Research Work

My research interest is to study the soil–rock mixture characteristics that constitute mine overburden dumps. Overburden dumps are formed by stacking a massive amount of earth material (soil–rock mixture) in a limited land area extracted during surface mining operations of coal and other ore deposits. This earth material consists of particles ranging from clay to big boulders. With the increase in opencast mines’ production capacity, millions of cubic meters of overburden are removed every year. Therefore, maintaining such a massive structure’s stability is essential for the proper functioning of mining activities.

The global objective of this research is to predict the shear strength of dump material from images. Field studies, laboratory experiments and computational works are designed to achieve this global objective. The initial step is to develop a robust algorithm that predicts an on-site particle size distribution from the dump images. An Artificial Intelligence (AI) based technique, Mask R CNN, a state-of-the-art in instance segmentation, was implemented to locate and classify each pixel of every particle in an image.

At the dump site, field experiments are conducted to determine the prominent factors that influence the shear strength of the dump material. A porosity profile was established to obtain the in situ porosity of dump materials from dump height. 

In the laboratory, dump samples are segregated into five groups based on their size ranges. The experiments generate a database of shear strength parameters. Four parameters, including the scale and shape of curves, intact density, and moisture content as input variables, and c and φ as target variables, are compiled into a database of tests conducted in the laboratory. Gaussian process regression, a probabilistic machine learning-based modelling technique, is implemented to predict c and φ along with the associated uncertainties for the new values of input parameters. 

To execute the entire workflow, the AI model predicts particle size distribution coefficients from an input image and the porosity profile yields in situ density. Subsequently, the aforementioned values are used to obtain the shear strength parameters of a dump slope.