Abstract
Vision-based diameter estimation systems are widely used during Czochralski growth of monocrystalline silicon, primarily for real-time process control. Although these systems also provide data for downstream activities, such as supporting resistivity measurements, their geometric accuracy has not been rigorously validated. This study investigates the reliability of diameter estimation by experimentally comparing theoretical three-dimensional reconstructions—obtained by revolving diameter profiles around the pull axis—with high-resolution 3D scans of real silicon ingot segments. Five ingots, differing in nominal diameter and length and originating from different crystals, were analysed under industrial conditions. Real geometries were acquired using structured-light 3D scanning and aligned with the corresponding theoretical models to enable point-wise deviation analysis. This approach allows a quantitative assessment of diameter estimation accuracy. In addition, the results show that such reconstructions can be effectively used to calibrate dimensional control actions and to support preliminary evaluations of surface roughness.