SMT Post-soldering AOI False Alarm Management: Using AI Second Gate to Reduce Re-inspection Volume

In the consumer electronics PCBA production line, traditional AOI detection has many problems, and WeLinkirt's DaoAI AI-AOI software provides an effective solution to these problems.

In the consumer electronics industry, the PCBA (Printed Circuit Board Assembly) production line is a crucial part. This production line is characterized by a fast production rhythm and a wide variety of board types. The post-soldering inspection of surface mount technology highly depends on traditional Automated Optical Inspection (AOI) equipment. Traditional AOI equipment conducts inspections based on rule and template matching. To a certain extent, it can detect soldering defects. However, with the expansion of production scale and the increase of product complexity, its limitations have gradually emerged.

Pain Points: Why It's Difficult

Traditional rule-based AOI has a serious false alarm problem. It generates a large number of false alarms, and real defects are often submerged in a sea of false alarms. From a quantitative perspective, the actual false alarm rate of a leading factory has been at a relatively high level for a long time. For example, in some periods, the false alarm rate even reached over 80%. This means that workers at the post-soldering re-inspection station need to invest a lot of energy in manual secondary confirmation. Taking this factory as an example, workers have to manually re-confirm tens of thousands of misjudged positions every day, with an extremely high work intensity. Moreover, due to the long-term high-intensity work, workers are prone to fatigue, and real defects are more likely to be overlooked during this fatigue, and the miss-detection rate will also increase accordingly.

The root cause of the high false alarm rate of traditional AOI lies in its detection method, which is extremely sensitive to solder joint reflection, component tolerance, and printing offset. In the actual production process, solder joint reflection is a common problem. Different lighting conditions and solder joint surface states can lead to changes in the reflection situation. Component tolerance refers to the deviations in dimensions, electrical performance, etc. of components during the production process. These deviations are easily misjudged as defects in rule and template matching detection. Printing offset refers to the position deviation between the printed pattern on the circuit board and the design pattern. A slight fluctuation will trigger an alarm. The combined effect of these factors makes it difficult for traditional AOI to accurately distinguish between real defects and normal production fluctuations, resulting in a high false alarm rate.

Technical Principle

WeLinkirt's DaoAI AI-AOI software uses advanced technologies to solve the false alarm problem of traditional AOI. In terms of algorithms, it uses APDT (Active Positive Data Training) positive sample/less sample learning technology. The advantage of this technology is that the production line only needs more than ten good product images and a small number of historical false alarm images to launch a model for one workstation, without the need to collect a large number of samples for each type of defect. Compared with traditional machine learning methods, which usually require a large number of positive and negative samples for training, APDT technology greatly reduces the workload of sample collection and annotation, and improves the efficiency of model launch.

In terms of false alarm filtering, it uses semantic-level false alarm filtering technology. This technology can identify reflections and offsets within the tolerance range as good products, and only leave real soldering defects such as bridging, insufficient solder, and misalignment. This is because semantic-level false alarm filtering technology not only focuses on the surface features of images but also understands the physical meaning represented by images, thus accurately distinguishing between normal production fluctuations and real defects. Compared with traditional rule-based false alarm filtering methods, which can only make judgments based on pre-set rules and cannot adapt to complex and changeable production environments, semantic-level false alarm filtering technology has stronger adaptability and accuracy.

Typical Application Scenarios

• Bridging detection: During the soldering process, adjacent solder joints may be connected together to form a bridging defect. Traditional AOI is prone to misjudging normal reflection areas as bridging due to its sensitivity to solder joint reflection. DaoAI AI-AOI software can accurately identify the characteristics of bridging through semantic-level false alarm filtering technology, distinguishing reflections from bridging and improving the accuracy of detection.
• Insufficient solder detection: Insufficient solder means that the amount of solder on the solder joint is insufficient, which will affect the electrical performance and mechanical stability of the solder joint. Traditional AOI's judgment of insufficient solder is often affected by component tolerance and printing offset, and is prone to false alarms. DaoAI AI-AOI software uses APDT positive sample learning technology to train the model with a small number of good product images and historical false alarm images, which can accurately detect insufficient solder defects and reduce false alarms at the same time.
• Misalignment detection: Components may be misaligned during the surface-mount process. Traditional AOI is sensitive to printing offset and is prone to misjudging normal printing offsets as component misalignment. DaoAI AI-AOI software's semantic-level false alarm filtering technology can distinguish offsets within the tolerance range from real component misalignment, and only alarm for real misalignment.
• Cold solder joint detection: Cold solder joints are relatively hidden soldering defects, and traditional AOI is difficult to detect them accurately. DaoAI AI-AOI software can identify the characteristics of cold solder joints through in - depth analysis and learning of images, improving the detection rate of cold solder joints.

Implementation Case

A leading consumer electronics manufacturing enterprise has a large-scale PCBA production line and produces a large number of circuit boards every day. The enterprise has long faced the problems of high false alarm rate and large re-inspection workload of traditional AOI. To solve these problems, the enterprise decided to introduce WeLinkirt's DaoAI AI-AOI software. During the implementation process, the enterprise did not replace the original production line equipment. Instead, it retained the traditional AOI as the first rough screening and connected the DaoAI AI-AOI software as the second gate. Through APDT positive sample learning, each workstation only needed 10-20 good product images and a small number of historical false alarm images to successfully launch the model.

After the implementation, the detection efficiency of the enterprise was significantly improved.

WeLinkirt's Solution and Product

WeLinkirt's solution is to retain the original traditional AOI as the first screening and use the DaoAI AI-AOI software as the second AI re-judgment gate. The DaoAI AI-AOI software has the following characteristics: First, APDT positive sample learning. Each workstation only needs 10-20 good products + a small number of historical false alarm images to be launched, which greatly shortens the time and cost of model launch. Second, semantic false alarm filtering, which can accurately distinguish between'reflection/offset within tolerance' and'real soldering defects', effectively reducing the false alarm rate. Third, 5-minute zero-code model change. New board types can be switched on the same shift without stopping the line, improving the flexibility and production efficiency of the production line.

In terms of quantitative results, after the implementation, the number of positions that need manual re-inspection decreased by about 80%, which means that the workload of workers was greatly reduced and work efficiency was significantly improved. The post-soldering re-inspection station was compressed from three shifts, saving labor costs. The overall false alarm rate decreased by about 80%, making it easier to find real defects. The miss-detection rate was controlled within 1%, ensuring product quality. The model change time was shortened from tens of minutes of stopping the line for parameter adjustment to about 5 minutes. Engineers no longer need to rewrite rules for new board types, improving the response speed and flexibility of the production line.

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.