Research On Consistency Check Of BIM Model And Intelligent Component Identification

The construction field is facing the reform requirements of information digitization and life-cycle carbon emission management.The specific requirements are to meet the interactivity and coordination during the digital disclosure and acceptance of each stage of the project,and further form a life-cycle information flow closed-loop.In this process,the BIM model plays a key supporting role.The existing software in the field of building digitization uses IFC as an open data exchange format,but IFC components are often misclassified during the export process,which interferes with the accuracy of BIM model analysis and compliance review.Therefore,this article takes the IFC category information in BIM models as the research object,and combines the current research status of compliance in the construction field with information delivery issues in the BIM application process.The classification recognition algorithm based on Multi View Convolutional Neural Network(MVCNN)is applied to BIM model consistency check,and in-depth research is carried out.The main research work is as follows:(1)To reduce information ambiguity during the interaction process,ensure seamless exchange of BIM information throughout the entire cycle,and ensure the accuracy of the IFC semantic system,this article designs a consistency check scheme for BIM models.Improve the BIM model consistency check process by establishing a BIM element view collection submodule and an image category check submodule.At the same time,we will explore the core methods of the BIM element view collection sub module,utilizing technologies such as Dynamo and Blender to achieve batch extraction of BIM elements and automatic multi view image generation,improving the automated view collection process.(2)An IFC component classification and recognition model based on improved multi view convolutional neural network is proposed.The model introduces self attention module and short-term memory network.Aiming at the limitations of feature fusion of multi view convolutional neural network model,LSTM＿ATT is designed.This module fuses the input view data to obtain a 3D shape descriptor with clear identification,which helps to improve the classification and detection performance of similar IFC components.(3)Verify the classification performance of the intelligent component recognition method based on improved multi view convolutional neural network for 20 commonly used component categories in the construction field.The overall accuracy of this model on the overall validation set reached 88.27%,which is 9.46% higher than the original multi view convolutional neural network.The results indicate that the proposed method can significantly improve the efficiency of BIM model consistency check tasks.(4)Compared with traditional multi view convolutional neural network models,the improved multi view convolutional neural network model significantly improves the number of units for effective information capture,which helps to improve the classification and recognition efficiency of IFC components that are difficult to distinguish in the construction field.It benefits from the precise optimization of sub class information in the model,which enriches the semantics of BIM by supplementing sub class information,and meets the needs of pan professional information in the BIM application process.