At present,the threshold segmentation method of traditional image processing technology is adopted in the two-dimensional pore recognition of permeable concrete,the recognition accuracy is unstable and the efficiency is low,and only one or two independent machine learning models are used in the prediction of permeability and mechanical properties.the prediction accuracy is limited.The mix design only takes the general index of porosity as the design parameter,fails to consider other characteristics of the pore structure,and can not achieve the double-objective optimization of water permeability and mechanical properties to form a good balance.This paper conducts deep research on the above three key issues.Firstly,the standard image data set of permeable concrete is obtained and constructed by CT scanning technology and image enhancement technology.By using the image processing technology based on depth learning,the optimal depth learning model of permeable concrete is obtained by training and verifying the model on the standard data set.At the same time,an efficient and accurate automatic analysis method of two-dimensional pore structure of permeable concrete is established,and then the three-dimensional reconstruction of permeable concrete is carried out by using three-dimensional image reconstruction and analysis software,and the threedimensional characteristics of permeable concrete are fully excavated and analyzed.Finally,a systematic and comprehensive characterization parameters of two-dimensional and threedimensional pore structure of permeable concrete are formed.Then,based on the depth learning optimal model,an automatic analysis method for particle recognition and gradation of mixed aggregate is established.Then,90 groups of permeable concrete test blocks with different ratios and different porosity are designed and prepared systematically,and 90 groups of slice image sequences are obtained by CT scanning,2D and 3D characterization data are obtained based on depth learning technology,and then the measured permeability coefficient and compressive strength are obtained through experiments to form the data set of 90 groups of permeable concrete blocks.Furthermore,big data technology is used to establish the prediction model of multi-parameters of pore structure and the relationship between permeability and mechanical properties of permeable concrete,and the non-dominant sorting genetic algorithm with elite strategy is selected for double-objective optimization to obtain the optimal Pareto solution set.at the same time,several common machine learning algorithms and traditional empirical formulas are compared.Based on the optimization of pore structure parameters and ensuring the permeability,a double-objective optimization strategy is proposed to maximize the mechanical properties.The research results of this paper include the establishment of automatic identification method of pore structure of permeable concrete,the prediction model of the relationship between pore structure and macroscopic performance of permeable concrete,and a doubleobjective optimization scheme of permeability and mechanical properties is established to guide the mix proportion optimization design of permeable concrete.The research results have certain theoretical guiding significance and engineering application value. |