【作者】 王鑫；

【导师】 高英； 曹荣吉；

【作者基本信息】 东南大学 ， 交通运输工程（专业学位）， 2022， 硕士

【摘要】 油石比是评价沥青路面施工质量的关键指标之一,其数值波动直接影响到沥青混合料各性能指标,因此在施工阶段,需对沥青混合料油石比有一个及时且较为精确的检测。目前工地上多采用离心抽提法对油石比进行检测,该方法检测精度不足,需使用大量具有一定毒性和挥发性的三氯乙烯,不利于环境保护和人员健康。本文基于图像处理技术和随机森林算法,对沥青混合料油石比检测方法进行研究,为探索一种新的油石比检测方法做出一定尝试。首先在沥青混合料油石比检测方法研究中,选取不同级配、不同油石比的沥青混合料:AC-13(油石比分别为3.76%、4.54%、5%、5.5%、6%)、AC-20(油石比为3.7%、4.2%、4.7%、5.2%、5.8%)、AC-25(油石比为3%、3.3%、3.8%、4.3%、4.8%),对各沥青混合料击实成型标准马歇尔试件并进行CT扫描。依据CT获取的多张沥青混合料断面数字图像,采用阈值分割法进行图像分割。利用Avizo软件识别了马歇尔试件中部截面的空隙与集料,分析了马歇尔试件中部空隙面积、空隙平均直径、空隙率等特征的分布规律。其次,基于马歇尔试件中部的空隙与集料信息,应用相关系数对特征进行筛选,并结合随机森林算法对特征重要性进行了排序。为探究随机森林算法预测的准确性,将随机森林算法、决策树算法、梯度提升树算法、线性回归算法四种算法的预测结果进行了对比,最终建立了基于随机森林算法的油石比检测模型,对随机森林算法预测结果进行了分析。最后,采用Avizo软件对室内试件与施工阶段马歇尔试件的切割图像进行了空隙与集料识别,基于获取的空隙信息与集料信息,应用基于随机森林算法的检测模型,对室内试件与施工阶段沥青混合料的油石比数值进行了预测,分析了基于随机森林算法的油石比检测模型在实际工程中的可行性,结果发现预测结果与实际数据之间的绝对误差一般在±0.3%以内。换言之,采用图像处理技术对沥青混合料油石比进行检测是可行的,满足工程应用的需要。更多还原

【Abstract】 Asphalt-aggregate ratio is one of the key indicators to evaluate the quality of asphalt pavement construction,and its numerical fluctuation directly affects the performance of asphalt mixture.Therefore,it is necessary to have a timely and accurate detection of asphalt-aggregate ratio in the road construction.At present,centrifugal extraction method is mostly used to detect asphalt-aggregate ratio in construction sites,which is not accurate enough and requires the use of a large amount of trichloroethylene with certain toxicity and volatility.So,it is not conducive to environmental protection and human health.In this paper,based on the image processing technology and random forest algorithm,asphalt-aggregate ratio detection method has been studied.This study has made a certain attempt to explore a new method of the detection of asphalt-aggregate ratio.Based on this,this paper carries out research on the detection method of asphalt-aggregate ratio and asphalt-aggregate ratio fluctuation of asphalt mixture based on asphalt-aggregate ratio detection method.Firstly,in the study of asphalt mixture asphalt-aggregate ratio detection method,asphalt mixtures of different gradations and different asphalt-aggregate ratios are selected: AC-13(3.76%,4.54%,5%,5.5%,6%),AC-20(3.7%,4.2%,4.7%,5.2%,5.8%),AC-25(3%,3.3%,3.8%,4.3%,4.8%).The standard Marshall specimens are shaped by compaction and CT scanning is carried out.According to the digital images of asphalt mixture cross-section obtained by CT,the thresholding method is used for the image segmentation.Avizo software is used to identify voids and aggregates in the middle section of Marshall specimens.The distribution of the void area,mean void diameter and void fraction in the middle section of Marshall specimens are analyzed.Secondly,based on the information of voids and aggregates in the middle of Marshall specimens,correlation coefficient is used to screen the features and the random forest algorithm is used to rank the importance of features.In order to explore the accuracy of the prediction of the random forest algorithm,the prediction result of the random forest algorithm,decision tree algorithm,gradient lifting tree algorithm and linear regression algorithm are compared.Finally,the prediction model of asphalt-aggregate ratio based on the random forest algorithm is established,and the prediction result of the random forest algorithm are analyzed.Finally,Avizo software is used to identify voids and aggregates in the cutting image of Marshall specimen in the laboratory test and construction stage.Based on the obtained voids information and aggregates information,the prediction model based on the random forest algorithm is used to predict the asphalt mixture asphalt-aggregate ratio in the laboratory test and construction stage.The feasibility of the prediction model of asphalt-aggregate ratio based on the random forest algorithm is analyzed.It shows that the absolute errors between the prediction result and the actual values is generally within ±0.3%.In other words,it is feasible to use image processing technology to detect the asphalt mixture asphalt-aggregate ratio and meet the demands of engineering application.更多还原