Testing And Evaluation Method For Appearance Quality Of Fair-faced Concrete And Engineering Application

Fair-faced concrete is one-time formed and without decoration.The sense of the essential beauty of architecture are expressed by the combination of natural texture of concrete and designed seams and holes.China started the application of fair-faced concrete in the 1990 s,but the quality control and acceptance inspection standard for fair-faced concrete are not mature enough,especially for the acceptance and evaluation of appearance pore and chromatic aberration.And there is a lack of scientific appearance quality evaluation system.This thesis is based on National Natural Science Foundation of China(Integrated Structural-functional Cement-based Composites for Green Building and Intelligent Manufacture in Building Construction,51738003)to carry out the relevant research.The main aspects of the appearance quality of fair-faced concrete were addressed,including pore,color difference,cracks,etc.UAV(Unmanned Aerial Vehicle)and orbital large-format scanners were used for data collection of appearance quality information and the influencing factors were analyzed.Based on convolutional neural networks,an automated identification and analysis model was established and then a human-computer interaction system were constructed.The system provides an automated analysis and evaluation of the appearance quality of fair-faced concrete under different states,working conditions and environmental conditions,and finally provides an efficient and convenient big data analysis platform for fair-faced concrete workers and all concrete researchers as a reliable reference for the design and acceptance inspection.Through experiments,with the UAV camera sensitivity of 1600～3200,shutter time of 1/50～1/20 s,aperture value of 2.8～3.5,the collection of images is closer to the real situation.Image size,single shoot area,and distance from the surface are directly related to each other.To ensure the collection accuracy of not less than 1 mm/pixel,the distance between the UAV and the surface should not be greater than 4 m.The line connecting the center point of the single area and the lens should be perpendicular to the collection plane.The wind vector should be as small as possible in the direction parallel to the collection plane.Light intensity should be controlled at 2000～10000Lux.The concrete surface should be dry and the surface moisture content should not be greater than 5%.The collection method was adjusted accordingly for irregular surfaces,colored concrete and designed seams and holes.Based on the geometry correction and data sets preparation,a database of appearance quality information of fair-faced concrete was established.Based on the basic network structure of VGGNet,a pixel wise segmentation structure based on convolutional neural network was proposed and a pixel classification module was established to make pixel wise recognition of appearance pore,cracks,peculiarity targets and general surface.RGB and CIELab color systems were compared and the spatial conversion method was clarified.The chromatic aberration indexes based on CIELab was proposed.The model was optimized and fixed with a variety of performance evaluation indexes,combined with a variety of current neural networks of the highest level in the field.Error and robustness analysis was carried out with the main influencing factors.The results demonstrate that the model for both target recognition and chromatic aberration analysis are robust.The human-computer interaction system design and implementation were carried out with the principles of scientificity,accuracy,easy to understand and easy to operate by users.The architecture of the front and back-end interfaces was built and the analysis and evaluation system of appearance quality of fair-faced concrete was established.The collection method and the system were applied to Xiong’an Station of Beijing-Xiong’an Intercity Railway,Chaoyang Station of Beijing-Shenyang high-speed Railway,Nanjing-Jurong Intercity Rail Transit.And the efficiency and accuracy of the collection method,the highly automation and ease of operation of the system were verified.