Title:
DIGITAL AGGREGATE: Analyzing Concrete Aggregate Physical Properties using Neural Network at Ready-Mix Plant
Author(s):
Hanmore
Publication:
Web Session
Volume:
Issue:
Appears on pages(s):
Keywords:
DOI:
Date:
11/3/2024
Abstract:
The physical characteristics of aggregates, such as shape (circularity, roughness, roundness, etc.) and size (minimum diameter, maximum diameter, aspect ratio, etc.), have a significant impact on the fresh and hardened state properties of concrete. Although only aggregate particle size distributions are regularly tested at ready-mix concrete plants by sieving of representative samples, morphological properties are not commonly considered when proportioning concrete mixtures. Some characteristics like flatness, elongation, and texture indexes of aggregates are still manually measured, however, this happens infrequently for standards compliance reasons and is not used as a reference point when proportioning concrete mixtures. These additional morphological properties of aggregate, that fluctuate constantly, remain unknown during concrete production, requiring additional safety margin to ensure the mix-design meets the fresh and hardened state requirements. To address this limitation, an innovative technique known as DIGITAL AGGREGATE is proposed, which involves installing a computer vision system at the ready-mix plant to capture and analyze images of aggregates falling into the weighing bin for a batch of concrete in real-time. A convolutional neural network (CNN)-based model is trained on the image dataset to predict the segmentation mask areas of the aggregates. Statistical metrics are applied for evaluating the model's training and prediction performances. This novel inline aggregate data offers an enhanced opportunity for optimizing cement paste proportions in real-time from batch to batch, reducing the need for high safety margins and field adjustments to meet fresh state requirements.