Original title: industrial Union Liu Zongchang: value opportunities brought by Industrial artificial Intelligence and Industrial Internet to Manufacturing Industry | CCF-GAIR 2020

August 7-August 9, coordinates Shenzhen, China. The 2020 Global artificial Intelligence and Robotics Summit (CCF-GAIR) kicked off at the JW Marriott Hotel in Qianhai overseas Chinese Town, Shenzhen, hosted by the Chinese computer Society of China (CCF), co-hosted by Lei Feng.com, the Chinese University of Hong Kong (Shenzhen), and co-sponsored by Pengcheng Laboratory and Shenzhen Institute of artificial Intelligence and Robotics.

At the "Industrial Internet Special Show" held on August 9, Liu Zongchang, chief data officer of Foxconn Industrial Fulian and general manager of the science and technology services group, made a major speech entitled "Industrial artificial Intelligence and the value opportunities brought by the Industrial Internet to the Manufacturing Industry."

Industrial Fulian, Foxconn Industrial Internet Co., Ltd. They have been listed on the Shanghai Stock Exchange for two years since they were listed on the Shanghai Stock Exchange in June 2018 as representatives of industrial Internet companies who not only understand industrial manufacturing, but also can act as "pilot fields". During this period, they officially launched the service master architecture of Industrial Cloud platform (Fii Cloud), Professional Cloud (Micro Cloud) and Industrial artificial Intelligence (IAI). At the same time, it also establishes the two-wheel drive strategy of "intelligent manufacturing" and "industrial Internet".

In his speech, he first introduced the lights-out factory of USI in Shenzhen, and talked about how to achieve zero downtime, zero defective products, zero waste, and inheritable worry-free manufacturing. Then it interprets how to use artificial intelligence technology to help manufacturing enterprises mine data value and enhance competitiveness.

He said that the main scenarios can be summarized into three opportunities: first, the interconnection between AI and equipment to enable it to achieve transparent management; second, enterprises have set up a large number of their own information systems, including ERP and MES, such as automatic scheduling in resource management and resource allocation in mainframe production. Now there are a lot of industrial software, Siemens, ICT and so on have been used a lot. Third, AI and human-computer interaction give birth to new ways of work, such as the current use of AR technology to do equipment maintenance, including equipment maintenance assistance, but also the extensive use of iPhone or Pad, its assembly has reached the accuracy of watches.

Speaking of smart manufacturing systems, Liu Zongchang said:

"We say that we have to go through five stages of development, and there is no way to make a jump in these five stages of development. When our Lean is not done well, it may be very difficult to do informatization, not to mention big data and intelligence. "

"at present, there are specialized scientific and technological service groups that serve many enterprises externally. We have contacted hundreds of enterprises, and we have paid attention to many excellent enterprises, both in terms of scale and profitability. They have turned the first two parts into the head position of the enterprise."

Then, Liu Zongchang systematically introduced the concept of Foxconn lighthouse factory: from "no one" to "worry-free", the three major business sectors of industrial Fulian technology service, and how to realize the full scene integration of intelligent manufacturing system.

The following is the full text of Liu Zongchang's speech, Lei Feng.com(official account: Lei Feng net)It has been edited and sorted out without changing the original meaning:

Good afternoon, everyone! It is a great honor to participate in this event on behalf of USI today. The theme I share today is "artificial Intelligence and the value opportunities brought by the Industrial Internet to the Manufacturing Industry".

As I am giving a speech on behalf of industrial manufacturing enterprises, first of all, I have a sense of substitution, which is a light-out factory pushed within the group.

In this factory, after a large number of automation is realized, the equipment is interconnected with a lot of IoT technology, and the lights-out production is realized in many factories. We give each production department a "light-out rate index" every year, and how many lights-out rates can be achieved in a workshop or a complete product line production process this year.

After the lights are turned on, we can see that in the whole workshop, the automation of a single equipment and the collaborative automation of habitual sequence, including the process of loading and unloading, should be said to be relatively perfect on the basis of hardware. The AGV, the loading and unloading manipulator and the intermediate machining center are independently developed and produced by Foxconn.

With regard to on-site mistakes, in fact, many of them are invisible to us, and the invisible content is the focus of our control. For example, there is no abnormal vibration of these equipment, whether there is parameter deviation in the production process, including the setting of parameters and whether the on-site operation is standard; because the life cycle of electronic products is getting shorter and shorter, the mode of manufacturing has changed from mass manufacturing to small batch and multi-variety manufacturing in the past, and the market has changed greatly, with frequent changes in orders. So, how can the coordination of our resources be allocated optimally all the time?

These invisible problems will bring a lot of losses, including downtime, defective products and efficient use of resources. In the past, the management of these losses depended more on experience and the management ability and leadership of the front-line team. What we are doing now is whether we can put the data through modeling and analysis methods, some industrial AI systems, and turn it into the basis to support our decision-making. In this way, we can always know what is happening on the front line, and when there is a risk, we can predict and prevent it very early, and the value it brings to us or the goal we are pursuing is called "zero downtime, zero defect, zero waste".

In the past, the experience of front-line teams, excellent managers and excellent technicians can be passed on. Whether it is the industrial Internet, or the use of AI, this is a very simple goal and idea as a manufacturing industry.

The key to the use of artificial intelligence technology is how to help enterprises tap such value and constantly enhance their competitiveness. The main scenarios are summarized into three opportunities:

First, AI and devices are connected to each other, so we can achieve transparent management. Whether it is predictive maintenance or very automatic and efficient quality inspection of AI, including process parameters are also analyzed in the way of AI to achieve automatic anomaly identification. This is actually the method we use on AI and devices, which uses a lot of edge operations.

Second, a large number of enterprises have established their own information systems, including ERP and MES, such as automatic scheduling in resource management and resource allocation in mainframe production. Now there are a large number of industrial software, such as Siemens, ICT and so on. If the enterprise has achieved information upgrading, there are many digital threads, there will be some key points to do some scheduling manually, including manual decision-making, the key nodes of key threads can make decisions intelligently by introducing AI technology. We are now doing the advanced scheduling of APS, and we have realized flexible mixed-line production, including the electronic products we use. In fact, all BOM has become more and more complex. We have received a lot of orders on how to organize the field equipment resources and the supply chain, including the mainframe of production resources, including the scheduling of hourly online equipment, which used to be done by people using Excel. Now we put this system online. And the processes of daily, weekly and hourly scheduling are all done in the way of AI. Turn the human scheduling production planning line of the past 4 hours into a minute level, and the efficiency of scheduling is more than 10% higher than that of manual scheduling and the use of resources in the past.

Third, AI and human-computer interaction give birth to a new way of working. For example, AR technology is now used to do equipment maintenance, including equipment maintenance assistance, as well as iPhone or Pad, which is widely used now, and its assembly has reached clock-level precision, so we now use VR when training front-line or even assembly workers, so that he has a more real environment to understand how to do this kind of assembly. Our training cycle can be shortened. For example, in the past, a team did five days of training, but now it has become less than three days. These are all ways to reduce costs and increase efficiency by using AI technology.

As an enterprise, we pay more attention to application-oriented R & D, which can be classified into five categories from an application point of view:

1. Model-based optimization. We combine the algorithm of AI with the mechanism and control theory to realize the on-line intelligent adjustment of the machine, whether the CNC machining center achieves automatic compensation, or now a large number of injection molding equipment adjusts the molding parameters, this is the process of automatically identifying the abnormality of the process and going to the process of self-closed loop adjustment after people found that there are defective products in the past, and people come to the scene to judge what is the reason for adjustment.

2. Decision-making oriented to key processes. Like the APS just mentioned and now in the procurement strategy, such as monitoring the water level of inventory, forecasting the next order situation, so as to optimize the scheduling of the entire supply chain.

3. There is also enhanced analysis, which is now widely used. Whether it is the intelligence pause room or the interactive terminal of the first-line equipment, there will be some reporting methods, including automatically pushing the information of interest on the spot or abnormal information to assist personnel to make decisions.

4. Intelligent working mode, there are also technologies that talk about flexible robots and AR/VR hybrid reality. Now to automate the knowledge, based on the knowledge graph, automatically recommend to the front-line personnel, use the question and answer method to enter a character, recommend to him what caused it, and what to do next? this is the application of our more intelligent way of working.

5. Manufacturing process management, more PHM, using predictive methods to manage quality and equipment condition maintenance.

The above are the five major application scenarios we have summarized. When we do the development of these scenarios, including the internal AI development team, it is difficult to achieve large-scale value delivery in industrial scenarios, and there are many challenges.

As you can see in the figure on the left, if we look at the time dimension and field value, the value curve brought about by the energy we put into and the funding for research and development, at first we find that AI application development will do a lot of experiments, and all five major scenarios will be piloted. In the process of pilot implementation, its development cycle is relatively long, basically 6 months to 1 year, which has a lot of problems.

For example, it will be a long process for us to solve what problems, what data to prepare, to choose the appropriate algorithm model, to constantly tune the parameters, to choose the appropriate computing environment after tuning, and to turn it into the whole program and system. After the basic completion, introducing the practical application of field production, the problem we are faced with is that many factors are taken into account in the laboratory environment or R & D environment, which may far exceed its complexity in the actual production site.

The reliability and interpretability of the results will bring a lot of deficiencies, in the initial introduction, such as the introduction of a new technology by a production line, all the stability and fluctuations in capacity and yield brought to the production line will be adjusted. This process will take 3-6 months, and after we have passed it, it is now basically stable in our actual field. When we want to popularize it on a large scale, we will say that its model generalization ability and migration cost are too high, and this production line has its own characteristics. The next product is not a similar product, but when it is similar to the scene, there is still a lot of development work to be done.

How to shorten the previous time and reduce the marginal cost? Using a systematic engineering method to do artificial intelligence, including such as the industrial Internet, very often people regard the industrial Internet as the concept of the Internet, or look at the artificial intelligence in industry in the way they used to develop artificial intelligence in the past. but we find it challenging to do it that way. We want to do it in a systematic way. First of all, we think that the standard development process is very important. In addition to introducing platform tools like DevOps, we also study some algorithms and use cold start and iterative mechanisms.

There are many typical small data environments in the industry. In the small data environment, how our algorithm iterates quickly, including fast convergence, is the direction of our research now. Including the systematic technical architecture, based on the CPS 5C architecture, each layer of tasks are clearly defined, which do device connections, which do network communication, which do edge operations, which do model management and iteration, and how to distribute results and interactions when connecting with OT systems.

The deep integration of OT and Know-How used to focus on the dimensions and four V elements of data in the process of doing big data, but I think the physical meaning is very important when it is used in industry, including how to define and select appropriate features in preprocessing, including full correlation with the mechanism of process and equipment, which is all related to the Know-How of industry. Including industrial reusable, transferable design, which is an effort as a systematic engineering approach.

In a systematic approach, we must first understand the positioning of artificial intelligence and industrial Internet technology in past systems.

From the earliest method, through people to manage things and things on the spot, this is the earliest method. If you go back to the 60's and 70's, the whole Japanese manufacturing has a concept called TPS production management system. Pay attention to the full mobilization of everyone on the front line, around the equipment on the site, around the elements of the production process on the site to do management, at that time more emphasis on third-line management, first-line leadership and staff mobilization.

In the past, it was how people manage things and things at the scene, relying on people's experience and sense of responsibility, and then there was a system. The system was first born in six Sigma, using data to manage and define the uncertainty brought by the manufacturing system, including another four schools, which is also originated from Toyota's lean manufacturing system in Japan. in the whole process design, management method and system design, and even how to fully lean in the whole production line design process, at that time we pay more attention to the design of culture, system and on-site equipment. Later, with the informationization of manufacturing, the introduction of information systems, including the beginning of MES, ERP, to follow up the offline process with the software system to undertake.

Now with the industrial Internet of things, the industrial Internet of things does not represent the industrial Internet, nor does it represent AI. Its main positioning should be how to inform people of the current situation of things and things in a timely manner, so as to liberate people from a lot of front-line operations or inspection of these affairs. This is what we were doing in the industrial Internet of things in the past. I was pushing the content ten or twenty years ago.

Later, we began to organize things and things in a systematic way, and the content used at this time is called CPS, through the basis of the Internet of things, and then to the upper edge of the operation, and then to the entire physical world of things or physical objects of the digital expression, including its current state of cognitive prediction and how to feedback to the human interaction system. Many of the things we are doing right now, whether it is digital egg laying or CPS, are doing such things, which is also a very basic element for us.

Can it be changed to use the system to assist people to do all the decision-making and management? In the past, people or people made the main drivers or decisions in the system, but now can we use the system to assist people in decision-making? This involves many applications of AI and the situation awareness and risk early warning of the current system, including enhancing the ability of analysis and management, including how to make decision optimization under complex conditions, and systematizing people's fragmented experience with the technology of knowledge graph, which in turn helps to deal with front-line problems. On the whole system, the relationship between people, things, and systems, as well as the roles that different technologies should play in each relationship connection.

When it comes to intelligent manufacturing systems, we say that we have to go through five stages of development, and there is no way to make a jump in these five stages of development. when we do not do a good job in Lean, it may be very difficult to do informationization, not to mention big data and intelligence.

We now have a special science and technology service enterprise group to serve many enterprises externally. With contact with hundreds of enterprises, we pay attention to many excellent enterprises, both in terms of scale and profitability. They have become the head position in the enterprise when they can improve the first two parts.

We are in lean improvement how to achieve rationalization and standardization, how to do a good job in the construction of staff culture, including the improvement of end-to-end value, through the systematic way and management way to improve the management level of the enterprise. With these two as the basis, we will make the process transparent later, putting all the processes online and to what node the process is going to, so that its data performance can be reflected digitally. You may pay attention to the views of the front-line MES, including the enterprise operation decision center can be done on this.

After we have achieved digital manufacturing, the vast majority of domestic enterprises, including the Ministry of Industry and Information Technology, have identified many intelligent manufacturing benchmarking as being in the third stage. To the next stage, we focus on digital networked manufacturing, in which we focus on how to use data to drive cross-process collaboration and intelligent decision-making.

If we can do that, in the past, the human-driven process has now become a data concatenation of all processes, and the key processes use data to drive what to do next, and whether the process nodes should run down, which are all being built now. If we also achieve it, we think we can achieve independent wisdom in the next five years or less, because we can achieve a record after the completion of each automatically transferred event to realize the automatic accumulation of data. If we solve the closed loop of feedback, then we can know whether the decision made automatically by the system is good or bad, then we use this data to continuously optimize our decision-making system behind our back. in this way, we can achieve a sustained and autonomous state of wisdom.

Intelligent manufacturing systems have several key features:

The goal to achieve is to increase efficiency and reduce cost, and to fully release the production capacity per unit of equipment, including the optimization of cross-process coordination.

Break through the bottleneck of quality in the key process

Transparent management in the operation process, all parts can be standardized

How to speed up the building of personnel capacity, there are fewer and fewer front-line operators and experienced technicians, but there are more and more application development and IT system maintenance.

If we want to achieve such a goal, we actually need a bearer, which is an excellent end-to-end operation, and we need to build a full-process information platform.

From Foxconn's point of view, we have two key threads, one is the main line of order management OTD, the other is the main line of product R & D management. Because we help our customers to do ODM, they may just be the concept of the product, and we help it do the whole product ID design and manufacturability design, this part is the main line of product research and development. On the two main lines, there will be a digital decision center to collect and analyze all the key information in the middle, and then distribute it to each process node after the decision. This is what we are doing to build the entire end-to-end excellent operation information platform.

There are many typical application scenarios, which are not analyzed one by one, including factory planning, production line design and manufacturing process management, to the digital expression of our factory, and to the management of front-line equipment. There are many key scenarios. For example, in our end-to-end operations, such as the acquisition of orders, the generation of delivery plans, supply chain management, the execution of production plans, front-line management of production and manufacturing, including our final transportation delivery, there are also many scenarios that can be optimized with industrial Internet and technology.

In the past, the concept of lighthouse factory has always been the pursuit of unmanned, using equipment to replace human labor, now we pay more attention to worry-free, worry-free is how to grasp the risk that did not occur at the scene in time, and avoid it through prediction and prevention. We focus on 3W, how to reduce waste, how to reduce labor, how to reduce anxiety.

Finally, I would like to share with you a video, this is our WEF World Lighthouse Factory. This is a very typical household production line of electronic products, including in addition to automated production, automated site review, this is the use of AOI technology to do patch quality inspection. Including the use of on-site AGV, on-site equipment management, and now a large number of assembly links can also be completely autonomous. In the soldering process, the application of solder head Patch can monitor the welding process, when to do solder joint defects and solder head prediction and prediction system. The intelligent correction of the equipment will be automatically pushed to the mobile end when there is a problem with the equipment, which will be dealt with by the staff. Now the packaging link, even the very complex packaging process can also achieve complete automation.

Some production lines used to use a lot of labor, each production line is generally 318 people, now reduced to 38 people. The labor saving is very obvious, in addition, the improvement of production efficiency and inventory cycle is very obvious.

We now have three major technology service sections, including the overall solution of the lighthouse factory, from the planning, design, detailed design of use cases and overall package implementation of the lighthouse factory; in addition, there is the introduction of the entire operation system and personnel training, and there are hardware and software integration solutions in different scenarios; and there are 1bn industrial Internet platforms to provide services on the supply chain.

Our vision is to realize the full scene integration of the intelligent manufacturing system, from the front-end intelligent hardware to the network, and then to the above software applications, including cloud data analysis and our massive AI usage scenarios, and finally, after all the scenarios are integrated, to achieve the improvement of customer manufacturing capability and end-to-end wisdom value. Thank you!Lei Feng net

The original article of Lei Feng net is forbidden to reprint without authorization. For details, please see the reprint instructions.