In the MLCC production process, micro-crack detection is a key link to ensure product quality. However, traditional detection methods face many challenges. DaoAI's APDT positive-sample learning technology provides an effective solution to this problem.
Industry background and user scenarios: Multilayer ceramic capacitors (MLCC) are widely used basic components in electronic devices. They are used in large quantities and each capacitor is extremely small in size. In many fields such as smartphones, tablets, and automotive electronics, MLCC plays a crucial role. On the MLCC production line, in order to ensure product quality, each capacitor needs to be inspected for micro-cracks. Micro - cracks may appear on the end face and the body of the capacitor. These cracks may only be micron-level in width, but they can have a serious impact on the performance of the capacitor. Therefore, accurately detecting these micro-cracks is crucial for ensuring the reliability and stability of the product.
Pain points: Why is it difficult?
From a quantitative perspective, the yield rate of MLCC is already very high, and the probability of defects such as micro-cracks is extremely low. It may be that only one product with a micro-crack appears after producing thousands or even tens of thousands of capacitors. This makes it difficult to accumulate enough defective samples to train a conventional classification model. Traditional visual inspection methods require a large number of defective samples for training to ensure that the model can accurately identify micro-cracks. However, due to the rarity of micro-cracks, the inability to collect enough samples has become a major problem.
Manual visual inspection also has serious limitations. In terms of production capacity, the MLCC production line is extremely fast, with a production capacity of thousands of capacitors per minute. Manual visual inspection simply cannot keep up with such a speed. Moreover, the consistency of manual visual inspection is poor. Different inspectors may have different judgments on micro-cracks due to factors such as fatigue and eyesight. This makes manual visual inspection unfeasible on high-capacity MLCC production lines. In addition, the size of micro-cracks is extremely small, only micron-level, which is extremely difficult for manual visual inspection to accurately identify these subtle cracks.
Technical principle
DaoAI's APDT positive-sample learning technology only learns from 1-20 good-product images. The principle is to establish a criterion of “normal” through the analysis of these good-product images. In this process, the system will learn the texture, features, etc. of good products and determine a normal distribution range. When the image of the detected capacitor deviates from this normal distribution, it will be marked as suspicious. This method bypasses the dilemma of traditional methods that require a large number of defective samples, because it only focuses on what is “good” rather than collecting a large number of “bad” samples.
Compared with traditional methods, traditional visual inspection methods need to spend a lot of time and energy to collect and label defective samples. And in the case of insufficient samples, the accuracy of the model will be greatly affected. DaoAI's APDT positive-sample learning technology can establish an effective detection model with only a small number of good-product samples. At the same time, combined with DaoAI's high-resolution imaging technology, the system can achieve micron-level resolution. High - resolution imaging can clearly capture the subtle texture on the surface of the capacitor, enabling the system to identify hair-like fine cracks on the end face. In terms of detection speed, single-piece detection can be completed in milliseconds, which matches the rhythm of high-speed incoming material inspection and pre-taping inspection, greatly improving the detection efficiency.
Typical application scenarios
- High - speed incoming material inspection: Inspection is carried out when MLCC raw materials enter the production line. Due to the high speed of incoming materials, it is difficult for traditional methods to complete the inspection in a short time. With its millisecond-level single-piece detection speed and micron-level resolution, the DaoAI system can quickly and accurately detect micro-cracks, ensuring the quality of raw materials entering the production line. The difficulty lies in ensuring the accuracy of detection under high-speed operation.
- Pre - taping inspection: Inspection is carried out before MLCC is taped and packaged. At this time, a comprehensive inspection of each capacitor is required to ensure the quality of the final product. The DaoAI system can identify micro-cracks on the end face and body of the capacitor, even the subtle hair-like fine cracks can be detected. The difficulty lies in the extremely small size of micro-cracks, which requires high-resolution imaging and precise algorithms for identification.
- Inspection of new-specification capacitors when they are launched: When new-specification capacitors are launched, traditional methods need to collect defective samples again and rewrite the inspection process, which will consume a lot of time and resources. The DaoAI system can achieve 0-code and 5-minute model change without collecting samples again, greatly shortening the introduction cycle of new specifications. The difficulty lies in how to quickly adapt to the characteristics of new-specification capacitors.
- Spot - check during the production process: Regular spot-checks are carried out during the MLCC production process to monitor product quality. The DaoAI system can quickly detect the spot-check samples and timely discover potential micro-crack problems. The difficulty lies in how to ensure the representativeness of spot-checks and the timeliness of detection.
Implementation case
There is a medium-sized MLCC production enterprise with a production capacity of 3000 capacitors per minute. Before adopting the DaoAI system, the enterprise relied on a combination of traditional visual inspection and manual visual inspection for micro-crack detection. Due to insufficient samples, the detection rate of micro-cracks by traditional visual inspection was only 80%, and the false-alarm rate reached 15%. Affected by speed and consistency, the missed-detection rate of manual visual inspection was relatively high, about 10%. Moreover, when new-specification capacitors were launched, the model-change cycle was as long as 2 days.
After the DaoAI system was launched, the detection rate of micro-cracks increased to 98%, the false-alarm rate decreased to 2%, and the missed-detection rate decreased to 2%. The single-piece detection time was stably within milliseconds, fully meeting the rhythm of the high-speed production line. When new-specification capacitors were launched, 0-code and 5-minute model change could be completed, greatly shortening the introduction cycle of new specifications and relieving the manpower and consistency pressure of the manual-inspection station.
WeLinkirt's solution and product
The DaoAI system provided by WeLinkirt is centered around the APDT positive-sample learning technology. The system includes high-resolution imaging hardware and intelligent algorithm software. The high-resolution imaging hardware can clearly capture the subtle texture on the surface of the capacitor, providing accurate image data for subsequent detection. The intelligent algorithm software establishes an accurate “normal” criterion through the learning of good-product images, realizing efficient detection of micro-cracks. In addition, the system also has the 0-code model-change function, which can quickly adapt to the detection needs of new-specification capacitors.
Quantitative results: The DaoAI system has achieved significant quantitative results in MLCC micro-crack detection. In terms of the detection rate, it has increased from 80% of traditional methods to 98%, greatly improving product quality. The false-alarm rate has decreased from 15% to 2%, reducing unnecessary misjudgments and improving production efficiency. The missed-detection rate has decreased from 10% to 2%, ensuring that more products with micro-cracks can be detected. The model-change time for new specifications has been shortened from 2 days to 5 minutes, significantly shortening the introduction cycle of new specifications and improving the enterprise's market response speed.
FAQ
What is the method for DaoAI to solve the sample dilemma in MLCC micro-crack detection?
DaoAI uses APDT positive-sample learning and only learns from 1-20 good-product images. By analyzing these good-product images, it establishes a “normal” criterion. Anything deviating from the normal distribution is regarded as suspicious, bypassing the dilemma of difficult collection of defective samples and realizing the detection of micro-cracks.
What are the advantages of DaoAI in detecting MLCC micro-cracks?
The DaoAI system has micron-level resolution and can identify hair-like fine cracks, enabling accurate detection of micro-cracks. Single - piece detection is completed in milliseconds, meeting the rhythm of high-speed production lines. For new specifications, 0-code and 5-minute model change can be achieved, shortening the introduction cycle and relieving the manpower and consistency pressure.
What problems exist in traditional visual inspection of MLCC micro-cracks?
Traditional visual inspection requires a large number of defective samples for training. However, MLCC micro-cracks are rare, and it is difficult to collect enough samples, which affects the accuracy of detection. Manual visual inspection is limited by speed and consistency. It is difficult to ensure the detection effect under high-capacity production, and it is also difficult to identify micron-level micro-cracks.
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.