A Study On Intelligent Dimensional Quality Inspection Of Building Component Using Point Cloud Data

Towards an efficient,high-quality and green development of construction industry,China is now developing prefabricated building and intelligent construction technologies,and promting the transformation and upgrading of the present construction industry through technical progress.To date,in China,dimensional quality inspection(DQI)(i.e.,appearance size and surface flatness)of precast concrete elements(PCEs)is usually conducted by manual inspection using rulers and feeler gauges,which are time-consuming,labor-intensive and subjective.Recently,a non-contact dimensional measurement technology based on three-dimensional(3D)laser scanners is gradually attracting the attention of researchers in the construction industry.3D laser scanners can scan the inspected object in all directions and provide massive 3D point cloud data(PCD)of the scanned object.The DQI results are then obtained easily by using inspection algorithm on the scanned PCD.Towards efficient DQI,it is thus of great significance to promote the technical progress of prefabricated buildings by proposing intelligent algorithms based on the PCD of building components(Building load-bearing components and decoration parts)to achieve automatic dimensional quality inspection.This thesis proposes a set of intelligent algorithms for automatic dimensional quality inspection of building components based on PCD.The main contributions and innovative achievements of this paper are as follows:(1)In order to improve the speed of dimensional quality inspection of prefabricated components(PCs),a data acquisition method which scans multiple PCs simultaneously and a segmentation algorithm that handles PCD of multiple PCs based on image processing technology are proposed.In particular,the proposed segmentation algorithm includes four steps: 1)PCD preprocessing,including ground data filtering,data slicing,etc.;2)image mapping,i.e.,converting 3D data into two-dimensional images;3)image segmentation and 3D PCD backtracking,including clustering,edge recognition and the proposed active window algorithm,etc.;(4)3D PCD segmentation,including the radially bounded nearest neighbor graph(RBNN)algorithm.The efficiency and effectiveness of the proposed segmentation algorithm for PCD of multiple PCs have been verified on 5 sets of outdoor PCD.(2)For the segmented PCD,a recognition algorithm for PCD of multiple PCs based on as-designed models in building information modeling(BIM)is proposed.The proposed recognition algorithm includes two steps: 1)an analysis method of the normal vector distribution,distinguishing PCs from the background.2)model matching evaluation,realizing the accurate recognition for the type of each PC.The accuracy and effectiveness of the proposed recognition algorithm for PCD of multiple PCs are verified by the PCD of three scanning scenes,including PCs with the same type but different shapes,PCs with the different types,PCs with the different shapes and types.(3)To achieve semantic segmentation of PCD of finished construction scenes,an indoor structural PCD dataset including 56 sets of training data and 13 sets of test data has been created.The neural network model Point CNN is trained using the created dataset and the semantic segmentation performance of Point CNN in the real scanned PCD are evaluated based on the test results.(4)For constructed building and component surface data,an automatic flatness quality inspection(FQI)algorithm and a visualization method of the FQI results based on color coded deviation map are proposed.According to the proposed automatic FQI algorithm,which considers the characteristics of large differences in surface dimensions between buildings and components,two different data preprocessing methods are proposed.With respect to the construction surface,surface data segmentation and plane fitting are used to create standard reference planes.For the component surface,surfaces of as-designed BIM model are considered as standard reference planes and a hierarchical search radius method is developed to improve the matching efficiency between the model and scanned data.The practicability and efficiency of the proposed automatic FQI algorithm and the visualization method of FQI results are verified by an unleveled constructed room and two sets of full-scale PCEs(5)An automatic dimension extraction algorithm based on PCD has been proposed.The proposed algorithm includes three steps: extraction of primary structure dimension,extraction of detailed structure dimension and the model visualization.In the extraction of primary structure dimension,the estimation methods are proposed for the corners in the PCD of PCs,including the multi-faceted corners,the double-sided corners and single-sided turning points caused by the occluded surfaces;in the extraction of detailed structure dimension,an adaptive estimation method based on PCD density is proposed for round holes in the PCs;in model visualization,a secondary development of Revit software is carried out for the transmission of dimension information of PCs and model generation.The proposed automatic dimension extraction algorithm based on PCD for prefabricated components are verified by two sets of PCD of a full-scale precast concrete staircase(including 80 edges and 4 round holes)acquired by two different scan resolutions(1/5 and 1/8).