Clasificación estructurada de la ASTM Uniformat II de elementos BIM usando información semántica y Natural Language Processing

Abstract

This project aims to develop an automated method for classifying construction elements in BIM models using natural language processing (NLP) and the ASTM Uniformat II standard. The hypothesis suggests that advanced techniques such as NLP and machine learning models significantly improve the accuracy and efficiency of BIM element classification. The project’s importance lies in optimizing construction processes through standardization and reducing data management errors. For the development, the BERT model was fine-tuned to classify elements under the ASTM Uniformat II standard, and Dynamo was used for data extraction and organization. Techniques included the integration of deep learning algorithms, semantic data processing, and result validation using confusion matrices. Materials consisted of BIM models created in Revit, with analyses performed using Python and specific NLP libraries. The results showed that the BERT model achieved an accuracy of 48.77% at Level 1, but its performance decreased at more detailed levels, reaching only 4.33% accuracy at Level 4. These findings highlighted the model’s limitations in granular classifications. It is concluded that automated models improve classification at general levels but require adjustments to enhance precision in specific levels. This reinforces the relevance of natural language processing in BIM information management. Keywords: BIM, Uniformat II, natural language processing, automated classification.