Metrological validation of a deep learning pipeline for in-line detection and dimensional quantification of three-dimensional surface defects

Defect detection is critical in manufacturing processes such as casting, machining, and additive manufacturing, where imperfections can impair part functionality and reliability. This work proposes a deep learning-based methodology for the detection and dimensional inspection of three-dimensional surface defects. The approach integrates deep learning trained on pre-labelled data and applied to two-dimensional deviation maps derived from laser-based measurements, and a skeletonization algorithm to estimate defect dimensions. Accuracy is ensured by metrological validation using reference measurements from a multisensor coordinate measuring machine. Applied to die-cast components, the framework demonstrates robust performance, offering a reliable tool for integration into real-world quality control workflows. • Deep Learning pipeline for 3D defect detection and quantification in die-casting. • Dimensional assessment via skeletonization of CNN prediction of defects. • Metrological validation ensured by multisensor CMM measurement.