Geometallurgical modeling of mining complexes: testing causal hypothesis to improve plant performance

需要金属和工业矿物等原材料来维持当前的社会需求。尽管在回收和再利用材料方面做出了努力，但发展中国家和向低碳经济转型需要新的采矿努力，以满足对金属和其他战略性商品，如稀土元素的需求。为了使采矿业尽可能可持续，原材料的开采必须有效和高效。为了实现这一目标，本研究项目提出与跨国钢铁制造公司安赛乐米塔尔合作，开发基于人工智能的预测工具。特别是，我们建议为每个单元操作建立因果模型，以便将采矿周期建模为一个系统，可以联合分析和优化。 因果模型是基于图形表示和概率的数学模型，有助于分析因果关系。流程的图形表示需要确定对响应有影响的输入因素，并找到这些变量之间的联系。然后，这些模型使用统计分析来计算事件的概率。可以提出一个可信的因果模型，并用矿场现有的数据进行检验。如果推理模型被认为是正确的，就可以分析反事实，也就是说，可以计算出未观察到的现象的概率。因此，这些模型允许决策和预测考虑到现有数据中没有观察到的情景的可能性。这种可能性将使行业受益，因为它提高了对过程和操作的理解，并允许预测从矿山提取材料的过程中每个单元操作的反应，从而建立一个相互关联的操作系统，该系统可以共同优化，以提高效率，减少水和能源的使用，并最大限度地减少废物产生。

Raw materials such as metals and industrial minerals are required to sustain the current societal needs. Despite efforts towards recycling and reusing materials, the developing nations and the transition to the low-carbon economy requires new mining endeavours to satisfy the demand for metals and other strategic commodities such as rare earths elements. To make the mining industry as sustainable as possible, the extraction of raw materials must be effective and efficient. In order to achieve this, this research project proposes the development of predictive tools based on artificial intelligence, in collaboration with ArcelorMittal, a multinational steel manufacturing corporation. In particular, we propose developing causal models for each unit operation, in order to model the mining cycle as a system, that can be jointly analyzed and optimized. Causal models are mathematical models based on graphical representations and probability that facilitate the analysis of causal relationships. A graphical representation of a process entails identifying the input factors that have an effect on a response, and finding the links between these variables. The models then use statistical analysis to compute probability of events. A plausible causal model can be proposed and tested with the data available at the mine site. If the inference model is deemed correct, counterfactuals can be analyzed, that is, the probability of an unobserved phenomenon can be calculated. Therefore, these models allow decisions and predictions accounting for the possibility of scenarios that are not observed in the available data. This possibility will benefit the industry by improving the understanding of the processes and operations, and by allowing the prediction of responses of each unit operation during the process of the material extracted from the mine, therefore building an interconnected system of operations, that can be jointly optimized to increase efficiency, reduce the use of water and energy and minimize the waste generation.