Enhancing Panel Drawing Dimensioning in BIM Software using Recommender Systems
Simulainen, Johannes (2025)
2025
Tietotekniikan DI-ohjelma - Master's Programme in Information Technology
Informaatioteknologian ja viestinnän tiedekunta - Faculty of Information Technology and Communication Sciences
This publication is copyrighted. You may download, display and print it for Your own personal use. Commercial use is prohibited.
Hyväksymispäivämäärä
2025-05-02
Julkaisun pysyvä osoite on
https://urn.fi/URN:NBN:fi:tuni-202504304527
https://urn.fi/URN:NBN:fi:tuni-202504304527
Tiivistelmä
Dimensions specify crucial measurements for panel fabrication in panel drawing templates. The large number of available dimensions makes manual selection challenging, time-consuming, and inefficient. This thesis investigates the feasibility and effectiveness of using various recommender system techniques to simplify and accelerate this dimension selection process.
The research utilized a real-world dataset of panel drawing templates and dimensions from multiple Vertex BD user environments. Dimension parameters were represented using BERT embeddings to capture semantic similarities. Three core recommendation approaches (content-based filtering, collaborative filtering, co-occurrence analysis) and hybrid models combining them were implemented and evaluated. Performance was assessed using a custom template recreation simulation framework. This framework measured precision and recall under varying conditions, including user data availability, initial template completeness, and recommendation set size.
The results demonstrate that the optimal recommendation strategy is context-dependent. Collaborative filtering techniques performed best when sufficient prior user data was available, effectively handling complex templates. However, they performed poorly in cold-start scenarios. Content-based filtering proved most effective in cold starts due to its reliance on dimension features. On the other hand, it showed limited adaptability and suffered from redundancy issues in complex templates. Co-occurrence models generally performed between the other two approaches, but suffered from limited adaptability and significant redundancy issues in complex templates. The hybrid model combining content-based and collaborative filtering offered the most robust performance across diverse scenarios by leveraging the strengths of its components. All tested models demonstrated computational efficiency suitable for real-time application.
This study confirms the potential of recommender systems to enhance the panel drawing dimensioning task in BIM software. It offers a foundation for developing more intelligent selection tools. The findings highlight key trade-offs between different approaches, and the underlying recommendation techniques show potential for application to other selection tasks within the BIM domain, warranting further investigation.
The research utilized a real-world dataset of panel drawing templates and dimensions from multiple Vertex BD user environments. Dimension parameters were represented using BERT embeddings to capture semantic similarities. Three core recommendation approaches (content-based filtering, collaborative filtering, co-occurrence analysis) and hybrid models combining them were implemented and evaluated. Performance was assessed using a custom template recreation simulation framework. This framework measured precision and recall under varying conditions, including user data availability, initial template completeness, and recommendation set size.
The results demonstrate that the optimal recommendation strategy is context-dependent. Collaborative filtering techniques performed best when sufficient prior user data was available, effectively handling complex templates. However, they performed poorly in cold-start scenarios. Content-based filtering proved most effective in cold starts due to its reliance on dimension features. On the other hand, it showed limited adaptability and suffered from redundancy issues in complex templates. Co-occurrence models generally performed between the other two approaches, but suffered from limited adaptability and significant redundancy issues in complex templates. The hybrid model combining content-based and collaborative filtering offered the most robust performance across diverse scenarios by leveraging the strengths of its components. All tested models demonstrated computational efficiency suitable for real-time application.
This study confirms the potential of recommender systems to enhance the panel drawing dimensioning task in BIM software. It offers a foundation for developing more intelligent selection tools. The findings highlight key trade-offs between different approaches, and the underlying recommendation techniques show potential for application to other selection tasks within the BIM domain, warranting further investigation.