To capture temporal dependencies within the 1.5-hour windows, convolutional layers were incorporated into both encoder and decoder resulting in a Convolutional Variational Autoencoder (CVAE). Convolutional layers are particularly suitable for multivariate time-series data because they capture local temporal patterns and short-terms dependencies, extract hierarchical features across time, share weights across the temporal dimension, reducing parameter count, improve generalization compared to fully connected architectures (Theunissen, Bradshaw, Auret, & Muller, 2021). By combining convolutional feature extraction with probabilistic latent modeling, the CVAE effectively captures nonlinear, high-dimensional, and temporally dependent patterns present in geological and operational process data. 3.3.4. Gaussian Mixture Models (GMM) A Gaussian mixture is a probabilistic model that assumes observed data are generated from a combination of multiple Gaussian (normal) distributions, where each component of the mixture is defined by its own mean and variance (García, 2015). This approach enables the modeling of complex data distributions as the superposition of several underlying statistical subpopulations. Gaussian Mixture Models (GMMs) are primarily used for unsupervised clustering tasks, as they allow the identification of data subsets without requiring prior labels (Zhang, Jiang, Zhan, & Yang, 2019). Unlike deterministic clustering methods, GMMs assign posterior probabilities of membership to each component, providing a probabilistic interpretation of the resulting clusters. RESULTS 4.1. Clustering of variables The radar chart illustrates the characteristic profile of each cluster, integrating mineralogical, geochemical, and geomechanical competency variables. Each cluster represents a distinct mineral behavior pattern, highlighting clear differences in composition, hardness, and mechanical response.
RkJQdWJsaXNoZXIy MTM0Mzk2