【2020 Solutions】 AI Assistant - Zhihe Technology Develops the World's Smallest Diamond Sorting Machine
How should one inspect diamonds that are only half the size of human hair? The answer is to use AI for the inspection.
Zhihe Technology has developed the world's smallest diamond sorting machine. By integrating AI and machine learning, along with accumulating large sample data, the system becomes smarter, increasing the yield rate from 70% to 96% within two years. Moreover, by incorporating AI technology into the laser cross-cutting machine capable of handling 500-nanometer laser machining optical measurements, it becomes the first laser machine globally implanted with an AI system.
2016年從中國大陸回流的智合科技,是一家專門提供微測距影像量測系統服務的新創公司,其主要的服務項目包括微米級(µm)甚至奈米(nm)級的微測距應用、半導體檢測、鑽石相關產品、高米密度刀具量測及非標準件檢測。
Zhongxuan Li, General Manager of Zhihe Technology, has over a decade of experience in AOI inspection. After returning to Taiwan from mainland China, he established Zhihe Technology. At that time, semiconductor processes evolved from 8 nanometers to 3 nanometers. Due to high difficulty levels in processing, Li saw a significant market opportunity. Plus, AI's development accelerated after Google released the TensorFlow software in February 2017. TensorFlow is an open-source software library for machine learning applications in various perception and language understanding tasks.
World's Smallest Diamond Sorting Machine - Improved Yield to Over 96%
Consequently, combining his expertise in AOI and AI, Li chose the diamond sorting machine as their first proving ground. Zhihe Technology assisted a major industry company specializing in grinding, cutting tools, optics, wafer refurbishment, and other precision industries with automating human visual inspection process of diamond operations. This company's star product, the 'Diamond Disk,' uses diamonds the size of half a human hair. Previously, the company used traditional visual inspection, employing over 80 workers per production line, many of whom were foreign labor or older employees. The process was time-consuming, output was low, and labor costs were high. Most importantly, their semiconductor clients demanded automation, digitization, and increased precision for diamond inspections.
With Zhihe Technology's help, they supported a semiconductor equipment supplier in fully automating their manual diamond inspection process. They helped create the world's smallest AI-equipped diamond sorting machine, increasing the yield rate from below 70% to over 96%, gathering millions of diamond data points per day.

▲靠AI技術打造全球最小的鑽石篩選機

▲運用AOI及AI交互應用,大幅提升鑽石篩檢效率
此外,智合科技在2019年10月與雷射設備廠商雷科科技共同合作開發雷射十字加工機,簡單說,就是在只有髮絲二分之一的尖點上,打上十字,其精度及準度的要求更高。智合採用AOI方式自動標注,測量位置與角度對位及加工高度,再運用AI訓練位置與角度估算核心,反覆校調後,耗費4個月時間,開發出全球第一台植入AI系統的雷射十字加工機,有了AI技術的加持,讓原有的雷射機價格翻了三倍之多。

▲智合與雷科科技合作,共同打造全球第一台AI 雷射機
Three brilliant methods to make Auto-AI digital transformation so easy
Zhongxuan Li shared that Zhihe Technology is able to quickly integrate AI technology and develop Auto-AI, allowing enterprises to rapidly adopt and smoothly transition into digital transformation. There are three main methods of implementation:
Method 1, Simplifying training issues with an automated labeling platform: Use cameras to collect data from machine manufacturers, replace manual labeling with automated labeling, and progressively train to improve accuracy. The simpler the problem, the less data is needed for training.
Method 2, Parallel advancement of AOI and AI: In smart manufacturing processes, relying solely on either AOI or AI cannot achieve everything. First, AOI should be used to mark features and distinguish between good and defective parts, followed by AI for labeling and training. Using them in tandem enhances their effectiveness, and as the training data accumulates, the proportion of AOI decreases while that of AI gradually increases.
Method 3, Enhancing the integration capabilities of embedded system peripherals: Establishing new computation platforms (embedded systems or IPC platforms) continuously enhances the computing power of AI, thus lowering the industrial threshold for AI applications.
為降低AI使用門檻與成本,智合科技建立自主開發核心-Auto-AI(又稱為傻瓜系統),目前已經跟國內知名工控電腦大廠進行合作,提供使用者更簡易的AI 使用環境。李忠軒表示,台灣是全球最適合作AI系統的國家,擁有超強的電腦設計能力與系統整合能力,若能再加上軟核心平台,將可大幅提升AI落地應用的實證。

▲AOI與AI交互並行,將AI應用落地時程大幅縮減
智合科技有研發能力相當強的機械控制及AI演算法的專業團隊,主要是公司的薪酬制度不同,智合將70%的利潤分享給員工,讓員工共同享受公司的成長果實,因此能吸引優秀人才投入,即時協助解決客戶痛點,在不更新設備的情況下,藉由AI技術的導入,提升原有設備價值。李忠軒也自許,智合將從純粹業務推銷性質的設備商,轉變成為工業升級服務方案商,並將客戶的滿意度與安全感,轉變成為市場行銷上的卓越口碑。

▲智合團隊,圖(左二)為總經理李忠軒
「Translated content is generated by ChatGPT and is for reference only. Translation date:2024-05-19」