Standardized workflow to measure mineralogical composition and 3D geometry of particles


Standardized workflow to measure mineralogical composition and 3D geometry of particles

Gupta, S.; Da Assuncao Godinho, J. R.; Gotkowski, K.; Isensee, F.

Precise measurements of the mineralogical composition and 3D geometry of particles in mineral samples unlock the ability to systematically optimize ore processing procedures and thus paves the way for more efficient industrial ore processing and recycling of complex composite materials such as electronic waste. X-ray Computed tomography (CT) is a widely used method to acquire 3D images of such samples but so far lacks standardized methods to enable their interpretation. Here we introduce a new workflow to standardize the measurement of the 3D geometrical and mineralogical properties of particles. Importantly, our method is able to correct biases arising from partial volume imaging artefacts.

Specifically, our method consists of a combination of a deep neural algorithm known as nnU-Net [1], a state-of-the-art ready to use framework for segmentation of particles in the CT images, and MSPaCMAn [2], an automated method to extract precise mineralogical and geometrical properties on the particle level. We demonstrate that our method can be used out of the box to produce the particle segmentations independent of user biasness. These segmented images are used to calculate the 3D spatial properties of the particles including the mineralogical composition, surface liberation and a comprehensive list of geometrical properties. Results are validated using reference samples of known compositions. The proposed workflow is the first to enable a precise, unbiased and standardized semi-automated 3D analysis of particles using CT. The more comprehensive and standardized characterization is critical for the use of 3D particle properties in advanced ore processing techniques. Moreover, these 3D properties can be applied in the field of sedimentology for example to study the sediment transport and deposition.

Keywords: nnU-Net; MSPaCMAn; X-ray Computed Tomography; 3D characterisation

  • Lecture (Conference)
    Geoanalyses, 06.-12.08.2022, Freiberg, Germany

Permalink: https://www.hzdr.de/publications/Publ-35821
Publ.-Id: 35821