Framework for the automated generation of 2-dimensional architectural drawings from building information models using deep learning

While building information modeling (BIM) is becoming more prevalent in the architecture, engineering, and construction industry, the necessity for 2-dimensional (2D) drawing generation persists, with a 41% extra effort currently needed for conversion from BIM. This study introduces an advanced framework that leverages deep learning to automatically convert BIM models into 2D architectural drawings, capable of applying multiple drawing styles. The study developed a hybrid architectural drawing recognition (Hyb-ADR) program, which employs detection and classification models to identify elements within 2D architectural drawings. Complementing this, a parametric algorithm further automates the stylization of these drawings. Validation on two reference drawings in different styles demonstrated an 81.85% accuracy rate for Hyb-ADR, and the parametric algorithm generated 2D drawings with two different styles from a BIM model successfully. The proposed framework is anticipated to significantly boost construction efficiency by facilitating the automated generation of a spectrum of 2D drawings from BIM models.