In the electronics/PCBA industry, BGA/QFN packaging technology is widely used, but the inspection of hidden solder joints has always been a difficult problem. A leading electronics/PCBA manufacturer faced problems such as high false alarm rates and missed detections in this process, which seriously affected product quality and production efficiency.
In the electronics/PCBA industry, BGA (Ball Grid Array) and QFN (Quad Flat No - leads) are common packaging forms, which can effectively save space and improve electrical performance. However, the inspection of hidden solder joints under these packages is a critical process but full of challenges. In actual production, hidden solder joints are prone to defects such as cold soldering, short - circuits, and missing solder balls. Traditional inspection methods, such as manual visual inspection and rule - based automated optical inspection (AOI), have many drawbacks. Manual visual inspection is inefficient, prone to fatigue, and has limited ability to identify tiny defects. Rule - based AOI has difficulty adapting to complex and variable solder joint shapes, resulting in high false alarm rates and frequent missed detections, which seriously affect product quality and production efficiency.
Technical Principle
DaoAI's AI-AOI system combines 2D and 3D inspection equipment with advanced AOI software and uses deep learning algorithms to solve the inspection problem of hidden solder joints under BGA/QFN packages. In terms of imaging, 2D imaging can clearly capture the planar features of solder joints, while 3D imaging can obtain three - dimensional information such as the height and shape of solder joints. The combination of the two provides more comprehensive solder joint data. In terms of hardware, high - precision cameras and light source systems ensure clear and accurate images. At the algorithm level, DaoAI uses APDT positive sample/few - shot learning technology. With only 1 - 20 good product images, an accurate detection model can be quickly trained. At the same time, the DaoAI World model can perform semantic analysis on the features of solder joints to achieve semantic false alarm filtering, greatly improving the accuracy of detection.
- Deep learning algorithm: Trained with a large amount of solder joint image data, the model learns the normal and abnormal features of solder joints to accurately identify defects.
- 2D/3D imaging fusion: 2D imaging provides planar information, and 3D imaging supplements three - dimensional information. The combination improves the comprehensiveness of detection.
- APDT positive sample/few - shot learning: Reduces the dependence on a large number of defective samples and quickly establishes an effective detection model.
- DaoAI World model: Performs semantic understanding of solder joint features and filters out false alarms caused by environmental factors and noise.
- Micron - level accuracy: Can detect tiny solder joint defects to ensure high - precision detection.
DaoAI's technology provides a more accurate and efficient solution for the inspection of hidden solder joints under BGA/QFN packages.
In terms of solution implementation, DaoAI provided a complete AI-AOI system for the manufacturer, including 2D/3D inspection equipment and AOI software. First, using APDT positive sample/few - shot learning technology, the detection model can be trained in a short time with only 1 - 20 good product images. Second, the system has the ability to change models in 5 minutes without coding. When producing different product models, it can quickly adjust the detection parameters without complex programming operations. During the inspection process, the 2D/3D inspection equipment comprehensively scans the hidden solder joints under BGA/QFN packages to obtain the image data of the solder joints. The AOI software uses deep learning algorithms and the DaoAI World model to analyze the images, identify defects such as cold soldering, short - circuits, and missing solder balls, and reduce false alarms through the semantic false alarm filtering function.
By implementing DaoAI's solution, the manufacturer has achieved significant quantitative results. In terms of the false alarm rate, it has been reduced by 80% compared with traditional inspection methods, greatly reducing the manual re - inspection workload caused by false alarms. The missed detection rate is controlled within 1%, effectively preventing defective products from flowing into the next process. The detection rate has reached over 99%, ensuring high - quality products. At the same time, the ability to change models in 5 minutes without coding improves the flexibility and efficiency of production, enabling the manufacturer to quickly respond to market demand.