This paper presents the development and evaluation of an automated visual
inspection system applied to the analysis of battery solder joints in an electronic welding
process. In automated manufacturing environments, the quality of soldered connections is
critical in ensuring the electrical reliability, mechanical stability, and long-term
performance of electronic assemblies. Defects in the soldering process may lead to
intermittent connections, premature failures, or complete malfunctions of the final product,
making effective inspection mechanisms essential.
The proposed system focuses on the visual inspection of battery solder joints using
image-based classification techniques. Images acquired during the welding process were
analyzed using a convolutional neural network (CNN) trained to classify solder joints into
two distinct categories: acceptable (OK) and non-acceptable (NG). The inspection is
performed using a vision system based on a standard camera, allowing for the non-contact
and non-destructive evaluation of the solder quality.
The artificial intelligence model was trained and executed using a backend
application developed in Python that handled image acquisition, preprocessing, and
inference. A graphical user interface was implemented using Windows Forms in C# to enable
operator interaction, visualization of inspection results, and integration with the production
environment. This software architecture allows for a clear separation between the machine
learning processing layer and human–machine interface.
The experimental results indicate that the proposed system can reliably distinguish
between conforming and nonconforming solder joints under real production conditions. The
solution demonstrates the feasibility of applying convolutional neural networks for
automated inspection in electronics manufacturing, contributing to improved process
control, enhanced product quality, and reduced dependency on manual inspection.