Deep learning enabled particle analysis for quality assurance of construction materials.

Microspheres in fly ash are critically important for determining the properties and performance of fly ash dosed concrete, but facile and cost-effective analysis and quality control of fly ash microspheres remain difficult on construction sites. This paper describes a deep learning method to segment and analyze fly ash spherical particles using a simple optical microscope and a path aggregation network. The proposed method accurately detects microspheres and predicts their particle size distribution and volume fraction, outperforming traditional methods for particle analysis. The predicted results are directly linked to key properties that determine the quality of fly ash. This research establishes an automated and efficient method for rapid job-site fly ash spherical particle analysis, so that inexpensive and handy construction material quality control and assurance can be achieved for infrastructure construction. [Display omitted] • A simple microscope is presented to capture fly ash microspheres with all objects in focus. • A PANet model is trained and implemented for detecting fly ash microspheres. • The method accurately detects microspheres and predicts their particle size distribution. • Predicted results of the method are linked to properties determining fly ash quality. [ABSTRACT FROM AUTHOR]