Physical separation processes are best understood in terms of the behaviour of individual ore particles. Yet, while different empirical particle-based separation modelling approaches have been developed, their predictive performance has never been tested under variable process conditions. Here, we investigated the predictive performance of a state-of-the-art particle-based separation model under variable feed composition for a laboratory-scale magnetic separation of a skarn ore. Two scenarios were investigated: one in which the mass flow of the different processing streams could be measured and one in which it had to be estimated from data. In both scenarios, the predictive models were sufficiently general to predict the process outcomes of new samples of variable composition. Nevertheless, the scenario in which mass flow could be measured was ∼4% more precise in predicting mass balances. The process behaviour of minerals present at concentrations above 0.1% by weight could be accurately predicted. Our findings indicate the potential use of this method to minimize the costs of metallurgical testwork while providing in-depth understanding of the recovery behaviour of individual ore particles. Moreover, the method may be used to establish powerful tools to forecast mineral recoveries for partly new ore types at a running mining operation.

物理分离过程最好根据单个矿石颗粒的行为来理解。然而，虽然已经开发了不同的基于经验的基于粒子的分离建模方法，但它们的预测性能从未在可变过程条件下进行过测试。在这里，我们研究了最先进的基于粒子的分离模型在可变进料组成下的预测性能，用于实验室规模的矽卡岩矿石磁选。研究了两种情况：一种是可以测量不同处理流的质量流量，另一种是必须根据数据进行估计。在这两种情况下，预测模型都足够通用，可以预测可变成分新样本的过程结果。尽管如此，可以测量质量流量的情况在预测质量平衡时要精确约 4%。可以准确预测浓度高于 0.1%（重量）的矿物的过程行为。我们的研究结果表明，这种方法的潜在用途是最大限度地降低冶金测试工作的成本，同时深入了解单个矿石颗粒的回收行为。此外，该方法可用于建立强大的工具来预测正在运行的采矿作业中部分新矿石类型的矿物回收率。我们的研究结果表明，这种方法的潜在用途是最大限度地降低冶金测试工作的成本，同时深入了解单个矿石颗粒的回收行为。此外，该方法可用于建立强大的工具来预测正在运行的采矿作业中部分新矿石类型的矿物回收率。我们的研究结果表明，这种方法的潜在用途是最大限度地降低冶金测试工作的成本，同时深入了解单个矿石颗粒的回收行为。此外，该方法可用于建立强大的工具来预测正在运行的采矿作业中部分新矿石类型的矿物回收率。