Abstract

In mining, the detection of overload conditions in SAG mills is of great relevance to guarantee their operational continuity due to their economic and environmental impact. Various authors have tried to use Machine Learning techniques to identify the relationship between the variables and the underlying overload phenomenon. Using a combination of techniques integrated into a framework, we seek to establish a model that learns and detects overloads, taking care of aspects such as selecting variables, the generation of an encode that maximizes the learning using a Gram’s matrices approach, and that consider the imbalance of the classes to training a Convolutional Neural Network. Our proposed framework allowed us to establish a mechanism that statistically exceeds the metrics presented by other authors and opens an interesting space of exploration for the continuous improvement of predictive models.