Extracting BIM data to support a machine learning model for automated clash resolution

Abstract

Clash resolution is considered a critical step to resolve issues among the different disciplines for a construction design to be realized as expected. This step, however, continues to remain slow and manual which can significantly delay a project and drive-up costs. A combined machine learning model was proposed by Harode and Thabet (2021) to automate the clash resolution process. A large amount of labeled dataset is required to train and test the proposed model. The dataset is planned to be extracted from various industry-provided federated construction BIMs. Federated construction models are created from multiple subcontractor component models authored using different software. As a result, data is stored in various formats using different data structures making the extraction process difficult. In this paper, the authors demonstrate the use of commercially available software tools including iConstruct, Dynamo, and Talend to overcome this limitation and extract the necessary data. The paper first defines the required data structure followed by a data extraction process to capture required data from clashing elements in the federated BIMs. The paper also discusses a novel method of extracting end point coordinates and moveable area for clashing elements using bounding boxes. The paper concludes with future research directions.