Commissioning a Process Twin for Setpoint Decision Automation at Dugald River Mine

Abstract

In January 2025, Dugald River Mine (DRM) successfully commissioned a closed-loop process digital twin for their zinc rougher scavenger flotation circuit. The outputs of the system are directly integrated back into the control layer, allowing the control system to autonomously optimise 16 flotation control setpoints across the circuit, while giving the operator oversight on the scope of control of the digital twin.

The objective of the twin’s optimisation is to continuously update the suite of control setpoints which maximise the DRM net smelter return function over a short interval. The technology commissioned by DRM contrasts with traditional rule-based (fuzzy logic) systems and linear matrix model-predictive control systems (MPCs) on several fronts. Firstly, the underlying process model is represented by a complete multivariate model of the rougher and scavenger which models the circuit grade-recovery position as a function of upstream drivers (geometallurgical and grinding circuit), and the operators decision variables. Mass balance modelling combined with neural networks were utilised for the multivariate models. Secondly, the internal control relationships (e.g. recovery vs collector dosage rate) within the model dynamically adjust for different upstream conditions (feed rates, grind sizes, grades, etc.) Allowing setpoints to be optimised for different feeds, rather than fixed rules or step-test
regressions in the case of MPCs. Thirdly, the system internally tracks its own health and performance and seamlessly passes control back to the operator when confidence reaches a critical threshold. In the first two months post commissioning, the efforts of the DRM technical and operations teams resulted in twin utilisation exceeding 90%. Learnings and developments from the post-commissioning period were focused on model error management and dynamic feedback, additional flexibility for the operator to define the twin’s operational envelope, and ongoing evaluation of the system’s grade-recovery performance.