The application of digital twin machine learning models for Mine to Mill and Pit to Plant optimisation

Whilst the value of Pit to Plant and Mine to Mill optimisation has been proven since the 1990s, sustaining the productivity and efficiency gains typically requires ongoing test work, blast trials, plant surveys, and model calibration services. In contrast, the input to digital twin Pit to Plant or Mine to Mill optimisation is tens to hundreds of millions of tonnes of standard mine operating data. This paper presents a series of digital twin case studies from across the mine value chain, including machine learning testing methodology, model accuracy statistics and the application of Shapley Additive Explanations (SHAP) for model interpretation. These case studies demonstrate how machine learning and data fusion ore tracking connects hundreds of millions of tonnes of downstream performance data to produce digital twin models (operating models that are continuously updated and integrated into the wider Mine to Mill/Pit to Plant ore tracking system). Digital twin modelling software has been used to add block model variables for dig-rate, crusher and mill throughput, recovery (tailings grade), lump yield, product grade and product quality, fines yield and fines in lump. Block-by-block predictions are added to the geological resource and grade control block models as new block model variables. These new block model variables are used across all mine planning time horizons from life-of-mine planning to daily schedules. In addition to the materials handling and geometallurgical prediction case studies, the paper also presents digital twin case studies for drill and blast simulation and plant set point optimisation. The case studies show how digital twin simulation enables drill and blast engineers to simulate the effect of different blast designs in different blast domains using the mines operating data. Plant set point optimisation applies mathematical optimisation algorithms to the machine learning model to provide live operational decision support to metallurgists and plant operators. The digital twin machine learning case studies demonstrate a digital and sustainable paradigm for Mine to Mill and Pit to Plant optimisation.