• Image-based framework for machining stability monitoring. • Chip features extracted from contour-based video analysis. • Statistical indicators quantify chip-metric separability. • Multiple thresholding methods tested for state classification. • Average chip count with IQR threshold reaches AUC = 0.97. This paper proposes a camera-based measurement approach for the evaluation of chip formation and machining conditions in CNC turning. The objective of the study is to assess whether image-based metrics can provide reliable and statistically separable indicators of different chip formation states using a low-cost, non-invasive measurement setup. Controlled turning experiments were conducted in which the feed rate (0.0839–0.165 mm/rev) and depth of cut (0.0936–3.05 mm) were systematically varied to generate ideal and non-ideal chip formation states. The machining process was recorded using a fixed-position industrial camera, and chip contours were extracted from video frames using edge and contour detection. The outputs consisted of image-based chip metrics, including count- and area-based measurement quantities. These quantities represent the measurement layer, while statistical classification and thresholding constitute a subsequent decision layer. The discriminative capability of these quantities was evaluated using effect size analysis and receiver operating characteristic (ROC) analysis, the area under the curve (AUC), in combination with data-driven thresholding methods. The results reveal statistically significant differences between the defined chip formation states. Count-based quantities exhibited the highest robustness and statistical separability, with the average number of chips providing distinction between the two states. When combined with an interquartile-range-based threshold, this metric achieved the highest performance (AUC = 0.97, true positive rate = 0.95, true negative rate = 1.00). These findings demonstrate that image-based chip quantities can serve as effective quantitative indicators derived from an optical measurement process, supporting process monitoring and early detection of unstable cutting conditions in CNC machining environments.
Filep et al. (Sun,) studied this question.