Irregular characteristic analysis of 3D particles—A novel virtual sieving technique

Abstract

For digitizing the granular material systems, including asphalt mixture, the sieving size of the particles plays an irreplaceable role. However, the traditional methods based on equivalent conversion performed terribly in measuring the size of individual particles. This study combined digital image processing, ensemble learning, and over-sampling algorithms to develop a novel sieving method. Besides, the comprehensive importance ranking of variables was put forward to assist in optimizing the random forest. The results show that the data collection method proposed in this study can quickly gain bulks of digital aggregates without a complicated reconstruction process. The synthetic minority over-sampling algorithm was conducive to the accuracy ascension by balancing the database. By scoring variables' relative importance and function, the variable selection method can effectively reduce system complexity, thus improving computational efficiency and robustness. Based on the above-improving processes, the measuring precision of the aggregates' sieving opening was raised by over 20%.