Condition-Robust Tool Remaining Useful Life Prediction under Highly Variable Milling Conditions

Accurate prediction of tool Remaining Useful Life (RUL) is essential for reliable and cost-effective milling, particularly when machining commercially pure titanium (CP-Ti), where tool wear is highly irregular. In industrial practice, continuously varying cutting conditions further complicate tool condition monitoring and life prediction. This paper proposes a vibration-based monitoring framework for RUL prediction under strongly variable milling conditions. A hybrid deep learning model based on CNN–BiLSTM is developed to capture the non-stationary relationship between vibration signals and tool degradation. Performance is compared between a spindle-mounted, non-invasive sensor and a tri-axial accelerometer mounted on the machine table, and the benefit of sensor fusion is assessed. Results show that spindle vibration correlates strongly with tool degradation and achieves predictive performance close to that of multi-sensor configurations, while requiring minimal instrumentation. The proposed approach remains robust under variations in both operating conditions and wear mechanisms, enabling reliable RUL estimation in non-stationary milling environments.