"ADC (Automatic Image Defect Detection and Classification) can't run in our factory, and the on-site team of one of the world's largest information technology and business solutions companies is too slow, so you can try it."
In 2018, Chengdu Shuzhilian Technology Co., Ltd. (hereinafter referred to as "Shuzhilian") received a call from an electronics group in Xiamen, asking them to quickly prepare a set of ADC industrial quality inspection project plans to conduct industrial defect detection and solve industrial quality inspection problems. This was the first head-on collision between Digital Link and an international head company, and the result was the successful execution of Digital Link.
Logo of Chengdu Shuzhilian Technology Co., Ltd
Later, it came first: industrial AI quality inspection from the data
Industrial AI quality inspection is the "life gate" related to production yield, but this life gate has long been in the hands of foreign manufacturers.
Fu Yan, founder and chairman of The Company, said that in the image defect detection software market, the market share of foreign manufacturers has reached more than 60%, and in terms of big data analysis software, the four international leading enterprises occupy 80% of the market. Most of the giants have been established for a long time, the investment in research and development is high, and the monopoly is based on preemptively occupying an early position, the company behind the traditional vision detection V software has been established for 40 years, and another giant German M company has also been established for 25 years.
The high price and slow service of overseas enterprises have made it extremely painful for many Chinese companies to use. Secondly, traditional vision inspection is no longer enough to cope with the industry's ever-changing inspection requirements. However, the advantages of software and hardware technology accumulated by foreign companies for a long time are still obvious, and the domestic machine vision market is occupied by international giants with a large market share. With the gradual deepening of domestic brands in the field of automation, the proportion of domestic brand sales in 2019 has reached 48.7%, and it is expected to further enhance the localization rate with lower prices, better service and faster market response in the future, providing opportunities for the number of companies to enter the market.
Seeking domestic alternatives to the industrial testing industry has also become a top priority for major enterprises.
"The first contact with industrial testing was when a company called us and asked us if we could help them analyze the defect data." Fang Yuke, general manager of the smart industry division, recalled: "We completed the system construction while stationing at the site, and then they were quite satisfied, so we continued to do three phases of service. ”
In 2018, Sosnault beat competitors such as one of the world's largest information technology and business solutions companies to provide panel inspection services to BOE and Pegasus. "BOE's panel testing involves PCB boards, IC line detection and other content, coupled with the development of new energy and mobile Internet, we all realize that the rise of pan-semiconductor testing will be the future trend." Fang Yuke recalled.
DigitalLink ADC's service industry
Subsequently, Digital Union began to focus on image processing analysis, and its software product ADC (Image Automatic Defect Detection and Classification System) was introduced to the market. R&D is also starting out in this growth industry, weakening the empiricism that time brings and highlighting the advantages of the digital link. At present, Shuzhilian has served the panel manufacturing, PCB, packaging and testing, PCBA, new energy, automobile manufacturing and other industries, and has been the largest ADC product and solution provider in China in the subdivision field, and the market's high recognition of them is reflected in the average monthly order growth rate of 10%.
From software to hardware: Build the skeleton of machine vision
Although domestic image analysis has sprung up, it is undeniable that overseas companies are still in a strong position. Foreign high-end manufacturing and automation processes are earlier than the mainland, of which the two-way development of software and hardware is indispensable. Taking the machine vision of industrial inspection as an example, the software algorithm is its flesh and blood, which determines the adaptation to the specific scene, while the automation hardware device is its skeleton, including the understanding of the specific industry.
In the view of Song Anxing, the hardware director of Shuzhilian, the opportunity for iterative upgrading of vision inspection equipment equipped with AI algorithms has arrived. Before joining Shuzhilian, he had many years of experience in the development of semiconductor devices, and after joining, he deeply realized the industrial advantages of AI+ADC. "In the past, visual inspection was dominated by traditional algorithms, supplemented by convolutional neural networks; with the leap of deep learning frameworks, the advantages of AI rapid problem solving and rapid replacement of compatible products have gradually become prominent, and the corresponding equipment supplier situation will certainly be rewritten."
According to 36Kr's market research, this is also the best era of AI+ industrial detection. Since 2018, the market for industrial testing has been expanding. In 2020, the global machine vision market size was 12.5 billion US dollars, while the domestic market accounted for 20 billion yuan (about 3.16 billion US dollars), with a compound growth rate of more than 12%. At the same time, domestic industrial testing enterprises are also rising, TZTEK Technology, Matrix Technology, Huaxing Yuanchuang, Jingji Electronics and other listed enterprises, have achieved hundreds of millions of yuan in annual revenue.
Digital Link focuses on AI+ industrial testing in the pan-semiconductor industry
Pan-semiconductor inspection equipment has become a huge historical opportunity. This is the best time for the AI algorithm of the number of links to carry hardware and cut into hardware.
The high technology, high patents and high profits of hardware are built on high research and development and high cost, which is a threshold that all hardware industries have to cross for large-scale research and development and production, and it is also a major reason why the United States' research and development strength leads the world.
According to the European Commission's "2020 EU Industrial R&D Investment Scoreboard" data, among the 2,500 companies in the global R&D list, 775 companies in the United States were shortlisted, with a total R&D expenditure of 347.7 billion euros; followed by Europe, with 542 companies shortlisted, with a total R&D expenditure of 220.9 billion euros; and in China, 536 companies were shortlisted, with a total R&D expenditure of 11.88 billion euros, and the average number of single companies was only 0.49% of that of the United States.
The number of links to develop hardware equipment, can not avoid the road of heavy industry research and development. To this end, Song Anxing pushed the AOI (Automatic Optical Inspection Equipment) of self-built hardware detection in Shuzhilian, and strived to introduce hardware scientific research thinking into this software company.
10 years of experience in hardware equipment development has pushed Song Anxing to constantly explain hardware ideas to colleagues: "When I first pushed AOI, 98% of employees didn't know what AOI was. Just like they don't understand AI algorithms, the technical points of optical electromechanical technology make big data elites scratch their heads, and the material cost of hardware seems too expensive. Not to mention the equipment and processing, procurement, assembly, commissioning and so on, all the supply chains and methods have to start from scratch. The upfront investment and cost are high, but the economies of scale after mass production will bring high value. The AI algorithm of Shuzhilian is second to none in China, and the high-value AOI must be the ultimate track for quality inspection with AI algorithms. ”
He spent a long time explaining to colleagues in the Ministry of Smart Industry what is the difference between doing hardware: "In the architecture of big data architecture and industrial equipment, industrial equipment generally does the bottom layer through C, and C# does the upper layer. As a software company, Digital Link is accustomed to the central-end architecture. But now we need to adapt to the technical specifications of the device industry, in order to quickly be compatible with more hardware modules, and deploy inference on the device side to ensure responsiveness and reproducibility, thereby achieving lower costs. At the same time, the traditional image acquisition algorithm needs to process the picture through splicing, cutting or other means before it can do AI recognition, while the ADC is directly analyzed by the picture detected by the AOI, which is a completely different approach and improves the difficulty of the AOI system. But the synergy between the old and new algorithms must be solved, otherwise the image accuracy before the depth algorithm cannot be guaranteed. ”
After months of discussion, Shuzhilian was ready for Song Anxing to build an optical laboratory as an initial test of water. This office-transformed laboratory became the birthplace of AOI inspection hardware.
Complete the hardware and open the door to the standard
Based on the good reputation of Shuzhilian in the ADC software development period, they received many contact calls in the process of developing AOI hardware, and the blue ocean characteristics of the pan-semiconductor industry were also revealed. After several experiments and communications, Shuzhilian first conducted DEMO prototype testing with a foreign company, and then received an invitation from a new energy battery company in Qingdao to customize a software and hardware system for the detection of new energy vehicle batteries in the factory.
The rapid development of the new energy automobile industry has led to the rapid development of this battery company eager to find testing solutions, requiring the testing equipment provided to quickly detect battery defects on the high-dynamic production line, the accuracy needs to reach less than 0.05 mm, the detection rate is milliseconds, and they have not been able to find ready-made service providers in the market. In the end, It took three months to complete the design, prototype manufacturing and assembly, algorithm debugging and parameter adjustment, and modeling of this customized solution, and the product has been launched.
Digital Link's inspection applications in the automotive industry
The successful application of this hardware has opened a trickle of AOI standard rivers. Subsequently, Shuzhilian supported and transformed products on multiple projects, while accumulating experience, optimizing hardware, and building a standard product framework. Among the dozens of demand solutions, Digital Link has gradually completed the accumulation of new energy batteries, auto parts, circuit boards and so on.
In August 2021, Shuzhilian delivered the first fully self-developed productized AOI, which can solve the defect problem after PCBA assembly, realize all the work from development, design, processing, assembly to commissioning, and meet customer requirements at one time. On January 26, 2022, Shuzhilian focused on the field of PCBA (electronic printed circuit board) and officially launched an innovative hardware product based on deep learning - light-chasing AI-AOI, which solves the problems of traditional AOI inspection, high false detection, and slow product change, and correspondingly saves the manpower of enterprises.
The measured data of the digital link also shows their all-round improvement of industrial testing.
Compared with other AOI solutions, the Digital Link Light Chasing AI-AOI is equipped with the Digital Link deep learning neural network model, which is a model that collects multiple successful landing experiences and massive defect data in the pan-semiconductor industry, and trains a compatible solder joint morphological change and high generalization model, which can effectively solve the core pain points of traditional AOI in programming/debugging time is too long, misjudgment is too high, and different operation results from person to person.
Tracking AI-AOI's motion speed and detection size
While considering low cost, Sosolink Light Chasing AI-AOI offers the industry's fastest motion speed and largest detection size, and is compatible with production lines of different heights. Compared with ordinary AOI, the AI-AOI automatically identifies the position of the solder joint through the AI model, and a detection box can cover the detection of various types of defects, shortening the product changeover time of the previous 1-2 hours to 10 minutes.
Light chasing AI-AOI: Automatic identification of solder joint position
In addition, the solder joint recognition based on deep learning can accurately locate the defect location, make accurate judgments on a variety of solder joints with different forms, and solve the problem of OK/NG coarse classification of defects by traditional algorithms.
Light Chasing AI-AOI: Detect defect types
After market and customer verification, the detection rate of ai-AOI can reach 99.99%, and the false positive rate is less than 0.3%. Among them, AOI after the PCBA-DIP furnace has achieved industry leadership in core indicators, effectively improving the quality of factory processes.
Conclusion: Autonomy and specialization are still the core of industry
At present, the main players of machine vision are still giant enterprises such as the United States, Japan and Germany, but in recent years, the mainland has paid more and more attention to the high-end equipment manufacturing industry. The overall industry needs to get rid of import dependence and develop high-end manufacturing, machine vision technology is indispensable, and its strict requirements for accuracy must also be identified from the level of machine intelligence technology.
Fang Yuke and Song Anxing, as the main leaders of the digital alliance to enter the AOI, believe that the "light chasing AI-AOI" has reached the first step of the digital connection industrial hardware. "But there is still a long way to go." Song Anxing said, "The industrial testing hardware industry is also divided into upstream, middle and lower reaches, and the more upstream the higher the sophistication. Now we have been polishing for 2 years to create a light-chasing AI-AOI, which can meet the downstream PCBA detection, and we will maintain research and development in the future and quickly expand to PCB, packaging and testing and other industrial fields. ”
As for whether this is an important step for Digital Link to "become the largest industrial data enterprise in China in the future", Fang Yuke laughed and said, "I hope this is a small step for Digital Link to contribute to China's manufacturing in 2025." ”