Adaptive classification of aggregate morphologies using clustering for investigation of correlation with contact characteristics of aggregates. (26th September 2022)

• Record Type: Journal Article
• Authors: Jin, Can; Cheng, Yang; Yang, Xu; Li, Shuqin; Hu, Jingran; Lan, Gongrui
• Is Part Of: Construction & building materials. Volume 349(2022)
• Publication Date: 2022-09-26
• Subjects: Aggregate morphology -- AP clustering -- K-means clustering -- Mixture microstructure; Building materials -- Periodicals; 624.18
• DOI: 10.1016/j.conbuildmat.2022.128802
• Languages: English
• ISSNs: 0950-0618

Abstract

Highlights: Clustering integrated with AP and K-means methods was used to classify aggregates in morphology. Morphological label was developed to reflect the statistical distribution of the shape properties for an aggregate cluster. The correlation between the composition of labeled aggregates and the contact characteristic of aggregates was studied. Abstract: To investigate the mechanism of the influence of aggregate morphologies on the internal structure of asphalt mixtures, an accurate classification of aggregates in morphology using an integration of affinity propagation (AP) and K-means clustering is proposed. To achieve this goal, the similarities of aggregates in angularity, surface texture, and form were evaluated and then used in clustering. Furthermore, morphological labels for aggregate clusters were constructed to indicate major features of the shape of aggregates in a cluster from the statistical perspective. 2, 766 coarse aggregates in three specimens were three-dimensionally (3-D) reconstructed and classified in three ways. Compared with two other ways, K-means clustering using the cluster centers derived from the AP method performs best in the qualitative indices including silhouette coefficient, Davies-Bouldin (DB) index, and the Dunn index. Furthermore, the correlation between the contact characteristic and the morphological grades of cluster labels in three specimens was analyzed, which shows that aggregates of cubic form, relatively rounded angularity with more fractured facets, and rich texture are more likely to have contact relations and larger contact regions. In addition, a pre-processing operation before clustering was adopted to reach a balance between the quality and efficiency of clustering. Therefore, the proposed method facilitates the optimization of the internal structure by the configuration of aggregate morphologies in future.