.php/sasbe/article/vie w/2554/9 [7] International Organization for Standardization. ISO 16739-1:2024 Industry Foundation Classes (IFC) for data sharing in the construction and facility management industries Part 1: Data schema. Online: https://www.iso.org/standard/84123.html, Accessed: 01/30/2026. [8] H. Liu, G. Gao, and M. Gu. A parallel IFC normalization algorithm for incremental storage and version control. 30th International Workshop on Intelligent Computing in Engineering (EG-ICE), pages 511-520, 2023. 2093 [9] K. Dietmair, S. Esser, and B. Harder. Incremental model updates in BIM: a critical analysis of the collaboration platform Speckle. Proceedings of the 35th Forum Bauinformatik 2024, 2024. [10] F. Xue and W. Lu. A semantic differential transaction approach to minimizing information redundancy for BIM and blockchain integration. Automation in Construction, 118:103270, 2020. [11] L. Kong, R. Zhao, C. Anumba, W. Lu, and F. Xue. Open BIM exchange on Blockchain 3.0 virtual disk: A traceable semantic differential transaction approach. Frontiers of Engineering Management, 12:510-528, 2025. [12] S. Esser, S. Vilgertshofer, and A. Borrmann. Graphbased version control for asynchronous BIM collaboration. Advanced Engineering Informatics, 53:101664, 2022. [13] I. Nonaka, and H. Takeuchi. The Knowledgecreating company: How Japanese companies create the dynamics of innovation. Oxford university press, 1995. [14] H. Wang, and X. Meng. BIM-supported knowledge management: Potentials and expectations. Journal of Management in Engineering, 37(4):04021032, 2021. [15] E. Feigenbau. Knowledge engineering. Annals of the New York Academy of Sciences, 426(1):91-107, 1984. [16] J. Beetz, J. Leeuwen, and B. de Vries. IfcOWL: A case of transforming EXPRESS schemas into ontologies. Ai Edam, 23(1):89-101, 2009. [17] J. Oraskari, and S. Törmä. RDF-based signature algorithms for computing differences of IFC models. Automation in Construction, 57:213-221, 2015. [18] S. Zhang, S. Zhang, H. Liu, X. Wang, Z. Zhao, C. Wang, and L. Yan. Integration of BIM and ontologies for pumped storage hydropower design change management in EPC projects. Automation in Construction, 175:106189, 2025. [19] Q. Zhou, and F. Xue. Automatic information gainguided convergence for refining building design parameters: Enhancing effectiveness and interpretability in simulation-based optimization. Building and Environment, 275:112788, 2025. [20] Y. Shen, and Y. Pan. BIM-supported automatic energy performance analysis for green building design using explainable machine learning and multi-objective optimization. Applied Energy, 333:120575, 2023.