Abstract
Advancements in data science, machine learning, and artificial intelligence are revolutionizing the mining and metals industry, making it more algorithm-intensive. The paradigm shift is moving from detection and control to prediction and optimization, where advanced soft sensors play an essential role. While conventional soft sensors were mainly based on linear and physical models, modern machine learning techniques provide an opportunity to improve commonly used soft sensors and develop new types that couldn't be achieved through conventional methods.
This paper will discuss two industrial case studies on advanced soft sensor applications in a pelletizing plant. The first one is a soft sensor that estimates the cold crushing strength (CCS) index of pellets, helping operators make better decisions in real-time to improve pellet quality. The economic and environmental value propositions of the CCS soft sensor application have been studied. The results demonstrate that improving pellet CCS leads to an increase in profitability and a reduction in emissions in both the pellet plant and downstream Direct Reduction Iron (DRI) plant. The second soft sensor predicts the amount of fired pellet FeO%, which incurs a penalty if it goes beyond the target. An economic analysis shows the potential for revenue improvement if the FeO% soft sensor is implemented in a 5 MTPA pellet plant.