Artificial intelligence (AI) technology has developed very rapidly in recent years, but there are not many real landing projects at present, and the most used image recognition and speech recognition may be. If the market scale of artificial intelligence is to be larger, it has to enter many subdivisions and land in more subdivisions. The industrial field is a topic that cannot be avoided.
Artificial intelligence (AI) technology has developed very rapidly in recent years, but there are not many real landing projects at present, and the most used image recognition and speech recognition may be. If the market scale of artificial intelligence is to be larger, it has to enter many subdivisions and land in more subdivisions. The industrial field is a topic that cannot be avoided.
In fact, many application scenarios in the industry require the support of artificial intelligence, such as the industrial Internet, smart factories, edge computing, and so on. So can artificial intelligence successfully land in the industrial field? What are the requirements? At the China Robotics Summit not long ago, Professor Zhu Yunlong, a distinguished professor of the Institute of Engineering and Applied Technology of Fudan University, shared his views. A small area achieves certain task autonomy, which is to follow the path of violent calculation of “big data, small tasks”. In the future, if you want to successfully land in the industrial field, you want to take the road of precise calculation of “small data, big tasks”.
What is the next explosive point for artificial intelligence?
In Professor Zhu Yunlong’s view, there are two development directions for robots in the future, one is extremely large; the other is extremely small. Professor Zhu Yunlong took the small killer robot shown at the United Nations Weapons Convention Conference in Geneva as an example. The robot is only the size of a bee, but its analysis and processing capabilities are 100 times faster than that of humans, and it can avoid all kinds of human tracking.
Why is this small robot so capable? Because of the blessing of artificial intelligence. So where will the next burst of artificial intelligence be?
Professor Zhu Yunlong first talked about the existing artificial intelligence model, “At present, everyone talks about big data and deep learning when they talk about artificial intelligence. Let’s look back now. Under the current circumstances, deep learning algorithms claim to have reached, even It is over 1 million times in 2000, how to calculate? Simply, the algorithm breakthrough here is a hundredfold breakthrough. The leap in computing power and the surge in computing power are exactly 100×100×100.”
That is to say, the current artificial intelligence technology basically revolves around big data, and performs calculations based on calculations for deep learning. Simply put, it is a very small task that needs to be calculated through a large amount of data. He believes that this is a violent calculation.
In fact, when human beings make a decision, they do not need a lot of data. It is a “small data, big task” model. For example, if a human is exposed to dangerous goods, he will react quickly based on some experience without the need for a lot of calculations.
He believes that the future of artificial intelligence should move towards precise calculations and adopt the model of “small data, big tasks”. Only in this way can artificial intelligence have a new world.
To realize the “small data, big task” model, algorithms are the key. There are various algorithms. How can algorithms find breakthroughs in the future? Professor Zhu Yunlong believes that there are two ways, one is to imitate human intelligence as much as possible; the other is intelligence from the universe and the intelligence of biological evolution.
Therefore, in the view of Professor Yunlong Zhu, the next burst of artificial intelligence in the future lies in algorithms.
What are the pain points of the manufacturing industry?
The problem of artificial intelligence, in the manufacturing industry, is also facing a common problem, that is, the fragmentation of manufacturing process data, and the large amount of data has not been fully explored and applied.
Professor Zhu Yunlong said in the sharing that artificial intelligence may produce profound changes in all levels of people’s production and lifestyles, especially in the industrial field. The manufacturing industry must be the most important point in the industrial field. “Although we have been working for so many years, we still can’t see how to integrate the massive data of the manufacturing industry to produce a profound change in the product design, quality monitoring and efficiency improvement of the manufacturing industry.”
Therefore, he believes that industrial big data presents its concealment, low quality, fragmentation, dynamic multi-mode, complex and strong correlation, low signal-to-noise ratio and other characteristics. In this case, it has distinct data field management characteristics and fields. Application characteristics, and traditional manufacturing enterprise information integration technology cannot meet the requirements of manufacturing enterprise big data, efficient organization and in-depth application. In this case, this is the current problem of the entire industrial big data.
In the manufacturing industry, there are real-time production scheduling optimization, edge data integration and network control, multi-layer supply chain network business process decision control models, parameter control in the manufacturing process, etc. These are all complex optimization problems in intelligent manufacturing. It seems that Professor Yunlong Zhu can be simply boiled down to the problem of optimization and control.
How to integrate artificial intelligence with manufacturing
How does artificial intelligence integrate with manufacturing? At the beginning, artificial intelligence was mainly centralized. All big data was processed in the cloud. However, in the actual use process, because all the data needs to be transmitted to the cloud for processing, the amount of calculation is too large. Make the processor a bit unbearable. Therefore, edge computing came into being.
With the complexity of industrial application scenarios, there will be a large amount of data to be processed at the edge, otherwise the amount of data processed in the cloud will be very large. In this case, the future applications of edge computing may bring about profound changes, and people’s decisions at this level are resolved by edge computing.
So, what is the more important difference between edge computing and cloud computing? Cloud computing focuses on non-real-time, long-period big data analysis; while edge computing is closer to the execution unit, emphasizing low latency and fast response, that is, the perception, autonomy and intelligence of the device… This requires edge computing to coordinate cloud computing.
In addition to edge computing, visual perception is also very important. Data shows that with the development of communication network technology, 5G can reach 20Gbps, 6G can reach 100Gbps, and the image information received by the human eye is 30 frames per second… The information of the surrounding environment received by the future machine far exceeds that accepted by humans. Speed, then in this case, that is to say, we can feel that the photoacoustic current media can pass through the machine vision angle and perhaps greatly exceeds the speed accepted by the human eye.
This brings about a problem. The amount of information received by machines exceeds that of the human brain. Machine learning is currently a small task of big data. How to solve the problem in a scenario where the human brain is a small data and large task must be an algorithmic problem. “We think Group intelligence computing is a more important issue.” Professor Zhu Yunlong said.
Therefore, he believes that the future manufacturing state is based on CPS manufacturing, which is data-driven, software-defined, and platform-supported manufacturing. It will be a real-time interactive manufacturing between physical manufacturing and virtual manufacturing. Its evolutionary process has changed from fragmentation to integration, from partial To the overall situation, from static to dynamic; data flow gradually covers the entire process of R&D, design, manufacturing, and service operation. It is a process of continuous extension and expansion of the closed-loop data flow system, and gradually forms an interactive complex data network space.
Moreover, it realizes cross-system and cross-platform interconnection, intercommunication and interoperability, and facilitates the closed-loop automatic flow of integration, exchange and sharing of multi-source heterogeneous data.
Finally, comprehensive information perception, in-depth analysis, scientific decision-making and precise execution will be realized on a global scale, and horizontal, vertical and end-to-end integration will be realized.
In general, the future manufacturing model will be the integration of manufacturing systems, the reconstruction of manufacturing systems, and the reconstruction of manufacturing models.
“Under this situation, if we say that in the future, people will play a weaker role in the manufacturing system and the large-scale manufacturing system. Then I personally think that it may be based on visual perception, with hybrid cluster intelligence computing and optimized control as the core. Our unmanned workshop/smart factory has truly entered the era of unmanned operation.” Professor Zhu Yunlong said.
Concluding remarks
In the view of Professor Zhu Yunlong, whether it is robots or artificial intelligence technologies, they are facing a technological critical point. From the perspective of the manufacturing industry, there may be a critical point in the software architecture. A new breakthrough; as far as the robot itself is concerned, the traditional robot sees tangible objects. In the future, the eyes, brain and arms of the robot may be scattered in the manufacturing system and become a ubiquitous intelligent robot. In this scenario, it will definitely be Bring a corresponding technological breakthrough and theoretical breakthrough, and then, these breakthroughs may bring major industrial applications.