Major electronics manufacturers already expect their equipment suppliers to provide well-developed AI tools. On the other hand, some are cautious about how to adapt to working with AI while maximizing its potential to accelerate workflows and eliminate human error. From either perspective, using AI as an inspector in SMT assembly can offer several well-known advantages, including its unerring attention to detail, consistent judgment, and faster decision-making.
A Lifetime Inspector
: Maintaining qualified inspectors to review boards rejected due to AOI "No-Good" (NG) judgments is a well-known challenge for electronics manufacturing companies. The work is repetitive yet stressful, demanding sustained high performance while carrying significant responsibility. A single defective unit leaving the factory puts the employer's reputation at risk.
The problem for production managers is that the typical human involvement cycle follows a familiar sequence. An inspector joins the team and begins acquiring knowledge and experience by analyzing a large number of AOI images labeled OK or NG. As experience grows, the familiarity and repetitive nature of the work can lead to fatigue. A common pattern is for inspectors to reach a high level of proficiency and then begin looking for a different position within the company, or even leave altogether. The value of their expertise, both within the team and the company, is then lost.
Managers can break this cycle by training AI to become an ever-present, ever-attentive, and permanently engaged expert. If properly implemented, trained AI can effectively store the experience gained by inspection experts who have worked on the team and become a consultant, ready to support new inspectors joining in the future.
Training for the Secondary Judgment Process:
One way to achieve this is to integrate AI into the workstations where inspectors review rejected assemblies following automated optical inspection (AOI). Units incorrectly identified as defective can be returned to production for final assembly, while those with verified defects can be sent for remanufacturing.
When adopted in the secondary judgment process, AI initially joins the inspection team as a new member and must learn on the job from the company's experienced inspectors. When rejected assemblies arrive, human inspectors review the AOI data and images to apply their own judgment. If their decision indicates that the AOI has flagged a false NG (Near Missing), the board can be sent to complete the remaining production processes. As the AOI images and human inspector decisions are saved, the AI server uses this accumulated data to train inspection models. As training continues, using the data acquired daily, the accuracy of the models improves, and the AI can begin to support human judgment during real-time inspections.
Training the models in this way allows them to quickly reach a level of confidence sufficient to assist human inspectors in their decision-making. When inspecting a defective point, the "AI Assessment Station" can present the AI's opinion with an associated confidence score, as shown in Figure 1. This level of support can be especially valuable for less experienced inspectors, as it allows them to rely on the AI for assistance in challenging cases, helping them refine their expertise while simultaneously improving both accuracy and performance.
Figure 1. First level of AI assistance in the secondary judgment process.
As each model reaches a stable confidence level, indicating near-perfect accuracy, the AI is ready to provide automated judgment for specific component types and defects. At this stage, the AI's superior judgment is combined with the human inspector's skills to ensure a faster process while maintaining high accuracy and reducing operator fatigue.
Where to begin?
Software applications with features like these are now entering the market and are extremely effective at detecting welding defects, such as insufficient welding or inadequate wetting, flagged by 2D AOI inspections. Yamaha Robotics created its AI Judgement Station environment to enable customers to independently create and train AI models using their own production data. After selecting suitable samples for training, users can leverage the software's features to automatically generate AI models and then train them within a machine learning framework using data collected from daily inspection shifts. Users can check the accuracy as training progresses to identify when the model has stabilized. After analyzing enough good images and 30 or more defective images, the model can reach a level of accuracy suitable for assisting human inspectors in their visual analysis. This can be achieved in a short time, although a larger number of images provides further improvement. After processing more than 1,000 images, the model's quality can become extremely high. AI assessments take approximately one second per inspection.
AI and OCR:
One area where AI can already provide reliable, human-like judgment and effectively alleviate the inspector's workload is optical character recognition (OCR). Conventional rule-based machine vision can misread characters with low contrast, obscured by scratches or other damage, or covered by foreign material. These are common risks in a surface-mount environment and can affect important markings such as component values, date codes, polarity markings, and board identification (Figure 2).
Figure 2. AI OCR can read markings that would otherwise require visual confirmation by an inspector.
By making inferences based on prior learning, just as humans can determine characters when readability is poor, AI can overcome these problems. The Yamaha AI Judgement Station package includes pre-trained AI OCR models that can be used out of the box, without any adjustments or additional training.
Automated Online Assessment:
As AI becomes integrated into company practices, working alongside various human inspectors, the number of models created to cover different components and defect types continues to increase, and the training achieves a high level of confidence. Thanks to this progression, fewer and fewer automated inspections require confirmation from a visual inspector. As reliance on human expertise decreases, the secondary assessment process can be fully automated and eventually integrated online, as illustrated in Figure 3.
Figure 3. Second fully automated evaluation after training of all AI models.
By working with human visual inspectors, providing guidance, and delegating decision-making when necessary, this aligns with both proactive and cautious approaches to introducing AI into inspection. More cautious adopters can maintain human oversight to prevent missed defects and over-inspection in the next process, while also reducing inspectors' workload. On the other hand, packages like Yamaha's AI Judgement Station present a roadmap toward full automation to meet the most demanding customer requirements.
Conclusion:
As industries worldwide explore how to leverage AI in manufacturing processes, surface mount assemblers are applying the technology to boost productivity in optical inspection. Beyond helping to create model libraries and generate programs for online AOI systems, AI is also improving inspection by identifying erroneously rejected boards.
A three-tiered approach, using AI to assist, partially automate, and then fully automate the secondary evaluation process, effectively trains a permanent resident inspection expert to manage this repetitive and demanding task with consistently high speed and accuracy.
