<![CDATA[Manufacturing quality management depends on timely detection of nonconformities, reliable process measurement, and corrective action before defective units advance through production. This study presents a descriptive comparison of manual visual inspection and AI-vision inspection records from a machine-component factory. The supplied dataset covers five production lines over twelve months, comprising six manual-inspection months and six AI-vision months, together with a before-and-after defect-category dataset. The analysis validates 60 line-month records and calculates weighted defect density, percentage reduction in defect density, mean detection-time reduction, first-pass-yield change, and defect-category reduction. During the manual period, 363,845 units and 5,562 defects produced a weighted defect density of 15,286.73 PPM. During the AI-vision period, 357,506 units and 1,195 defects produced 3,342.60 PPM. The observed comparison corresponds to a 78.13% lower weighted defect density, an 87.51% shorter mean detection time, and a 4.28-percentage-point higher mean first-pass yield. Every production line recorded a lower weighted defect density in the AI-vision period. The before-and-after defect-category records show the largest percentage reduction for surface cracks, 86.57%, followed by dimensional nonconformity, 85.86%. Four source-derived figures distinguish line-level defect-density changes, monthly quality trends across the inspection transition, defect-category shifts, and the association between detection time and defect density. The study does not establish that the inspection technology alone caused the observed differences because the records are an observational pre/post comparison without controls for product mix, operator effects, or other concurrent changes. Its contribution is a transparent quality-management evidence base for identifying where deeper process validation is needed]]>
