What about artificial intelligence, each side simply makes sense, since the definition of the Dartmouth Conference “AI” after the word, artificial intelligence itself has gone through several peaks and valleys, it is not popular today, this wave tide, but the tide has gone through more, and because it does not contribute anything and quiet, but humans never stop researching artificial intelligence, has been trying to use it to solve some problems, realistic business value, even though today artificial in business there are many smart success, however, in terms of its industrial sector, but it seems lackluster, do not know which scenes have better. The market also brought a lot of sound “fuzzy” confusing, optimism believe artificial intelligence will subvert manufacturing, and other ideas and think AI is not really so amazing – people expect too much of artificial intelligence, Especially in the industrial field, there are many different voices control Engineering Copyright , as there are now artificial intelligence including machine learning, deep learning and other industrial scene and applied a greater difference, industrial people seem more conservative and based on realistic considerations, not on artificial intelligence remain overly optimistic, however, cross-border areas there is always space, let us try to analyze how to bring new opportunities for the industry through the integration of science and technology innovation.
1. Extreme 1.1 on artificial intelligence to analyze several industrial and innovative thinking to avoid a commercial rather than think for dry AI industry, one is that AI can subvert the industry. Like Musk Conference on Artificial Intelligence in Shanghai, said this year, in fact, Musk said machine this thing is not a real person than clever, AlphaGo chess this matter are reasoning under uncertainty learn the rules, and the industry has not like that scenario, uncertainty, disturbance, Go is complex can not be compared.
AI think 1.2 is replaced by the era of automation, and the automation, information, intelligence as dynastic history, like that automation is the “past”, is actually a history of AI development and industrial application of AI is not clear, behaviorism school Wiener is typical, industry is the norm rather than a linear, or linear state is even referred to as “special case.” AI want to subvert the industry had erected “Flag” is not only a little, even Minsk, Herbert A. Simon this big brother constantly being hit artificial intelligence set a milestone own face, therefore, do not easilyTo say “subversion”, because those new economy to subvert the traditional industries, we can not survive more than C round. The new economy has too much leverage and capital strength amplification factor in it, the more money to burn, but also relatively fast, if this logic is equivalent to ten months a woman to have children, then the woman whether it is ten a month can give birth to a child it? Many logic can be used in the business world, but not in the spirit industry, this is normal. 1.3 Pan AI there is nothing to talk about the necessity of the algorithm is that it is the lack of AI is easy to be confused with the fuzzy boundary caused. For the current definition of AI it seems more and more broad, presumably to be able to catch the AI this outlet, right? What do algorithms, software, just to catch up with the outlet when almost unable to pay wages to the closure of the company can be labeled as “Industry 4.0 benchmark” the same, so do a rented cyberspace will become a “edge computing” subject, this is the entertainment capital, and find open source algorithm, and then find a training scenario at the data, you can claim “AI start-up companies”, then go to the capital market misappropriating, some time ago to see a program, probably similar to Xu Xiaoping and other Lord chiefs to see the project of entrepreneurs, alas! People really feeling, we do industrial circles might really do not understand those who do not fly Why can apply to be heavyweights fancy, willing to invest a few million, accounting for a 10% stake …, capital is so “there money, self-willed, “whether it be fly do not fly, but no one than this tricky time, this is tricky. “AI must be defined” – this is a scientific attitude, what is AI? This is critical, not just to get some personal algorithm, to put forward a vision known as AI, AI’s core is going to do? – Problem solving their own decisions, that is, its hard core that “decision”, if only sensor perception, it is just a question of automation, if in order to control, and that is the automation of control problems, if there is no reasoning, analysis and ability to make independent decisions, this can not be called AI, you can be called “expert systems”, “dictionary” but you can not just arbitrarily defined as AI, especially when used in industry. 2. Consider the problem of differences in logic and industrial and commercial business in a greater difference in the application of artificial intelligence in industrial applications particularity of 2.1, as communication is in accordance with “the highest performance (Up to the IT field ) “, and the people at the manufacturing site of the talk about” the worst situationCondition (Worst Case) “, which reflects the fact that two completely different way of thinking and attitudes, further, the biggest difference is the stability and reliability, as emphasized in industrial control data transmission of” certainty ” is “accurate”, “interpreted”, many of these applications are currently in the field of artificial intelligence, neural networks connected either doctrine or previous symbolism of reasoning are unable to meet industrial applications for “reliable”, ” precision “,” robust demand “, which makes it necessary to combine industry itself, to develop effective applications of artificial intelligence, which in turn requires a combination of several important thought: AI people have high hopes, however, the industry seems to there is no artificial intelligence in the end to be able to see what to do? icing on the cake if artificial intelligence can do some things rather than temporary relief, then artificial Intelligence is a dispensable beautiful girl, in the business world, you can find a large number of users. , anyway, there are so many people, as long as 1% of the people can be interested will be able to create a very The market, however, the industry is another scenario, you must meet in order to have a high accuracy rate with customers – as recognition of defective products, like 1% inaccuracy for mobile phones in terms of production, there may be a number million and hundreds of thousands of defective products be punished, which is clearly unacceptable.
2.2AI industrial applications is a systematic project which is a really big problem, difficulties exist, AI applications industry such as Li Professor Jay thought “is a systematic project,” sensory information, transmission, data cleansing, data feature extraction , and, ultimately, control and execution are all must work closely with you even if there is AI but can not function effectively, because, you collect accurate or not depends on the accuracy of the sensor, and feature extraction of your data depends on the technology industry, and at the time they need to consider the implementation of mechanical actuators … AI characteristics of successful application of AI is not a problem in itself, and indeed is a systematic project, many aspects involved, the complexity of the relationship between each other, even at present Unknowable, how to build AI application itself requires a lot of collaborative logistics planning is a very interesting question, this matter so I was surprised – some time ago and a logistics industry friends talk to a large number of industrial site warehouse known as “smart warehouse”, but, in fact, veryMulti-warehouse and not a smart plan, but a FIFO (First In First Out) of the same cohort, the problem is not the optimal storage location based on frequency of use, weight and other programs, is said to obstacles encountered is “encoded” the problem is that many factories will be the lack of good coding system, I think of OPC UA in the function of AutoID, if each product to do this, then the data can be applied to real intelligence analysis and optimization of industry, therefore, this perspective, the industry must be a system of intelligent engineering.
3 Innovation: to solve economic problems, even if there are cognitive limitations or expansion between the two produced a variety of emotions, in fact, do have to realize that the introduction of new methods and tools to solve the traditional problems, is still a worthwhile shop and go thing, where there will be opportunities for cross-border, which is also numerous facts have proven, innovative also occurred on the border, to abandon prejudices into the other side of the world, perhaps the AI and industry need a peaceful state of mind , it is not in the chain despised different objects. This need to think about the following questions? 3.1 Which scenario is emerging problems too? In industry, the very sensitive for AI and AI, or for networking and networking behavior, all the goals are to improve quality, reduce costs, improve delivery capabilities to address these lean production target personalized, there is no objection to this problem, because the real business operations can not take a national project to invest a lot of manpower buckle down to this thing. (1) problems of the manufacturing site of the connection if the existing production under mechanical constraints has reached a limit, by tapping the potential of “connection”, it is a problem, but a lot of problems to be solved and which non-mechanistic model is difficult to resolve, therefore, need the help of “learning” to solve nonlinear problems, then this problem is a lot of it? Scheduling and planning: This requires an existing model or a model to learn it? (2) Fine demand management generated which is the angle of approach demand-driven, in order to solve the quality problems, the use of vision for product defect detection, but the detection needs to involve recognition of the problem, and require a certain algorithm to process the data, and after extensive training, pre-judgment can be given, however, for existing vision applications, the mostThe complex is on the complex conditions of production, it requires a lot of configuration settings in which people participate, definition, calibration, etc., the need for people with high professional requirements, whether it exists makes it more cost-effective way to do that? 3.2 How to reduce dependence on people? Many industrial applications, whether it is data-driven model or mechanism, it is actually for people with very high demand, just as predictive maintenance, must rely on a very professional international certification vibration analysts can participate in this prediction, Therefore, the problem of artificial intelligence to solve the industry should focus on how knowledge explicit – how knowledge can be packaged as one of APP, let AI algorithm training data simultaneously combined experience and wisdom of the human form of knowledge automation package this is a need to work on machine learning experts and experts in the field of engineering aspects of the common algorithm carried out. Whether traditional process can be shortened? AI, including machine learning, reinforcement learning, neural networks these, how they will change the manufacturing sector? This needs to think from a question: – What can we really give us the possibility by learning? Many of the traditional industrial accumulation CONTROL ENGINEERING China Copyright , such as the commonly used vibration analysis for predictive maintenance, depending on the expert system, and Worcester leveling accumulated yarn parameters 200 years, while the spring and sheet metal bending process, etc., are required to accumulate a lot of decades of engineering data accumulation, the process may be whether to change based on learning data? It can be replaced if the original design of mechanistic models based on the new method? Without spending huge time cost to the accumulation of these problems is feasible? These methods are new tools can play a range. The likelihood of these scenarios are dependent on a number of areas of common people to explore, including joint efforts in many fields of machine learning, process engineering, electrical and mechanical, sensor and communications, because, AI is a must for industrial applications systems engineering.
3.3 more economical conditions whether they have a? AI application in industrial economics from these considerations is the former technology used in industry because of force to bring the count, the maturity of software such uncertainty, the cost is too high and has not really been applied, the manufacturing sector is a “pinch pennies “the field, therefore, there must be economic support. (1) calculate the force problem: in fact veryMostly in traditional industries in the solution to the problem is the “most economical” approach, because if you want to achieve maximum precision and control of the rehabilitation of industrial scene, then you need a very strong algorithm design, and this is again a strong operator support force, as it simplifies the nonlinear system to a linear problem, in turn, can reduce the calculation amount, including if the complex algorithms to control, then the processing resources required operator force is not a common controller can be done under , and for industrial purposes, but also to meet reliable, quantities not large, in fact, considered a powerful force for industrial applications is still very expensive, if this operator force It was resolved – as computing power of mobile phones are now on top of the Manhattan project as a whole.
(2) the model algorithm and problems: because after extensive training of model able to be reused. This could be considered potential knowledge reuse. In the past many years, it may not have considered the use of data-driven, machine learning, deep learning method to solve industrial problems, including the knowledge of how it is expressed in software, models, and form a real value – the current thinking It is the first data, which is not excessive, and knowledge of how to play the dominant value is a potential application. (3) improvement of new tools and methods: The problem is that there is a problem “tool”, in fact, the biggest consumer and commercial software is powerful, “ease of use” design, once the tube and Microsoft, Mr. Zhen chatted about “Toy “- said someone said OT do IIoT platform toys, I said that we do look at IT platform is also a toy – like to say is not practical floor applications, however, Mr. Guan said to me,” we have to do is toys ah! we play together, “I’d been convinced him, they saw the machine learning tool, that is really simple, really easy to get started, they might learn a few days I could get this, but I think the real the difficulty is still the process, but at least the tool itself is simple, easy to make popular entry-up, is a good thing. Therefore, in this perspective, offers convenient tools and methods, in itself solve the economic problem – how to make use of experts in the field of industry is more integrated with low-cost or low cost to the industrial system, making the overall economical, and that costs for existing AI is a dilution process. 4. Analysis: Which scene requires intelligentSystem to dry it? In the field there are a large number of process issues, however, there will be differences in vertical industries, however, question whether the industry can be common problem? We can provide universal model for these common problems and find them widely available training model it? 4.1 parameter optimization using either supervised or unsupervised learning, the problem of finding the best parameters is a relatively common problem, the problem is for the industry, especially the production process, and hope into a discrete continuous process in terms of production are very large accumulated amount, just 1% of the energy savings for a large metallurgical enterprises are significant – Fu teacher Yuya big data is determined here, it is indeed a very industrial feel significance of the place – as CHAI academician called “small data, large application”, can solve the big problems that might accrue each year is staggering cost savings. 4.2 Lean production level of 1% of the upgrade so that the problem of AI focal point in the “Operation” operations, rather than a problem of control now, because of operational problems can tolerate “a mistake”, and the problem is not in control allowed, because once is not up to defective products. Mechanistic models – whether chemical reaction model flow production, or physical mechanical transmission control model, in the traditional mechanistic model is often economical and effective, it has the advantage of a simple algorithm model solves 90% or even 99 % of the problem, then, for the remaining 1% of the problem need to solve it? This issue also has a reasonable judgment, that is, if the average level of 95%, then to 96% can be achieved in terms of yield, it will mean a larger profit margin, because that 1% difference in yield terms for conversion profits may not 1% but 10%, therefore, in this sense also exist, however, put much power can be generated this upgrade? AI certainly need to do? This can be a problem as, obviously, the scene gods, needs to be judged by people who live, but the current problem is that both sides not to know each other, AI do not understand the scene, the scene does not understand AI, it is a difficult conversation. 4.3 Process matching problem? For each print, the process parameters need to be adjusted according to the conditions, whether there is an optimal process parameters to match with existing technology it? Textile yarns are carried out by the thousands of possible combination of fiber, thatWhat if there is an optimal combination of process parameters, such as setting the optimum value of the draft ratio calculated? For all kinds of plastic particles melt, plastics molding process, whether there is an optimal matching parameters? To do this automatically calculate the optimal drive control parameters can it? 4.4 path planning problem? Path planning problem in many scenes exist, logistics is one aspect, however, path planning may not need to learn in most applications, the artificial intelligence, programming models are available to realize, however, it has the versatility training model can solve the problem of planning a variety of occasions? 4.5 Analyzing and simple classification of such problems, by various parameters including the visual, sound or the like on the object recognition and judgment, and then classified, including statistical classification thereof by a robot, and the corresponding feedback feature for the processing of adjustment values, which are 4.0 for the industrial production of small and medium volume in a dynamic intelligent analysis and judgment, and implementation issues. 5. What are the premise of the need to solve the problem? 5.1 the need for innovation and evaluation of national money should be in basic theory and engineering methods, and companies must have a self-driving force. (1) must be clear, AI industrial applications is a systematic project, not an AI can solve the problem, whether it is machine learning, or neural networks, the depth of learning, is just a tool, but to solve the problem must be required of the system, and more interdisciplinary be built, and the use of scientific methods and engineering process to achieve. (2) must be based on the value assessment, rather than to cater to this trend, or to declare the capital invested in a project in which basic issues must be resolved, artificial intelligence must be established in an efficient, standards of production technology and process base above. 5.2 cross-border talent, especially the problems of industrial software development involves knowledge of mechanical and electrical objects, software, technology, long cycle, and complex but the industry did not open high salary, the equivalent of an AI expert training in the industrial field takes the cost of cost several times in the business world, and this market is not big enough, because 1% of the population as a user that is a lot of data in the commercial sector, and for the industry is another scene, you spend the cost is very high, but it is nothing more than small application and paid the price, because this particular big obstacle for the industry.
(2) AI software algorithm to understand more, fewer understand the process, you need complex, even the basic resultsType itself is small. Of course, it is that these people are often reluctant to dry in the industry, because the industry would like to have a little difficult achievement, need to understand the process, you need to understand the various complex site, such as acquisition, signal processing, communications, feature extraction doing things … very much, but not as concerned about obtaining a large market potential in other areas, investors, and tend to sink like a monk, like, a little difficult to attract talent into it. 5.3 Data standardization is the difficulty of this integration of IT and OT, of course, if we are to reach agreement on this issue, using the OPC UA or other uniform standards and specifications, which are can be resolved, however, the two worlds people really lack of basic mutual understanding, always in a state of waiting for each other, IT people usually question is “What is your data?” and people tend to OT question is “What data do you want?”, two after between individuals on this issue for many years did not pull clear, everyone thinks he is very clear, IT believe that you have the data I can train and learn, can be optimized, while OT people think, you do not do not understand the scene, the scene complicated than you might think, you do not think you can take it for granted, like a lot of people think that IT is already 10mS real-time tasks, and you ask people OT end, they did not think that called real-time systems. In short, this is an exploration, we focus on the integration of innovation, through multi-party cooperation and jointly explore the potential of new technologies, new methods, tools can play the AI for the manufacturing, welcome to explore.