3D AI AOI: APDT Few - sample Self - training for BGA Void Detection

Wemio's 3D AI AOI device (self-developed 3D camera + 3D topography reconstruction/point cloud, detecting hidden solder joints, coplanarity, micron-level topography, air holes and other 2D optical blind-zone defects, 2D-3D fusion) reduces the false-alarm rate of BGA void detection by -50% through APDT few-sample self-training.

In the electronics/PCBA industry, BGA (Ball Grid Array) is a common and important packaging form. However, the detection of internal void defects in BGA has always been a difficult problem. Wemio's 3D AI AOI device, with APDT few-sample self-training technology, can accurately detect BGA voids and reduce the false-alarm rate of detection from a high level by -50%. A certain electronics manufacturing company's production line mainly produces various PCBA boards, among which a large number of chips with BGA packaging are involved. Traditional detection methods cannot meet its requirements for detection accuracy and efficiency, and there is an urgent need for a more advanced detection method.

Pain Points: Why Is It So Hard?

From the perspective of quantitative dilemmas, firstly, the miss-detection rate is relatively high. Under traditional methods, the miss-detection rate of BGA voids can reach about 5%. This means that a considerable number of unqualified products may enter the market, bringing potential quality risks. Secondly, the false-alarm rate remains high, exceeding 60%. A large number of false alarms force workers to spend a lot of time on manual re-judgment, and the manual re-judgment working hours account for more than 30% of the total detection hours. Moreover, the machine shutdown for model change is frequent. Whenever detecting different models of BGA products, traditional equipment needs more than 30 minutes for debugging and calibration, which seriously affects the production rhythm.

The reasons for these dilemmas mainly lie in multiple aspects. In terms of process, the internal structure of BGA packaging is complex, and the positions and sizes of voids are random. Traditional 2D detection technologies are difficult to obtain the real topographic information inside, and it is impossible to accurately judge the void situation. In terms of imaging, 2D optical imaging is easily interfered by factors such as surface reflection and shadows, resulting in poor imaging quality and affecting the detection results. In terms of materials, the material characteristics of BGA packaging also increase the difficulty of detection. Different materials have different reflection and absorption characteristics of light, making it difficult for traditional imaging to clearly present the internal structure. From the perspective of production rhythm, the low efficiency of traditional detection methods cannot meet the needs of high-speed production, resulting in overly long machine shutdown time for model change.

Technical Principle

Wemio's 3D AI AOI device uses a self-developed 3D camera for data collection. Through 3D topography reconstruction and point-cloud technology, it can accurately obtain the 3D structure information of BGA. At the algorithm level, the APDT few-sample self-training technology plays a key role. This technology only needs 1-20 pieces of good samples to automatically learn the characteristics of good products and establish an accurate detection model. During detection, the device compares the collected 3D data with the model to quickly and accurately identify defects such as voids. The effectiveness of this technology lies in its ability to overcome the dependence of traditional methods on a large number of samples and adapt to different models of BGA products, improving the flexibility and accuracy of detection.

Compared with traditional rule-based AOI and manual visual inspection, Wemio's 3D AI AOI device has obvious advantages. Rule - based AOI detects based on fixed rules. For complex and changeable BGA void defects, it is difficult to adapt and accurately identify, while Wemio's device can continuously optimize the detection model through AI self-learning. Manual visual inspection is not only inefficient but also easily affected by human factors, and the detection accuracy cannot be guaranteed. Wemio's 3D AI AOI device can achieve a detection rate of over 98%, greatly improving the reliability of detection.

Typical Application Scenarios

• BGA void detection: Wemio's 3D AI AOI device obtains the 3D topographic data inside BGA through a 3D camera and uses point-cloud analysis technology to accurately identify the positions and sizes of voids. The difficulty lies in that the voids may be hidden inside the packaging, which is difficult to find by 2D detection. 3D detection requires precise imaging and algorithms to distinguish between normal structures and voids.
• Hidden solder joint detection: The device uses 2D-3D fusion technology. First, it obtains the approximate positions of solder joints through 2D images, and then accurately judges the quality and connection status of solder joints through 3D reconstruction. The difficulty lies in that hidden solder joints may be blocked by other components, and traditional detection methods are difficult to reach. Multiple technologies need to be combined for detection.
• Coplanarity detection: Through 3D topography reconstruction, the height information of BGA pins or the packaging surface is measured, and the coplanarity is calculated. The difficulty lies in the need for high-precision measurement and data analysis to ensure that the coplanarity is within the allowable error range.
• Micron - level topography detection: Using the high resolution of the self-developed 3D camera, it detects tiny defects and topographic changes on the BGA surface. The difficulty lies in the extremely high requirements for imaging accuracy and algorithms, which need to be able to identify micron-level differences.

Implementation Case

A medium-sized electronics manufacturing company, whose production line mainly produces PCBA boards, involves the detection of a large number of BGA - packaged products. Before introducing Wemio's 3D AI AOI device, the false-alarm rate of BGA void detection was as high as 65%, the miss-detection rate was 4%, the proportion of manual re-judgment working hours was 35%, and the machine shutdown time for model change was about 40 minutes. After the implementation of Wemio's 3D AI AOI device, the false-alarm rate was reduced to 50% of the original, that is, reduced by -50%. The miss-detection rate was reduced to less than 1%. The proportion of manual re-judgment working hours dropped to less than 10%. The machine shutdown time for model change was shortened to less than 5 minutes.

The application of Wemio's 3D AI AOI device brings an efficient and accurate BGA detection solution for electronics manufacturers.

Wemio's Solution and Product

Wemio takes the 3D AI AOI device as the core and uses APDT few-sample self-training technology to quickly build models. In terms of model change, the device can complete the parameter adjustment and model switching of different models of BGA products within 5 minutes. In terms of deployment, it supports various methods such as SDK / API / Docker and can achieve 100% local private deployment to ensure that data does not leave the factory. At the same time, the device can also cooperate with the DaoAI AI AOI software system to achieve semantic false-alarm filtering and further improve the detection accuracy.

In terms of quantitative results, Wemio's 3D AI AOI device reduces the false-alarm rate of BGA void detection by -50% and the miss-detection rate to <1%, greatly improving the detection accuracy and reducing the risk of unqualified products entering the market. The model-changing time is shortened from more than 30 minutes to less than 5 minutes, improving production efficiency and reducing production costs. The proportion of manual re-judgment working hours drops from more than 30% to less than 10%, saving a large amount of labor costs.

This article was generated by AI. Customer cases are simulated scenarios based on real product capabilities and figures are illustrative; see product pages for official benchmarks.