Image-Based Dimensional Measurement of 3D Printed Features

Sammanfattning

This thesis develops a practical, end-to-end framework for image-based dimensional measurement of additively manufactured parts, aimed at closing the “metrology gap” between slow, high-accuracy systems and fast but operator-dependent manual tools. Using a digital microscope and, for larger features, an optional standard camera the approach converts high-resolution images into millimetre-level measurements through a calibrated MATLAB pipeline that combines robust preprocessing (CLAHE), adaptive thresholding, and morphological refinement for reliable segmentation. A compact, custom-trained CNN acts as a feature router, classifying regions as circles, rectangles, or squares and dispatching them to shape-specific geometric routines. The workflow spans image capture, interactive calibration, automated measurement, accuracy evaluation against ground truth (calipers and CAD), and consolidated PDF reporting, and it includes a module for incremental finetuning to improve the classifier with new data. Design choices are justified with clear engineering rationale, and known challenges in optical metrology lighting variability, specular reflection, and threshold sensitivity are mitigated with explicit countermeasures. The validation programme quantifies performance in terms of accuracy, repeatability across sessions, and a formal measurement-uncertainty analysis aligned with international GUM guidance. The result is a fast, low-cost, and repeatable alternative suitable for integration into industrial quality-control workflows, documented to ensure transparency, reproducibility, and future extensibility.