Optimization of the SAG Grinding Process Using Statistical Analysis and Machine Learning: A Case Study of the Chilean Copper Mining Industry

Abstract

Considering the continuous increase in production costs and resource optimization, more than a strategic objective has become imperative in the copper mining industry. In the search to improve the efficiency in the use of resources, the present work develops models of a semi-autogenous grinding (SAG) mill using statistical analysis and machine learning (ML) techniques (regression, decision trees, and artificial neural networks). The hypotheses studied aim to improve the process’s productive indicators, such as production and energy consumption. The simulation of the digital model captures an increase in production of 4.42% as a function of mineral fragmentation, while there is potential to increase production by decreasing the mill rotational speed, which has a decrease in energy consumption of 7.62% for all linear age configurations. Considering the performance of machine learning in the adjustment of complex models such as SAG grinding, the application of these tools in the mineral processing industry has the potential to increase the efficiency of these processes, either by improving production indicators or by saving energy consumption. Finally, the incorporation of these techniques in the aggregate management of processes such as the Mine to Mill paradigm, or the development of models that consider the uncertainty of the explanatory variables, could further increase the performance of productive indicators at the industrial scale.

Prediction of energy consumption in grinding using artificial neural networks to improve the distribution of fragmentation size [Predicción del consumo de energía en la molienda utilizando redes neuronales artificiales para mejorar la distribución del tamaño de la fragmentación]

“…Figure 6 shows the scatter plot between the real values and the prediction of Wi (kWh/t), which shows a determination coefficient R2 of 98%. Saldaña, et al (2023), in their study, the adjustment through the application of artificial neural networks, for its part, turned out to be the model with the best adjustment indicators (MAE,RMSE and R2), where it obtained an R2 of 89%, whose architecture of the RNA corresponded to a multilayer perceptron. In this study, an R2 of 98% is a very strong indicator that the model is capable of making predictions very close to the real values of the variable Wi.…”

“…Otsuki and Jang (2022), in their study collected a total of 56 data sets and randomly divided into training data sets (40 data sets: 70%), validation (8 data sets: 15%) and training (8 data sets: 15%), for their grinding study, 56 experimental data sets were considered more than enough. Saldaña, et al (2023), in their research divided the historical data into two groups, that is, the training set (70%) and the validation set (30%), while the fitted model was used to estimate the production after the application of the M2M strategy and simulate production, at different values of the mill rotation speed and lining age factors. Like Azizi, et al (2020), in their study to investigate the application of three powerful Kernel-based supervised learning algorithms to develop a global model of the wear rate of grinding media based on input factors such as pH, percentage of solids, the loading weight of the balls and the rotation speed of the mill and the grinding time, the models were trained using 40 randomly selected data (representing 80% of the total data) and the remaining 10 data (which represent 20%) were applied for testing purposes.…”

“…The goodness-of-fit indicators are MAE, MSE, RMSE and R2. All indicators shown quantify how well a model fits a data set(Saldaña et al, 2023). Table4shows the adjustment and prediction capacity of the model to estimate the Bond index accurately, which is mainly evaluated by the RMSE and R2, which have reached a value of 1.182 and 1.181 respectively in training and a value of 0.984 and 0.984 in validation.…”

Machine Learning Algorithms for Semi-Autogenous Grinding Mill Operational Regions’ Identification

“…[18] for intelligence-based hybrid approaches, and Ref. [19] where a compelling case study is provided that leverages machine learning techniques to enhance SAG mill productivity indicators such as production and energy consumption. A great review on this subject can be found in [20], where a rigorous discussion on the topic is provided.…”

Development MPC for the Grinding Process in SAG Mills Using DEM Investigations on Liner Wear

“…Delaney et al [47] used various methods to implement predictive analytics techniques as a result the model estimates mass loss with particle evolution and included cumulative damage and destruction. Saldana et al [48] considered optimization of particle fracture processes using machine learning, which can significantly improve the quality of control.…”