Sieveless particle size distribution analysis of particulate materials through computer vision

Abstract

This paper explores the inconsistency of “length-based separation” by mechanical sieving of particulate materials with standard sieves, which is the standard method of particle size distribution (PSD) analysis. We observed inconsistencies of length-based separation of particles using standard sieves with manual measurements, which showed deviations of 17–22 times. In addition, we have demonstrated the “falling through” effect of particles cannot be avoided irrespective of the wall thickness of the sieve. We proposed and utilized a computer vision with image processing as an alternative approach; wherein a user-coded Java ImageJ plugin was developed to evaluate PSD based on length of particles. A regular flatbed scanner acquired digital images of particulate material. The plugin determines particles lengths from Feret's diameter and width from pixel-march method, or minor axis, or the minimum dimension of bounding rectangle utilizing the digital images after assessing the particles area and shape (convex or nonconvex). The plugin also included the determination of several significant dimensions and PSD parameters. Test samples utilized were ground biomass obtained from the first thinning and mature stand of southern pine forest residues, oak hard wood, switchgrass, elephant grass, giant miscanthus, wheat straw, as well as Basmati rice. A sieveless PSD analysis method utilized the true separation of all particles into groups based on their distinct length (419–639 particles based on samples studied), with each group truly represented by their exact length. This approach ensured length-based separation without the inconsistencies observed with mechanical sieving. Image based sieve simulation (developed separately) indicated a significant effect ( P < 0.05) on number of sieves used in PSD analysis, especially with non-uniform material such as ground biomass, and more than 50 equally spaced sieves were required to match the sieveless all distinct particles PSD analysis. Results substantiate that mechanical sieving, owing to handling limitations and inconsistent length-based separation of particles, is inadequate in determining the PSD of non-uniform particulate samples. The developed computer vision sieveless PSD analysis approach has the potential to replace the standard mechanical sieving. The plugin can be readily extended to model (e.g., Rosin–Rammler) the PSD of materials, and mass-based analysis, while providing several advantages such as accuracy, speed, low cost, automated analysis, and reproducible results.