Deep Learning-Based Detection of Shield Presence in Industrial Laser-Marking Applications

This paper presents the development and implementation of a computer vision-
based system aimed at improving the reliability and operational safety of an industrial laser-marking process applied to electronic boards. The investigated process involves a board
containing six integrated circuits, over which metallic shields must be correctly positioned
prior to the laser-marking stage to protect sensitive internal components. If a shield is absent
during this operation, the laser may directly affect the chip surface, leading to irreversible
damage, functional degradation, and significant production losses.
To mitigate this risk, an automated visual inspection system was designed to verify,
in a non-invasive manner, the presence or absence of the shield on each individual chip
before authorizing the laser-marking process. The proposed solution was implemented using
the Python programming language and was based on deep learning techniques for object
detection. Specifically, a YOLO-family model in its nano configuration was employed and
trained to classify two distinct conditions: shielded chips and unshielded chips. The use of
a compact model enabled the efficient execution of standard computing hardware,
achieving an average inference time of approximately 300 ms. Although industrial processes
do not impose strict real-time or cycle-time constraints, this inference latency is considered
adequate for reliable integration into the production workflow.
Experimental results demonstrate that the proposed system operates consistently in
an industrial environment, effectively preventing improper laser-marking operations. The
solution acts as an additional layer for fault prevention, contributing to enhanced process
robustness, improved product quality, and increased overall reliability of the manufacturing
system.