Predictive maintenance of semi-autogenous grinding (SAG) mills reduces unplanned downtime and improves throughput. This study develops a data-driven prognostic model for production SAG mill using four years of operational data (temperature, voltage, current, motor speed, etc.). We follow a MATLAB (R2025a)-based prognostics and health management (PHM) workflow: data cleaning and synchronization; feature engineering in time and frequency domains (statistical moments, spectral power, bandwidth); normalization and clustering to separate operating regimes; and labeling of run-to-failure sequences for a recurring electrical failure mode. A health indicator is derived by scoring candidate features for monotonicity, trendability, and prognosability and fusing them into a condition index. Using MATLAB Predictive Maintenance Toolbox, we train and validate multiple Remaining Useful Life (RUL) learners including similarity-based, regression, and survival models on run-to-failure histories, selecting the best via cross-validated error and prediction stability. On held-out sets, the selected model forecasts RUL consistent with observed failure dates, providing actionable lead time for maintenance planning. The results highlight the practicality of deploying a PHM pipeline for SAG mills using existing plant data and commercial toolchains.