A Three-Dimensional BGA Solder Ball Defect Detection Method Based on YOLOv8 and Metric Analysis

With the increasing integration and miniaturization of electronic devices, the detection of solder ball defects in ball grid array (BGA) packaging has become extremely important. In this study, a three-dimensional (3D) defect detection method for BGA solder balls was proposed. Computed tomography (CT) scanning was used to reconstruct internal 3D images of BGA chips. The YOLOv8 algorithm was utilized to develop a 3D target detection model. The void size ratio in the training dataset was adjusted to enhance sensitivity to void defects. This method identified void defects in 3D BGA images and generated candidate targets. A defect size measurement algorithm was designed to segment the internal voids in the solder balls. The void ratio was calculated to identify defects that met predefined criteria. The measurement algorithm was integrated into the dataset construction process. This integration of automated defect labeling reduced the workload of 3D annotation. Experiments were conducted on a 3D BGA chip image dataset. This method achieved high detection rates and low false detection rates. These results validate the effectiveness and reliability of the proposed method.