Abstract:Two-dimensional (2D) random mesostructure models have been broadly utilized to study mechanical behavior and damage mechanisms of asphalt concrete due to their high availability and efficiency. The 2D aggregate gradation that characterizes the area fractions of the simulated aggregates of different sizes significantly affects the simulated mesomechanical properties of asphalt concrete. However, there are very few methods for determining the 2D aggregate gradation from the actual three-dimensional (3D) one. This study proposed a 2D aggregate gradation conversion framework integrating a 3D random aggregate method and machine learning models. First, 3D asphalt concrete mesostructures were generated using polyhedral coarse aggregates with random gradations and shapes. Virtual cutting and sieving procedures were then developed to obtain the 2D aggregates at sections with 1 mm intervals in each mesostructure and acquire the area fractions of aggregates of different sizes, respectively. By averaging the results for the sections in each cutting direction, so-called 2D representative area fractions of aggregates within different grading segments in different directions were obtained. The gradation conversion data sets were established by collecting the volume fractions of coarse aggregates in the 3D mesostructures as input variables and the corresponding 2D representative aggregate area fractions as output variables. Based on the data sets, machine learning models, including linear regression, support vector regression, bagging of regression tree, and neural network (NN), were trained using a 5-fold cross-validation approach and then tested. The results showed that the NN model generally provided the most accurate predictions of the 2D representative aggregate area fractions. The effectiveness of the proposed framework was validated by using both the test sets regarding different sizes of random mesostructures and a mesostructure rebuilt from computed tomography (CT) images of actual asphalt concrete.

Deep learning-based software and hardware framework for a noncontact inspection platform for aggregate grading

A Coarse Aggregate Gradation Detection Method Based on 3D Point Cloud

Aggregate particle identification and gradation analysis method based on the deep learning network of Mask R-CNN

Deep Learning‐based Automatic Classification of Three‐level Surface Information in Bridge Inspection

Point Cloud-Based Dimensional Quality Assessment of Precast Concrete Components Using Deep Learning

A Deep Learning-Based Surface Defect Inspection System Using Multiscale and Channel-Compressed Features

A Smart Surface Inspection System Using Faster R-CNN in Cloud-Edge Computing Environment

Rapid pavement aggregate gradation estimation based on 3D data using a multi-feature fusion network

Interactive Defect Segmentation in X-Ray Images Based on Deep Learning

Rapid extraction of pavement aggregate gradation based on point clouds using deep learning networks

Research on Intelligent detection of concrete aggregate level based on monocular imaging

Research on Real-time Detection of Stacked Objects Based on Deep Learning

2D Aggregate Gradation Conversion Framework Integrated with 3D Random Aggregate Method and Machine-Learning for Asphalt Concrete

Intelligent Analysis Method of Dam Material Gradation for Asphalt-Core Rock-Fill Dam Based on Enhanced Cascade Mask R-CNN and GCNet

Automatic railroad track components inspection using real‐time instance segmentation

A deep transfer learning-based algorithm for concrete surface defect detection

A deep learning approach for construction vehicles fill factor estimation and bucket detection in extreme environments

Deep Learning-Based Generic Automatic Surface Defect Inspection (ASDI) with Pixelwise Segmentation

Sand gradation detection method based on local sampling

A systematic framework for DEM study of realistic gravel-sand mixture from particle recognition to macro- and micro-mechanical analysis

Research on Nondestructive Testing Technology for Drilling Risers Based on Magnetic Memory and Deep Learning