Quick extraction of recycled sand morphology parameters based on deep learning and their effect on mortar property

Abstract

Due to the limited application of the natural river sand in China, recycled sand (RS) has been considered as an important replacement of natural sand in sustainable construction. The morphology of recycled sand is critical to its effective application in mortar and concrete. An environmentally friendly approach is introduced in this work to improve the morphology of recycled sand during its crushing process. This paper presents a newly developed image acquisition device, which obtained 25,328 RS images to use as a dataset for the construction of a segmentation model based on deep learning technology with an automatic segmentation of RS characteristics. Evaluation indices including accuracy (ACC), Recall, intersection over union (IoU), and F1-score index are reaching up to 99.8%, 88.1%, 84.9% and 84.3%, respectively. RS morphological characteristics including the flat and elongation ratio (FER), angularity index (AI), roundness (R) are automatically extracted by the proposed model, and the predicted results agreed well with the experimental ones. Finally, the effect of FER of RS on mortar’s mechanical properties is investigated by a series of experiments. Results reveal that the flowability, flexural strength and compressive strength of mortar decreased invariably with the lower FER of sand.

Keywords

• Recycled sand
• Morphological characteristics
• Automatic segmentation
• Deep learning technology
• Mortar property