Modular digital metrology architecture for legacy CNC machines using visual acquisition and PCA implemented via SVD

Abstract This study presents a modular digital metrology architecture for computer numerical control (CNC) machines with restricted access to internal controller variables. A microscope camera supports visual localization of geometric entities, while a second camera captures the coordinates displayed on the native human–machine interface (HMI). Optical character recognition (OCR), integrated into the external metrology platform, converts these coordinates into structured records that form three-dimensional measurement point sets. Geometric entities are estimated using principal component analysis (PCA) implemented via singular value decomposition (SVD) to obtain geometric indicators associated with straightness, parallelism, and perpendicularity. The evaluation included three reference artifacts, each with a corresponding calibration certificate, and one CNC-machined workpiece, with three replicates and five points per entity. For the straightness-associated indicator, the reference artifacts exhibited standard deviations between 0.0077 and 0.0080 mm, whereas the machined workpiece exhibited a standard deviation of 0.3007 mm. For the parallelism-associated indicator, the machined workpiece showed the largest mean deviation ( $$0.1310^{\circ }$$ ); for the perpendicularity-associated indicator, it exhibited a standard deviation of $$0.0076^{\circ }$$ , lower than that of two reference artifacts. Compared with off-machine inspection using coordinate measuring machines, approaches dependent on manufacturer-specific interfaces or software, operator-dependent manual measurement, and optical systems affected by environmental and setup conditions, the proposed approach allows the workpiece to remain on the CNC machine and does not require internal controller access, although it requires external visual hardware and operator-initiated acquisition. The results support the functional feasibility of the architecture and characterize within-operator variability without constituting a complete metrological validation.