Investigation of Classification and Anomalies Based on Machine Learning Methods Applied to Large Scale Building Information Modeling

Building Information Models (BIM) capable of collecting and synchronizing all the data related to a construction project into a unified numerical model consisting of a 3D representation and additional metadata (e.g., materials, physical properties, cost) have become commonplace in the building sector. Their extensive use today, alongside the increase in experience with BIM models, offers new perspectives and potentials for design and planning. However, large-scale complex data collection leads to two main challenges: the first is related to the automatic classification of BIM elements, namely windows, walls, beams, columns, etc., and the second to detecting abnormal elements without manual intervention, particularly in the case of misclassification. In this work, we propose machine learning for the automated classification of elements, and for the detection of anomalies based on geometric inputs and additional metadata properties that are extracted from the building model. More precisely, a Python program is used to decipher the BIM models (available as IFC files) for a series of complex buildings, and three types of machine learning methods are then tested to classify and detect objects from a large set of BIM data. The approach is tested on a variety of practical test cases.