Deep Learning-Based Framework for Reconstruction and Optimisation of Building Information Models Containing Parametric Rules

The building information model (BIM) is maturing as a new paradigm for storing information and exchanging knowledge about a building. BIM models precisely show existing structural elements which can be conveniently extracted to support condition monitoring, quality assessment and design optimisation of engineering structures. However, current practice does not fully address the integration of 3D modelling and computational intelligence to support the automatic reconstruction and optimisation of BIM models. To resolve this challenge, this paper presents a framework for deep learning-based reconstruction and optimisation of BIM, with a case study to demonstrate its application for the parametric design optimisation of high-rise buildings. The proposed framework involves the development and application of deep neural network for automated reconstruction of content-rich building models subject to certain design rules. After obtaining the geometric details about a building, the performance of the generated 3D models is assessed towards identifying the optimum solution. In this study, attempts have been made to optimise the wind flow of a high-rise residential building, as the wind plays an important role in designing a high-rise. The findings provide a deeper understanding and interesting insights into 3D reconstruction and optimisation of BIM subject to parametric design rules.