This thesis investigates the use of Convolutional Neural Networks (CNNs) for the automatic detection of electronic components on industrial circuit boards, within a real manufacturing context at Leonardo S.p.A.. The proposed approach explores several state-of-the-art object detection models, including You Only Look Once (YOLO), RetinaNet, and Faster R-CNN with Feature Pyramid Networks. These models are tested using Python-based tools and libraries such as Detectron2 and Ultralytics YOLO, with special attention to the use of oriented bounding boxes (OBBs) for handling rotated components—a common challenge in industrial imaging. The results demonstrate the feasibility of deploying such models for automated visual inspection in industrial environments, significantly contributing to the enhancement of quality control processes in manufacturing.
Questa tesi indaga l'utilizzo di reti neurali convoluzionali (CNN) per il rilevamento automatico di componenti elettronici su schede a circuito stampato industriali, in un contesto di produzione reale presso Leonardo S.p.A.. L'approccio proposto esplora diversi modelli di rilevamento di oggetti all'avanguardia, tra cui You Only Look Once (YOLO), RetinaNet e Faster R-CNN con reti piramidali di feature. Questi modelli vengono testati utilizzando strumenti e librerie basati su Python come Detectron2 e Ultralytics YOLO, con particolare attenzione all'utilizzo di bounding box orientati (OBB) per la gestione di componenti ruotati, una sfida comune nell'imaging industriale. I risultati dimostrano la fattibilità dell'implementazione di tali modelli per l'ispezione visiva automatizzata in ambienti industriali, contribuendo in modo significativo al miglioramento dei processi di controllo qualità in produzione.
A comparative study of object detection approaches for quality control of printed circuit boards
Di Girolamo, Loris
2024/2025
Abstract
This thesis investigates the use of Convolutional Neural Networks (CNNs) for the automatic detection of electronic components on industrial circuit boards, within a real manufacturing context at Leonardo S.p.A.. The proposed approach explores several state-of-the-art object detection models, including You Only Look Once (YOLO), RetinaNet, and Faster R-CNN with Feature Pyramid Networks. These models are tested using Python-based tools and libraries such as Detectron2 and Ultralytics YOLO, with special attention to the use of oriented bounding boxes (OBBs) for handling rotated components—a common challenge in industrial imaging. The results demonstrate the feasibility of deploying such models for automated visual inspection in industrial environments, significantly contributing to the enhancement of quality control processes in manufacturing.File | Dimensione | Formato | |
---|---|---|---|
2025_07_DiGirolamo.pdf
accessibile in internet per tutti
Dimensione
72.01 MB
Formato
Adobe PDF
|
72.01 MB | Adobe PDF | Visualizza/Apri |
I documenti in POLITesi sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
https://hdl.handle.net/10589/239898