【2020 Solutions】 ASUS AI Deep Learning Image Recognition Makes Defect Detection Easier
For the manufacturing industry, replacing manual visual inspection with automated optical inspection is common, especially when the yield of 3C or semiconductor products is high. General automated optical inspections often face the bottlenecks of insufficient defect samples and difficulties in qualitative and quantitative recognition. Using AI deep learning for image defect detection has become increasingly significant!
AI detects minute defects, ASUS makes smart manufacturing 'visible'
'Initially, we hoped to promote upgrading with our 3C supply chain partners steadily, assisting the industry to enhance and face international competition,' said Chang Quande, ASUS Global Vice President and Co-General Manager of the Smart IoT Business Group. ASUS Smart Solutions Business Unit uses AI deep learning to perform various workpiece defect detections, and layout after accumulating experiences is a priority task.

▲華碩全球副總裁暨智慧物聯網事業群共同總經理張權德
For metal component manufacturers, detecting defects on surfaces is relatively difficult due to the reflection of light, which often causes actual defects to be overlooked. Mastery of optical properties and the specifics of component surfaces is crucial. The ASUS Smart Solutions Business Unit not only has AI experts but also a digital imaging technology team with unique post-processing skills and strong augmentation capabilities. They can achieve correct defect data collection and train AI models efficiently even with a very small number of defect samples. 'General optical inspection accuracy is about 85-90%, and high precision-seeking manufacturers would not use it, as it implies a +/-10% defect misjudgment,' said Chang Quande. Whereas manual visual inspection has an accuracy rate of about 93%, it is labor-intensive and carries occupational hazard risks. ASUS has now enabled AI to achieve 98% accuracy, fully capable of replacing manual inspections and certain traditional optical inspections.
Previously, it took three people to manage quality control across three production lines, now only one is needed
In recent years, many manufacturing industries have been returning to invest in Taiwan. Major metal structure stamping plants have also committed to establishing new factories. ASUS has designed their three-in-one defect detection stations, capturing images through edge computing, uniformly training an AI model, and utilizing the same AI inference workstation to perform defect detection calculations. Quality control stations across various production lines now monitor these processes in real-time. Previously, three production lines required three people for quality control; now, only one is sufficient, increasing the detection rate from 93% to 98% and reducing costs by 5%. Accompanied by the reallocation of human resources, the stamping plant has achieved smart manufacturing and has broken the curse of increased production costs due to returning investments.

▲ASUS IoT 應用產業
除了金屬機構件之外,塑膠成型件、印刷電路板等電腦周邊元件生產業及系統組裝業都能運用AI 深度學習影像瑕疵檢測做高精度品管,目前也有半導體業正在優化導入華碩AI 深度學習影像瑕疵檢測,以補足自動光學檢測在晶圓層所抓不到的瑕疵,盼藉由AI的助力突破良率瓶頸,降低人工目測或自動光學檢測已知的誤判所造成的損失,更能利用人工智慧大數據針對品質瑕疵種類做統計分類以歸納出瑕疵形成原因,從源頭改善進而減少製程瑕疵。
「Translated content is generated by ChatGPT and is for reference only. Translation date:2024-05-19」