Introduction
In the Architecture, Engineering, and Construction (AEC) industry, Building Information Modeling (BIM) serves as the digital backbone for project data. To ensure interoperability, the industry relies on Model View Definitions (MVD) to specify data exchange requirements. These rules are typically encoded in the mvdXML format. However, validating large-scale models against thousands of these rules often leads to significant performance bottlenecks, hindering the efficiency of digital workflows.
The Problem: Combinatorial Explosion
Traditional validation tools employ a "matching-then-checking" strategy. They first search for all possible subgraphs in a model that match a specific template and then verify if these subgraphs satisfy the associated rule statements.
As shown in Figure 1, the separation of templates and rules in mvdXML necessitates this two-step process. In complex models with millions of nodes, this approach results in a combinatorial explosion of potential subgraphs, making validation computationally expensive and memory-intensive. For large projects, this can turn a routine check into a process that takes hours.
The MVDLite Solution
MVDLite addresses these inefficiencies by shifting from subgraph matching to a unified filtering approach based on a rule chain structure. This method integrates constraints directly into the search process, allowing for "early pruning" of invalid data branches.
1. Rule Chain Structure
Instead of treating templates and rules as separate entities, MVDLite merges them into a continuous chain. This chain consists of three segment types:
- Attribute Segments: Define the navigation path through the IFC (Industry Foundation Classes) data structure.
- Metric Segments: Apply filters to specific values (e.g., checking data types or ranges).
- Compound Segments: Handle complex logical operations like AND, OR, and NOT.
2. Round-Trip Search and Deep-Caching
MVDLite utilizes a "round-trip" search algorithm that filters all relevant nodes simultaneously in a forward pass and identifies valid root entities in a backward pass. To further optimize performance, a deep-caching strategy is employed. By assigning unique string identifiers to rule chain prefixes, the algorithm can reuse results from previously processed paths, significantly reducing redundant computations across large rulesets. Mathematically, this shifts the complexity from $O(nmp)$ (where $n$ is entities, $m$ is subgraphs, and $p$ is attributes) to $O(e)$, where $e$ is the number of edges visited.
Performance and Results
The researchers evaluated MVDLite using three IFC models of increasing size—Duplex.ifc (52 MB), Office.ifc (193 MB), and B01.ifc (841 MB)—against rulesets containing up to 1,791 statements. MVDLite consistently outperformed existing industry tools like xBIM and BIMserver.
- Massive Speedup: On the 841 MB
B01.ifcmodel, the xBIM MVD plugin required over 145 minutes (8722.8 seconds) for validation. MVDLite completed the same task in just 73.2 seconds. - Scalability: Unlike traditional tools that exhibit exponential time increases with model size, MVDLite’s performance scales linearly, making it suitable for large-scale projects.
- Caching Impact: The deep-caching strategy proved highly effective, reducing processing time from 126.7s to 73.2s on the largest model, nearly doubling the algorithm's efficiency.
Conclusion and Future Work
MVDLite provides a high-performance alternative for BIM validation, turning a process that once took hours into one that takes seconds. While currently focused on structural and value-based validation, future work aims to extend this logic to complex geometric clash detection and partial model extraction. This lightweight representation offers a scalable path forward for ensuring data quality in increasingly complex digital construction environments.
Summary updated