439 Just-in-Time (ODJIT) system enables a detailed multivariate characterization of the material processed (Muñoz, 2022). To address the limitations of deterministic and linear clustering methods, this work proposes a Deep Clustering framework integrating a Convolutional Variational Autoencoder (CVAE) with a Gaussian Mixture Model (GMM). The CVAE extracts nonlinear and temporally structured latent representations from multivariate time-series data, while the GMM provides probabilistic clustering in the latent space. The proposed methodology captures geological, geometallurgical, and geomechanical variability in a unified framework, enabling robust ore characterization and supporting operational decision-making in SAG milling. 2. GEOLOGICAL AND OPERATIONAL CONTEXT The Toromocho mine, located in central Peru, is a copper–molybdenum porphyry system associated with Miocene intrusions and surrounded by skarn and hornfels units. The deposit exhibits significant mineralogical and metallurgical heterogeneity due to multiple hydrothermal alteration stages and structural overprinting (Ramos, Recines, Alcala, & Muñasqui, 2023). Geologically, the system comprises two main types of mineralization: the primary porphyry Cu– Mo system and a district-scale network of Cordilleran-type polymetallic veins that overprint the porphyry. The porphyry developed through multiple hydrothermal stages, evolving from early potassic alteration to quartz–molybdenite-rich veins and later sericite–quartz–pyrite assemblages (Bendezú, 2007). This geological complexity generates strong variability in ore competence parameters (BOND, DWI, HARD), directly influencing SAG mill performance. Figure 16 - Geological profile of the Toromocho mining deposit. (Ramos, Recines, Alcala, & Muñasqui, 2023)
RkJQdWJsaXNoZXIy MTM0Mzk2