Digital twin with automatic disturbance detection for an expert-controlled SAG mill

TL;DR

The paper addresses the need for proactive SAG-mill control by developing a digital twin that fuses fuzzy expert control, a discrete-time regulatory model, and a NARX-based mill simulator to predict key variables $y_1$ (bearing pressure) and $y_2$ (motor power) from inputs $u_1$, $u_2$, and $u_3$. An automatic disturbance-detection framework triggers retraining of the regulatory and SAG-mill components when predictive performance degrades, ensuring continual adaptation using data from 68 hours of operation for training and 8 hours for validation. The digital twin achieves mill-prediction horizons of 2.5 minutes with errors below 5% (and about 1% for the 30-second horizon), demonstrated under synthetic disturbances such as liner wear and ore-hardness changes. The approach provides a foundation for real-time, proactive optimization in SAG-mill operations, maintaining robustness against process variations without requiring changes to the expert controller.

Abstract

This study presents the development and validation of a digital twin for a semi-autogenous grinding (SAG) mill controlled by an expert system. The digital twin integrates three key components of the closed-loop operation: (1) fuzzy logic for expert control, (2) a state-space model for regulatory control, and (3) a recurrent neural network to simulate the SAG mill process. The digital twin is combined with a statistical framework for automatically detecting process disturbances (or critical operations), which triggers model retraining only when deviations from expected behaviour are identified, ensuring continuous updates with new data to enhance the SAG supervision. The model was trained with 68 hours of operational industrial data and validated with an additional 8 hours, allowing it to predict mill behaviour within a 2.5-minute horizon at 30-second intervals with errors smaller than 5%.

Figures (3)

• Figure 1: Proposed supervisory system architecture, featuring a digital twin and an automatic disturbance detection module. The 'real system' illustrates the closed-loop operation of the actual SAG mill, while the 'digital twin' simulates this behavior in real-time. Subscripts: P denotes prediction error, and T denotes training error.
• Figure 2: Digital twin simulation results: (A-B) test data and predictions; (C-D) prediction errors; (E-F) histogram of prediction errors; (G-H) prediction error intervals.
• Figure 3: Proportional errors of bearing pressure prediction and automatic learning detection indicator for two disturbance scenarios: (A) mill liner wear after 5 months of operation, (B) 10% increase in ore hardness.