Automatic BIM-SKEY Mapping Framework for Piping Spool Drawings: LightGBM-Based Component Group Classification Using Expert-defined BIM Features

This paper proposes a BIM-SKEY automatic mapping framework for piping spool drawing automation. The framework extracts 24-dimensional expert-defined features consisting of meta- data, geometry, and connector information from Revit family JSON data generated by a Revit API-based add-in. A LightGBM classifier classifies piping components into twelve groups. Model evaluation using Stratified 5-Fold Cross-Validation on 3,815 samples from 180 BIM models achieved 94.84% group classification accuracy without GPU resources, and the end-to-end mapping accuracy up to the SKEY symbol-pattern level reached 91.35%. To improve prac- tical applicability, a confidence-based routing strategy was introduced using the predict_proba output of LightGBM. High-confidence predictions were automatically applied, while uncertain cases were routed to engineer review or manual processing. The classified groups were assigned to SKEY symbol patterns using predefined mapping rules, after which the final SKEY codes were determined from component attributes and the Ez-ISO symbol definition file. A two-layer construction data integration model was also proposed to connect BIM design and construction progress data. Experimental results showed that the proposed framework improved the ISO drawing automation success rate from 58.3% to 91.8% and achieved 94.7% consistency between BIM design data and the construction data model, demonstrating its applicability to industrial BIM environments.