Niranjan Adhikari holds a double Masters in both Statistics and Physics.
Niranjan is a PhD student at the JKMRC working on a project of the Advanced Process Prediction and Control group under Dr Gordon Forbes, Associate Professor Mohsen Yahyaei and Dr Marko Hilden.
Niranjan completed his undergraduate and Masters' degrees in theoretical physics. After this, he worked as a physics lecturer in Purbanchal University, Nepal, for more than four years. Niranjan completed a further Masters degree in Statistics and Operations Research in 2017 where he specialised in Machine Learning. In 2018, he worked with Advantage Data consulting company as a Data Scientist, before commencing his PhD in early 2020, with the working title:
Machine Learning Algorithms for the Calibration of Comminution Models.
Empirical process models are frequently used in the mineral processing industry for designing circuits, selecting equipment, testing circuit configurations, and for process control. Many of these models are used in flowsheet simulation packages such as JKSimMet. For each model, there are often multiple parameters, some of which need to be calibrated. Furthermore, it is important to re-calibrate these models (updating the model parameters) in a timely manner when the processing conditions of the circuit, as well as ore properties, have changed.
In order to calibrate these models, one typically does a survey to collect relevant data then fits the model using the simulation package JKSimMet. This end-to-end process can take up to six months to complete. The existing calibration procedure is costly and time-consuming as it has to go through multiple processes such as plant survey, sample analysis, and model fitting, and therefore a better solution is required.
Instead, this project is utilising plant operating data (i.e. PI data) to extract essential information to calibrate the process models through a machine learning technique as guided by mineral process expert knowledge. However, there is still a need to do a base case survey to be able to develop an initial calibrated model, but after that, the goal is to update the model automatically based on the current operating regime identified using PI data.
The proposed approach can improve the accuracy of the prediction of process models by extending their prediction range and reduce the burden of frequent expensive surveys. This capability will put integrated flowsheet simulation in a unique space as it provides real-time prediction capability of the process models.