揭示循环荷载下级配碎石性能劣化机制对提高高铁路基服役性能具有重要意义.首先,基于机器视觉提出一种颗粒形状精细快速量化方法,通过自主搭建的图像采集装置构建颗粒图像数据集,选取U-Net、SegNet、PSPNet、DeepLabv3+4种典型语义分割模型进行颗粒分割实验,从准确性角度优选最佳模型并输出高精度二值图片,进而导入OpenCV(Open Source Computer Vision Library)快速量化颗粒形状.其次,开展室内循环加载试验,采用力学指标动刚度K表征级配碎石性能劣化特性,并通过颗粒形状量化方法明确劣化主控因素.最后,基于劣化主控因素建立不同劣化程度的双轴压缩DEM(Discrete Element Method)模型,揭示该主控因素对级配碎石性能劣化的细观影响机制.研究结果表明:U-Net模型的F1分数、平均像素准确率、平均交并比均高于其他模型,分别为98.03%、97.53%、97.02%,故将其选为最佳颗粒分割模型,并能在后续循环加载试验中准确分割颗粒形状.级配碎石性能劣化前后,粗颗粒等效粒径De、长细比Ei均无明显变化,而圆度Rc均值增大31%,故确定级配碎石性能劣化主控因素为粗颗粒研磨破碎.在双轴压缩DEM模拟中,随Rc增大,即颗粒研磨破碎程度增加,级配碎石偏应力、滑动率、强力链大小及占比、各向异性均逐渐减小,旋转量持续增大,粗颗粒嵌锁能力下降,导致颗粒骨架承载能力及结构稳定性降低,产生劣化现象.研究结果可为高铁路基长期服役性能评估提供理论依据.
Research on crushing mechanism and mechanical behavior of graded gravel based on machine vision
Revealing the deterioration mechanism of graded gravel under cyclic loading is crucial for improving the service performance of high-speed railway subgrade.First,a precise and rapid quantitative method of particle shape based on machine vision was proposed.A particle image dataset was constructed using a self-built image acquisition device.Subsequently,four typical semantic segmentation models,U-Net,SegNet,PSPNet,and DeepLabv3+were selected for the particle segmentation experiment.Then from the perspective of accuracy,the optimal model was selected to output high-precision binary images,which were imported into OpenCV(Open Source Computer Vision Library)to rapidly quantify the particle shape.Second,the indoor cyclic loading experiment was carried out to characterize the deterioration of graded gravel using the mechanical index dynamic stiffness K,and the key controlling factor of deterioration was determined through the quantitative method of particle shape.Finally,based on the key controlling factor of deterioration,biaxial compression DEM(Discrete Element Method)models with different degradation degrees were established to reveal the microscopic influence mechanism of performance deterioration of graded gravel.The research results indicated that the F1-score,average pixel accuracy and average intersection over union of U-Net model were 98.03%,97.53%and 97.02%,which were higher than other models.Therefore,U-Net was selected as the optimal particle segmentation model and was able to accurately segment particle shapes in subsequent cyclic loading experiment.After the performance deterioration of graded gravel,there was no significant change in the equivalent particle size De and aspect ratio Ei of coarse particles,while the average value of roundness Rc increased by 31%.Hence,it was determined that the key controlling factor of performance deterioration of graded gravel was the grinding of coarse particles.In the biaxial compression DEM simulation,as Rc increased,indicating an increase in the degree of particle grinding,the deviatoric stress,sliding rate,the value and proportion of strong contact force chain,and anisotropy of graded gravel gradually decreased,while the rotation amount continuously increased.As a result,the interlocking ability of coarse particles decreased,so the bearing capacity and structural stability of particle skeleton decreased,resulting in deterioration.The research results can provide a theoretical basis for the assessment of long-term service performance of high-speed railway subgrade.
万方数据
维普
