Entering 2022, intelligent manufacturing has become a hot word in the government’s “two sessions”.
On the one hand, eight departments including the National Development and Reform Commission, the Ministry of Industry, and the Ministry of Science and Technology issued a notice on the issuance of the “14th Five-Year Plan for the Development of Intelligent Manufacturing”, proposing a national top-level design; on the other hand, various local governments, such as Beijing, Shanghai and Guangzhou First-tier cities such as Shenzhen have issued corresponding policy rules to provide institutional guarantees for the promotion of intelligent manufacturing.
In the process of realizing intelligent manufacturing and promoting industrial upgrading, machine vision, as an indispensable technology, assumes the role of a bridge that allows machinery to “understand, see carefully, grasp firmly, and deliver quickly”.
In other words, in order for a machine to replace manpower, the machine must first be equipped with “eyes” so that it can “see”, and then it can work like a human. This is machine vision.
As a branch of artificial intelligence, machine vision has advantages that humans cannot have: high precision, applicable to dangerous working environments, high recognition efficiency, and uninterrupted work. In fact, machine vision is not a new thing and has been widely used in industrial processes such as appearance inspection and recognition, and goods sorting.
The current machine vision is still dominated by 2D, that is, taking pictures of the object plane through the camera, and then identifying the object through image analysis or comparison. The limitation is that only the features of the object plane can be observed, and the imaging accuracy is easily affected by lighting conditions. Therefore, it is suitable for some low-end manufacturing industries with low technical requirements.
In some high-end manufacturing fields, such as biotechnology, precision semiconductors and other industries that require extremely high measurement accuracy, traditional 2D solutions can no longer meet the needs, and 3D vision is gradually emerging as a new favorite in the market.
From 2D to 3D: the difference in more than one dimension
3D vision, that is, collecting the three-dimensional coordinate information of the object through a 3D camera, and realizing three-dimensional stereo imaging through an algorithm.
Compared with 2D vision system, the advantage of 3D vision is that one more dimension of information data (mainly spatial coordinates) can meet the needs of information measurement such as volume, shape, and distance. Moreover, 3D vision is not easily affected by lighting conditions, and its imaging accuracy is much higher than that of 2D vision. At the same time, its ability to quickly process information is not comparable to 2D vision systems.
To give a simple example: When measuring objects with curved surfaces and radians, 2D vision can only take a plan view, which is difficult to reflect the real situation of the object; , depth and other real information, which is more valuable for machines or humans.
With the intelligent upgrading of the manufacturing industry, the market demand for 3D vision is also increasing. According to the report of Grand View Research, a US market research institution, by 2027, the global 3D machine vision market size is expected to reach 3.46 billion US dollars. During the forecast period, the compound annual growth rate of the market is expected to be 14.7%, which is a potential blue ocean market.
Nuggets learned that the current application of 3D vision in intelligent manufacturing has developed from a single scene to the empowerment of the entire production line, involving multiple links such as positioning, guidance, production, sorting, and assembly.
Take the production process of smartphones as an example: in the era of 2D vision, the most widely used application scenario is quality inspection, that is, size and defect detection, involving three major parts: motherboard, components and packaging. 3D vision can directly cover these processes, and is superior in detection accuracy and speed, and expand the application to scenarios such as feeding, production, inspection, and packaging, and realize intelligent transformation of the original production line. Picking, handling and other links need to be planned and completed in real time according to different product types.
This actually provides convenience for manufacturers’ flexible production. Driven by the C2M business model, enterprises need to decide the production scale according to the real-time orders of users. In the past, mechanized production belonged to mass production, and the flexibility was very weak. 3D vision has improved the intelligence level of industrial robots and automation equipment, making it capable of meeting actual needs. Production needs to flexibly change the production of various products.
For example, during the Winter Olympics, there was a situation where it was difficult to find a “pier”. Then manufacturers need to adjust production strategies in real time, flexibly configure production raw materials, production quantity and quality inspection deployment, produce more “Bing Dun Dun” and less “Xue Rong Rong”. In the entire production process, 3D vision can be used to reduce labor costs, Increase productivity.
Therefore, 3D vision and 2D vision are not simply the information difference in one dimension,The change in production mode, efficiency, and business mode brought about by one-dimensional information is its core essence.
However, the above examples are ideal presets. The reality is that although 3D vision has many advantages, there are still many problems to be solved in order to achieve widespread application.
The difficulty of 3D vision: scene, cost
Different from consumer electronics, the application of 3D vision in the field of intelligent manufacturing is more complicated due to fragmented scenes.
Zhao Qing, founder of Entropy Technology, said in an interview with Leifeng.com that the application of 3D vision technology faces two major difficulties:
- 3D vision technology must have strong adaptability to application scenarios;
- The convergence of 3D vision technology and motion planning technology.
First of all, the production scene of the manufacturing industry is very complex, and the effect of 3D vision in the laboratory may not be reflected in the actual scene, which requires 3D vision to have strong adaptability to the application scene. For example, whether it can still accurately perceive and recognize objects under conditions such as reflection, darkness, film, and long distances.
Secondly, after 3D vision perceives the three-dimensional information of the object, it needs to be connected with the motion planning technology to complete the task. This in turn involves technologies such as collision avoidance detection, hand-eye coordinate conversion, beat optimization, and force control.
However, it is difficult for the machine itself to be like a human. The brain can issue instructions to complete the action; the machine needs to interpret the input information, and transmit the instructions to various parts to execute the order. An error in one of the links will cause the task to fail.
Finally, it is difficult for the technology itself to adapt to various scenarios through standardization, and even in the same scenario, the technical requirements are different. For example, in the defect detection of products, the standards of manufacturers are different, and the definitions of defects are also different. It is difficult to do a standardized defect detection process.
In addition to the problem of sceneization, the sensors (mainly cameras) that 3D vision relies on cannot achieve high resistance to ambient light interference, high ranging accuracy, and high resolution, while reducing costs and improving cost performance.
Therefore, the current application of 3D vision mainly selects the camera according to the usage scenario and budget, and then conducts customized algorithm development according to the camera imaging results. This high-cost and long-cycle development model severely limits the application of 3D vision in practical scenarios.
The Road to Domestic 3D Vision Technology: Difficulties
According to the statistics of the China Machine Vision Industry Alliance, the domestic machine vision industry is dominated by small and medium-sized enterprises, and companies with sales below 100 million yuan account for 83.5%, while Keyence’s sales have already exceeded 10 billion yuan (32.161 billion yuan in 2020). ), in comparison, the domestic company with a revenue of over 100 million yuan is Opte (642 million yuan in 2020, only 2% of Keyence’s).
It can be said that in the field of machine vision dominated by 2D vision, the global market has formed a monopoly situation of Keyence and Cognex, and the emergence of 3D vision technology is regarded as a technological thrust to change the current pattern.
As a new technology, 3D vision is faced with the difficulty of sceneization, which is a problem that all enterprises must solve. At present, whether it is foreign Keyence, Cognex, or domestic security giants such as Hikvision, or many AI vision companies and machine vision companies, they are all on the same starting line in the field of 3D vision technology.
However, compared with foreign giants, domestic enterprises inherently have three shortcomings.
- understanding of the scene.
Whether it is Keyence or Cognex, they have been established for decades and occupy most of the machine vision market; years of accumulation make them more advantageous in exploring the application of 3D vision, and many scenarios can be difficult. Based on past experience, try to reduce unnecessary expenses.
The establishment of domestic enterprises is relatively short, and the understanding of the scene needs to be explored step by step, and even detours are required, which costs a lot of time and capital.
- Lack of hardware capabilities.
The main logic of machine vision is to analyze and process the collected image information, and intelligent devices make corresponding judgments based on the processed information. In this process, the quality of the lens and the lens plays a very critical role in the accuracy of the obtained image information.
Most domestic companies start with software algorithms, focus on the application layer, and lack corresponding hardware capabilities. Most of the core cameras of domestic 3D vision are outsourced, including IDS, Cognex, Keyence, Canon, etc. In terms of lenses, the high-end market is still monopolized by foreign brands such as Leica, Schneider, Nikon, and Fuji.
The Links: 3HAC027697-001