CGA @ C3DIS: Managing, Visualising, and Understanding Data for Coal Characterisation

June 30th, 2021

Sample PreparationIntroduction

Characterisation of the microscopic structure of coal is fundamental to understanding its chemical and physical behaviour. The highly heterogeneous nature of coal and the fact that the characteristics and relative distributions of the different material constituents of this resource will determine its rank and quality, make a deep understanding of the composition of each grain critical. Analysis of coal samples allows benchmarking of potential

The software we developed allows the automatic analysis of large coal images which provide reliable statistics on the distribution of coal types and impurities. It has been successfully used to analyse hundreds of coal samples and is now commercially available.

 

Software Features

Unique features of our software include:

  • Single grain characterisation
  • Enhanced particle separation
  • Support for large images
  • Statistical analysis of 100,000+ particles, as compared with approximately 500 particles in standard coal petrography methods.

 

Image acquisition and segmentation

After careful preparation of the sample a high resolution image is acquired. The image segmentation stage is summarised in the following steps:

  • Identification and removal of background, polishing and imaging artefacts
  • Automatic separation of touching grains based on shape
  • Removal of shades on the edge of the grains that are due to a difference of hardness between coal grains and the mounting resin
  • Computation of grain structural statistics such as Feret diameters, grain volumes and size distributions for the total sample.

System operation

 

Characterisation of Coal components

Characterisation requires identification of the different components in the coal grains. Due to the mixing of blends and difference in cutting angles during sample preparation, there is a significant benefit in estimating the distribution on a per-grain basis.

We employ a non-linear least square algorithm (Levenberg–Marquardt) to robustly fit the underlying distribution components on each grain and then use the data obtained from this algorithm to accurately characterise the grain.

 

System overview

 

Current and Future Work includes…

  • Continued commercialisation of the software
  • Getting additional support from industry
  • Working directly with various microscope file formats
  • Handling of larger image files up to 150GB
  • Extending the use of colour information for characterisation
  • Intra-component characterisation for individual particles
  • Enhanced automation of image analysis with the help of machine learning algorithms
  • Machine learning characterisation of dust particles using Optical dust markers
  • Blend partitioningFuture work
  • SEM integration
  • SaaS features
  • Ongoing optimisation:
    • GPU processing
    • Distributed processing
    • Improvements to core algorithms

 

 

 

 

 

Contributors: Paul McPhee, Gregoire Krahenbuhl, Graham O’Brien, Karryn Warren, Priyanthi Hapugoda, Silvie Koval