IM November/December 2025 | 页面 23

GOLD PROCESSING necessity; it directly determines mill throughput, recovery efficiency and energy use, according to Tenzin Worden, Data Scientist at Eclipse Mining Technologies. The right blend can boost grinding efficiency and improve leaching performance, whereas poor blending can result in lower recoveries, wasted energy and lost revenue. Similarly, well-managed stockpiles act as buffers between the mine and the mill, smoothing out variability in feed quality and increasing operational stability.
Historically, blending decisions have relied on average grades or simplified models, leaving operations vulnerable to variability in mineralogy and ore quality, Worden says. The complexity of interactions between gold grade, sulphur content and recovery efficiency mean averages alone are rarely accurate predictors of these relationships. To establish a solid baseline, the process must begin with trusted sampling at the plant entry point. Yet, even when good samples exist, data integrity is not guaranteed. It is common to encounter problematic data, such as samples collected during plant downtime, misread instruments, or values that fall well outside the expected range. Without careful validation, such data can distort reconciliation and undermine decision making.
By turning raw data into actionable intelligence, SourceOne EKPS enables predictive modelling and optimisation that closes the loop between mine and mill, Eclipse says
This is where AI-enabled validation becomes essential, according to Worden.
“ The rise of sensors, cloud computing and AI has opened a new frontier in mining,” she says.“ Instead of relying solely on static models, operations can now access real-time predictive insights that adapt to changing conditions. Big data and AI don’ t just provide numbers; they create generalised solutions from patterns in the data, helping miners optimise decisions as ore properties shift.”
But not all AI is created equal. Generalpurpose models, such as large language models, may dominate headlines, yet they often lack the precision required for domain-specific challenges in mining. Instead, the real breakthroughs lie in combining statistical modelling, simulation and reinforcement learning( RL), a trio of methods that together can represent

Experience

SourceOne ® EKPS
SourceOne ® is an adaptable knowledge system. It enriches data with context, history, and digital workflow automation, empowering the foundation for AI; giving you the accurate answers you’ ve been looking for.
eclipsemining. com