Unlocking Tool Wear Information from Spindle Current Signals: An explorative Energy-Based Analysis using the NASA Milling Dataset

Reliability of milling processes is affected by accurate tool wear estimation due to its direct impact on product quality, energy consumption, process stability and, indirectly related to productivity and costs. In CNC milling, the progressively worn tool increases friction and this results in major mechanical resistance with a consequence of higher energy requirements on fixed cutting conditions. In this context, the continuous monitoring of energy-related signals, captured by sensors installed directed on CNC spindle, represents a non-invasive solution to derive Physical-informed Machine Learning (PiML) approach for a predictive production. It can derive wearing condition by alternative managing operational condition and production (as per material and task) schedule. This paper, starting from the NASA Prognostics Center of Excellence (PCoE) Milling Wear dataset, investigates the relationship between spindle electrical current signals, considering both the direct current (DC) and alternating current (AC) components, and tool flank wear (VB) in cast iron and steel bar milling operations. The separation of the DC and AC components enables modulation of the electrical signal in order to distinguish the contribution of the average energy load, associated with cutting conditions, from the dynamic variability of the process, related to tool-workpiece interaction phenomena and load fluctuations. The steady-state cutting phase is captured to avoid influence of transient effects associated to approach/ejection movement to material removal. From the AC and DC signals, by using Principal Component (PC) decomposition, authors generated the global energy behaviour associated with wearing progression. Results highlight a relationship DC-based indicators and flank wear.