Controller-Independent Monitoring of CNC Operating States Using Non-Intrusive Net Input Current Signatures

Legacy computer numerical control (CNC) machines often lack controller connectivity or sensing suites, making it difficult to monitor utilization and energy demand or deploy data-driven maintenance without mechanically invasive modifications. This study develops a black-box method that links net input current signatures to CNC operating states. Four current-sensor channels recorded 70,000 samples per second each, generating approximately 504 million samples across 44 repetitions of an unloaded Haas Mini Mill cycle containing dwell periods and bidirectional motion of the spindle and each linear axis. Cycle-to-cycle variability supported semiautomated, sequence-guided transition detection, while frequency-domain signatures were evaluated across similarity measures, baseline compensation, window functions, overlap, multicycle aggregation, and multichannel Bayesian analysis. The active cycle was distinguishable from a powered-idle baseline, and the reconstructed sequence represented 39.99 s of the measured 40.94 s cycle. Cross-correlation produced stronger intra-dataset screening results than Pearson correlation under the magnitude-based procedure; baseline compensation increased the Bayesian screening score from 36.1% to 81.0%. With a rectangular window and positive-match evidence across all four channels, the conservative minimum estimated confidence reached 95% across the eight states. This value describes segmented-signature distinguishability within the dataset, not independently tested classification accuracy. The results provide proof-of-concept evidence for controller-independent monitoring under one fixed unloaded program and a basis for broader experimental validation.