Abstract
Artificial intelligence (AI) stands as one of the most remarkable advancements in human history, with its integration spanning various sectors such as industry, healthcare, law enforcement, and finance. Despite this, the mining industry, from exploration to exploitation, lags in harnessing the potential of AI, especially in areas like monitoring underground operations and detecting anomalies such as seismic events and faults in mechanical systems. As the use backfill materials becomes imperative for mineral extraction, understanding and studying the diverse properties—mechanical, rheological, and physical—of these materials pose ongoing research challenges. Cemented paste backfill (CPB) emerges as the prevalent choice for mine backfill, with unconfined compressive strength (UCS) being a critical property defining its mechanical behavior. However, predicting UCS of CPB remains a formidable challenge.
Harnessing the wealth of big data generated by the mining industry presents an opportunity to develop intelligent systems. This research aims to develop an intelligent tool, leveraging Machine Learning (ML) and Deep Learning (DL), for predicting the UCS of cemented paste backfills. To achieve this, an extensive dataset was compiled from Agnico-Eagle's mines and augmented with laboratory data, resulting in a rich database (DB) of 10,050 CPB samples that consider the physical, chemical, and mineralogical properties of ingredients. Prior to model training, an exploratory data analysis was conducted. Intelligent models, including Random Forest (RF), Gradient Boosting Regressor (GBR), eXtreme Gradient Boosting Regressor (XGBR) and the Deep Neural Network (DNN), were employed. Based on performance indicators, the top-performing models, GBR and DNN, demonstrated coefficients of correlation (R) equal to 0.970 and 0.969, respectively. These models underwent validation in the laboratory through the preparation of new CPB mixtures.
To operationalize these models for the mining industry, a user-friendly web application was developed, ensuring accessibility and ease of use for industry professionals.