AI quality inspection is getting through the “last mile” of product landing.
On the production lines of some manufacturing factories, the use of AI vision to detect defects has begun to replace manpower, changing the traditional way of detecting defects only through human eyes and experience.
AI vision can realize 24-hour uninterrupted detection through equipment, without visual fatigue of human eyes, so there is a qualitative improvement in detection efficiency and quality.
According to IDC data, by 2025, the market size of domestic industrial AI quality inspection will reach 6.2 billion yuan. But at the same time, there are currently more than 2 million relevant personnel engaged in the quality and efficiency of production lines in my country, and the annual labor cost is as high as 140 billion yuan.
In other words, the industrial AI quality inspection market has a market space of up to 100 billion yuan, and this part happens to be the quality inspection needs of “neglected” small and medium-sized enterprises.
“Small and medium-sized enterprises are the capillaries of the economy. Their intelligent transformation is very important, but there is an imbalance between supply and demand.”
Zhang Zhiqi, CEO of Weiyi Intelligent Manufacturing & Zhiyun Tiangong, said that many manufacturing companies, especially small and medium-sized enterprises, are limited by factors such as their own scale and capital, coupled with the high cost and poor flexibility of AI quality inspection before, their quality The inspection demand has not been met for a long time.
Therefore, it is necessary to change the original implementation path of industrial AI quality inspection, and solve the problem of difficult implementation and high cost of AI quality inspection for small and medium-sized enterprises from both sides of technology and products, so as to further open up the quality inspection market and even promote the intelligent transformation of small and medium-sized enterprises .
AI quality inspection is difficult to implement in small and medium-sized enterprises: poor flexibility and high cost
As the last pass of quality control, the importance of quality inspection is self-evident: if the product is qualified, it can be shipped, and if it fails, it will be recast. The quality of quality inspection is related to the yield rate of the product and the brand image.
In the past, product quality inspection mainly relied on manpower, requiring quality inspectors to have keen eyesight and rich experience to detect the types of defects.
The disadvantage of this model is that the labor cost of an experienced quality inspector is not low, and the experience of the quality inspector is difficult to replicate. In addition, people have physical fatigue, and working for a long time will cause great wear and tear on the human body. For small and medium-sized enterprises, raising an experienced quality inspection team is tantamount to greatly increasing labor costs.
Therefore, many small and medium-sized enterprises directly choose to reduce the quality inspection team and choose the business strategy of “only replace but not repair” on the business side. The labor cost saved can even cover the cost of directly replacing new products. However, although this can effectively save costs in the short term, it reduces the product yield; in the long run, it will cause too much damage to the brand, and it is an act of drinking poison to quench thirst.
“In fact, quality inspection can be digitized.” Zhang Zhiqi believes that quality inspection is one of the best scenarios for the application of artificial intelligence to the industry.
Artificial intelligence can carry out a large amount of training through data, accumulate the experience of the previous “quality inspection master” into a model, and then let the machine equipment do a large number of defect detection. On the one hand, this directly saves labor costs; on the other hand, the machinery and equipment can work 24 hours a day, thereby improving detection efficiency.
But the traditional AI quality inspection solution, in Zhang Zhiqi’s view, has two major problems.
-
The cost of introducing AI in the early stage is too high, and the cost of subsequent operation and maintenance is not low
“Traditional AI quality inspection solutions require a large amount of data for model training, and also require a large number of algorithm engineers to go to the production line to communicate with customers, accumulate experience, and then tune the algorithm model to meet customer needs.”
For a similar quality inspection plan, the cost of the early production line transformation alone will cost hundreds of thousands of yuan, and the follow-up operation and maintenance will also require a lot of manpower and material resources. Its main customers can only be major manufacturers such as Huawei, BYD, Foxconn, etc. For more small and medium-sized enterprises, it is basically unbearable.
-
Asset-heavy mode deployment, poor flexibility
Zhang Zhiqi said that small and medium-sized enterprises have strong production flexibility, weak cash flow, and obvious characteristics such as multiple varieties, small batches, and seasonality; and it takes several months to deploy a set of solutions from demand to successful deployment. In the past few months, the market demand may have changed. When the production line transformation is completed, the factories may be close to closing down.
In addition, for small and medium-sized enterprises, a set of quality inspection solutions may only be used a few times, and the frequency of replacement is very fast. It is not cost-effective to deploy a traditional AI quality inspection solution.
“Traditional AI quality inspection deployment process, taking product surface appearance defects as an example: the average equipment introduction cycle is 6 months, the single product model training cycle is 2 months, and each project requires an average of 5 engineers to be on-site for implementation. 400,000 to 1,000,000 yuan; not to mention the frequent line change needs of some flexible manufacturing industries.”
Zhang Zhiqi believes that due to the cost and deployment flexibility of traditional AI quality inspection solutions, a large number of small and medium-sized enterprises in China cannot afford or use them, but the quality inspection needs of small and medium-sized enterprises are very strong. On the one hand, there are high landing costs, and on the other hand, the quality inspection needs of a large number of small and medium-sized enterprises cannot be met. This contradiction between supply and demand seriously hinders the intelligent transformation of small and medium-sized enterprises.
Therefore, in Zhang Zhiqi’s view, it is necessary to change the path of AI quality inspection to solve the problems of high cost and poor flexibility.
New solution: AI quality inspection returns to the original, light deployment, low cost
The implementation of the previous AI quality inspection solution required a large number of algorithm engineers to be stationed on the front line, build algorithm models, train and optimize, and then apply them to the production line. Basically, it can only be customized and developed, and reuse cannot be achieved, and there are slow line changes and maintenance costs. advanced questions.
This not only brings high customization costs, but for “small batch” products produced by small and medium-sized enterprises, the sample data may not be enough for the development of quality inspection models.
In addition, in terms of the training effect of the model, since the data collected by the front-end equipment can only be used after being “translated” by engineers, it cannot completely reflect the real data, which will not only cause partial data distortion, but also affect model optimization, The effect of quality inspection will also cause data waste.
“We used to do AI quality inspection projects more to meet the project space requirements, relying on programmers and algorithm experts to translate, and transform experience into productivity through models.”
This method adds intermediate links, resulting in increased labor costs and loss of model effects. If the experience of front-line workers can be directly transferred to the model and the model can be trained independently, various costs in the intermediate links can be reduced.
Zhang Zhiqi believes that the “base” of artificial intelligence is human experience and data, so it is necessary to “allow on-site production personnel to deeply participate in the production process of AI models, so that their ‘human experience’ can be precipitated”; while the industrial AI This is the production experience of the people on site and the massive production data. It is necessary to “center on the production site and immerse yourself in the manufacturing scene around the collection, calibration, transmission and iteration of the data flow on the production site.”
This is equivalent to reconstructing the entire business process of industrial AI “from data collection, to model training, to deployment”, and changing the original path of AI quality inspection.
for example:
For the quality inspection of some digital products, after the front-end collects the data, the equipment will directly match the algorithm library and compare the marks for common defect types; if there is no corresponding defect in the algorithm library, the system will actively mark it, and then the front-line workers will mark it , when the system detects similar defects again, it can directly mark the defects.
In the whole process, there is no professional deployment link, and the relevant algorithm models are directly deposited in the equipment. The process of importing the model into the production site is fast and light, and the experience can be replicated; the system will also optimize and learn based on the real-time data of on-site production. At the same time, front-line workers are freed from the laborious quality inspection work, and can also reduce their workload by doing work similar to that of annotators.
This light deployment, low cost, and flexible model is more suitable for a large number of small and medium-sized enterprises that are small in scale but have quality inspection needs.
AI quality inspection down: from large factories to small and medium-sized enterprises
AI quality inspection is changing from customization to lightweight deployment, and from the exclusive use of large manufacturers to become more “people-friendly”
Zhang Zhiqi believes that this change reflects the trend of intelligent transformation of domestic manufacturing enterprises: expanding from a small number of high-end manufacturing industries to broader low-end manufacturing industries.
As the former Vice President of SAP China, Zhang Zhiqi was one of the first people who came into contact with German Industrialization 4.0 and tried to introduce German Industrialization 4.0 into China.
During this process, Zhang Zhiqi found that due to the high degree of standardization of Industrialization 4.0, only a few high-end manufacturing and large manufacturers, such as Foxconn and BYD, are willing to invest. And after the investment, the business results cannot be seen in the short term, and the ROI cycle is too long, which is completely impossible for small and medium-sized enterprises.
In October 2017, Zhang Zhiqi joined Baidu as the deputy general manager of Baidu Cloud, responsible for the commercialization of Baidu’s AI. During this period, Baidu began to look for the application scenarios of artificial intelligence in the industry, and finally took AI quality inspection as the starting point, and launched the “Baidu Tiangong Internet of Things Platform”.
In July 2020, Zhang Zhiqi resigned from Baidu and joined DingTalk, responsible for the commercialization of DingTalk. Soon, Zhang Zhiqi chose to start his own business, and established Zhiyun Tiangong, a company engaged in digital and intelligent transformation of enterprises.
At the end of 2022, Zhiyun Tiangong and Weiyi Zhizao will complete the merger, and Zhang Zhiqi will serve as CEO.
Founded in 2018, Weiyi Zhicao takes industrial AI quality inspection as the core, and uses AI + machine vision to do intelligent transformation of factories.
Zhang Zhiqi believes that the intelligent transformation of industrial manufacturing only focuses on large enterprises and high-end manufacturing, while ignoring small and medium-sized enterprises, which is not correct. Small and medium-sized enterprises play a fundamental role in the economic structure. Only when the intelligent transformation of small and medium-sized enterprises succeeds can the entire industry be upgraded.
From a business point of view, there is a great demand for intelligent transformation of small and medium-sized enterprises. Even if the unit price is low, the market size is also very large. Among them, AI quality inspection is the “last mile” for products to go to the market. If AI quality inspection can be made “affordable and usable” by more small and medium-sized enterprises, it will also promote the intelligentization of the manufacturing industry to a certain extent. transformation.
Therefore, Zhiyun Tiangong and Weiyi Zhizao chose to merge, and launched two series of products, “Gong” and “Spirit”.
Specifically, the “Gong” series includes Gong Xiaojiang, Gong Xiaozhi and Gong Xiaohui.
Gongxiaojiang is defined as an “AI digital quality inspector”, which aims to replace traditional human quality inspection and can achieve flexible deployment; Gongxiaozhi is an “AI production line administrator”, aiming to realize visual management of the production process; Xiaohui is an “AIoT digital factory administrator”, which is used for intelligent management of the entire factory.
Taking the small worker as an example, it has the ability to quickly cut lines, and is compatible with different types of light sources, lenses, and robotic arms that have been deployed by the enterprise to meet different inspection needs. In terms of software functions, it also has one-key switching between the detection models of different products.
“Ling” series includes Lingmo OCT defect detector, Lingjing PMD high-reflective surface detector, Lingzhen OMX compound eye array module, suitable for 3C, semiconductor, precision optics and other industries, as well as highly reflective materials such as optical lenses and wafers .
Zhang Zhiqi said that the two series are modular in design and can be plug-and-play; the hardware products can be compatible with the existing information systems of small and medium-sized enterprises, and SaaS software such as Gongxiaozhi and Gongxiaohui can also be connected to the existing information systems of enterprises. Some hardware equipment, and achieve horizontal connection with the information system.
“The Gong series is a solution combining software and hardware, and the Ling series is a hardware product for defect detection. The two cooperate with each other, and both have flexible deployment capabilities. The cost is far lower than the customized AI quality inspection solution, which is suitable for small and medium-sized enterprises.” Multi-category, small batch’ quality inspection requirements.”
Zhang Zhiqi believes that the flexible deployment of AI quality inspection is an irreversible trend. In the past, only a few large companies were willing to try the transformation of the manufacturing industry. Small and medium-sized enterprises had needs but no funds. The transformation of the manufacturing industry was more like a direction and a slogan.
However, with the implementation of artificial intelligence, breakthroughs in various technologies, and product iterations, the cost of AI quality inspection has been further reduced, and more and more small and medium-sized enterprises can afford it and use it.
“The intelligent transformation of the manufacturing industry is gradually becoming a reality.” Leifeng.com Leifeng.com
This article is transferred from: https://www.leiphone.com/category/mandigital/UnkJScrjffHLK7I0.html
This site is only for collection, and the copyright belongs to the original author.