A new weakly supervised learning approach for real-time iron ore feed load estimation

Guo, L ORCID: https://orcid.org/0000-0003-1272-8480, Peng, Y ORCID: https://orcid.org/0000-0002-5508-1819, Qin, R and Liu, B (2022) A new weakly supervised learning approach for real-time iron ore feed load estimation. Expert Systems with Applications, 202. p. 117469. ISSN 0957-4174

Preview

Published Version Available under License Creative Commons Attribution Non-commercial No Derivatives. Download (21MB) | Preview

Abstract

Iron ore feed-load control is one of the most critical settings in a mineral grinding process. It has direct impact on the quality of final iron products. The setting of the feed load heavily replies the characteristics of the ore pellets. However, such characteristics are challenging to acquire in many production environments, requiring speical equipments and complicated modelling process with a high cost. To provide an low-cost and easier-to-implement solution, in this paper, we present our work on using deep learning models for direct ore feed load estimation from ore pellet images. To address the challenges caused by the large size of ore images and the shortage of accurately annotated data, we proposed to use a weakly supervised learning apporach with a two-stage model training algorithm and two neural network architectures developed. The experiment results show competitive model performance, and the trained models can be used for real-time feed load estimation for grind process optimisation.