Header | Big Whale AI closed-door meeting
Produced by | Tiger Sniff Think Tank
Problems such as the lack of data, the high accuracy of the results, and the need for the flexibility and scalability of the model in factories are all problems that hinder the implementation of generative AI in industrial scenarios.
In order to solve the application problem of AI in the industrial manufacturing industry. On June 20, Tiger Sniff Think Tank invited AI decision-makers from leading enterprises in the industrial field such as China Academy of Information and Communications Technology, TCL Zhonghuan, Guoxuan Hi-Tech, Midea Group, LONGi Green Energy, Schneider, Baidu Intelligent Cloud, etc., to give us an in-depth analysis of the application trend of AI in intelligent manufacturing, the implementation of AI applications in specific scenarios such as quality inspection, energy management, and employee standard operation improvement, as well as the cases of large models in the industrial field, as well as how enterprises can identify the effective landing scenarios of large models.
I hope that this excerpt of the views of the conference can give some inspiration to those who are not present, and readers and friends are welcome to have in-depth exchanges with us.
The following is a summary of the views shared by the panelists:
Dong Xiaofei, Deputy Director of the Platform and Engineering Department of the Institute of Artificial Intelligence, Chinese Academy of Information and Communications Technology
The new generation of artificial intelligence, represented by large models, is one of the important engines for the development of new quality productivity
There are at least three "new" new quality productivity, namely new workers, digital employees, new production factor data elements, and new production tools and intelligent tools. The large model is one of the important engines for the development of new quality productivity, which is mainly reflected in the fact that the new agent with the large model as the core gradually replaces repetitive and procedural physical and mental work, and becomes a new worker with new quality productivity.
Large model technology is applied to all aspects of social production to fully release the value of data elements. Driven by large models, machines evolve from automation to intelligence, and are endowed with perception and cognitive capabilities, constituting new production tools for new quality productivity.
TCL Central CIO Qu Benqiang
The integration of AI+IT+OT+IE is the key to enhancing the competitiveness of photovoltaic manufacturing
In recent years, the output of all links of China's photovoltaic industry has increased significantly, entering the era of T-watts. In the future, market competition will intensify, and how to enhance core competitiveness has become a problem that photovoltaic companies have to face.
Qu Benqiang, CIO of TCL Zhonghuan, said that the key to enhancing the core competitiveness of photovoltaic manufacturing is to promote the integration of IT+OT+IE, so as to accelerate the digital transformation and competition of enterprises, flexibly respond to market changes, and enhance core competitiveness.
Promoting the integration of AI + IEOT is a systematic project, which needs to be comprehensively considered and coordinated from the aspects of strategic planning, technology implementation, talent training, and organizational change. At the strategic level, enterprises need to define convergence goals, such as improving production efficiency, reducing costs, and improving product quality, and clarify the respective roles and responsibilities of AI, IT, OT, and IE. At the technical implementation level, we will build a data-driven decision-making system, use IT and IE capabilities to collect and analyze data, provide a high-quality data foundation for AI, identify production bottlenecks through data analysis, and use IE and AI for optimization.
In terms of talent training and organizational change, enterprises need to form an interdisciplinary team composed of experts in different fields to cultivate compound talents who understand both AI and IEOT, while encouraging innovation and continuously exploring new applications of AI and IEOT.
Liu Shengxiang, a senior expert in forward-looking manufacturing technology at NIO
Starting from the business pain points of manufacturing, we will achieve 80% of business AI policy
The application path of industrial AI can be divided into three stages, namely, technology-led point empowerment, demand-led complex multi-scenario empowerment, and general empowerment of multi-modal scenarios, and the application value of AI is continuously amplified with the progression of development stages.
At present, AI-empowered industry is still in its infancy, and the combination of AI and industry is scenario-driven, and enterprises should look for opportunities to solve problems from small scenarios. Examples include AI quality inspection, intelligent analysis, and energy management for automobiles. In terms of AI quality inspection, factories used to have problems such as poor consistency of personnel operations, inconsistent standards, and poor cost efficiency, and AI vision can solve the pain points of this scenario. Intelligent analytics, enterprises can use AI tools to solve complex multi-factor production bottleneck analysis and predictive analytics; In terms of energy management, AI can help factories optimize energy management and reduce energy waste.
Fu Xu, President of Midea Group's Intelligent Manufacturing Research Institute
Research on sensing technology and automation solutions to promote the application of AI algorithms in the manufacturing field
Midea shared a number of AI application solutions based on sensing technology, including the detection of appearance defects of home appliances, the appearance inspection of products before they go offline, the improvement of employees' standard operations, the intelligent real-time analysis of action waste and line balance, the detection of abnormal sound of products, and the predictive maintenance of production equipment.
For example, the appearance defect detection of air conditioning panels: image acquisition is done through multi-dimensional cameras, and visual analysis is carried out at the same time to achieve rapid detection of multiple products, which not only improves the detection efficiency, but also ensures the detection results.
In terms of improving the standard operation of employees: in order to solve the frequent production changeover and complex process operation caused by mass customization, the AI camera is used to capture fingertip/millisecond movements in real time, analyze based on the real-time data of the production line, and deploy an expert system to identify problems through AIGC and provide suggestions for the waste of operation actions and the improvement of production line efficiency, which reduces the operation error rate of employees and poor operation quality, and improves the line balance rate.
Focusing on the intelligent real-time analysis of action waste and line balance: Based on AI visual inspection and 3D digital twin technology, real-time dynamic analysis of production line operation cycle and production line balance problems can improve operation cycle and line balance.
Product noise detection: Based on AI, the noise detection of the motor can be carried out, and the standard and consistency of the product can be improved in the application through self-developed testing equipment and algorithms.
Predictive maintenance of production equipment: The generalized hardware and software integrated operation and maintenance solution of production equipment in the home appliance industry is adopted to carry out intelligent perception and virtual measurement of real-time equipment status and key process status, and realize functions such as system fault early warning, root cause reasoning, and predictive maintenance.
Zhang Hongyu, head of the AI algorithm system R&D department of Guoxuan High-tech Engineering Research Institute
In the intelligent transformation of manufacturing, enterprises need to pay attention to the development of four key technologies
In the context of the development of artificial intelligence, enterprises need to pay attention to the development of four key technologies, namely perception technology, reasoning technology, decision-making technology and interaction technology. Perception technologies include visual algorithms, signal processing algorithms, natural language processing, and speech recognition algorithms. Inference techniques include causal analysis, association analysis, clustering algorithms and prediction algorithms, and decision-making techniques include decision trees, Bayesian networks, Markov decision processes, genetic algorithms, fuzzy logic and predictive learning. Interactive technologies include AGI, LLM, Gen AI, etc.
The common algorithms of the four technologies are used to solve the needs of business scenarios such as process, quality, equipment, and R&D, so as to promote the intelligent transformation of enterprises.
Zhang Ningning, Director of New Manufacturing Industry Solutions of Baidu Intelligent Cloud
The exploration of large-scale model capabilities in various scenarios has gradually matured
When exploring the application of large models, enterprises need to clarify the capability boundaries of various models, including large language models, CV large models, multimodal large models, and scientific computing large models. Enterprises can only better solve business problems by clarifying the capability boundaries of various models and matching different business scenarios.
During the meeting, Baidu shared a number of cases, including using large model capabilities to solve the construction of enterprise knowledge base, reducing the learning cost of employees' learning standards and knowledge; Use the large model to connect to the database to carry out certain data governance, so that the business data can be answered immediately, and the production situation can be comprehensively and quickly analyzed. The CV model is used to realize the safety control of hot work; Through natural language dialogue, the ability to integrate large and small models, and system APIs to achieve end-to-end business closed-loop service of device O&M.
Liang Zibo, analyst of Tiger Sniff Think Tank, Li Yuanbo, head of big data AI of LONGi Green Energy, Mao Chunjing, chief digital designer of Schneider Electric, and Zhang Ningning, director of Baidu Intelligent Cloud New Manufacturing Industry Solutions
In the roundtable session, Tiger Sniff Think Tank and the guests discussed the current implementation cases and effects of large models in the industrial field, as well as the future development trend of large models and small models that enterprises are concerned about.
Mao Chunjing, chief digital designer of Schneider Electric, said that the landing of large models in industrial scenarios faces a dilemma. On the one hand, the production-related data, knowledge, documents and experience of manufacturing enterprises are one of their core competitiveness, which cannot be shared, resulting in poor data flow, and it is difficult to appear a large model of vertical industries, which benefits industry; On the other hand, if a single enterprise does a large-scale vertical fine-tuning model for privatization, it will face high training costs and difficult maintenance, resulting in high costs related to the enterprise and the large model, and it is impossible to rationalize its investment decisions.
Zhang Ningning, director of Baidu Intelligent Cloud New Manufacturing Industry Solutions, believes that the exploration of large models is still in its infancy, and in the actual contact process with customers, the three directions of large model applications can be continuously explored in the early stage and can achieve results. First, knowledge management, that is, knowledge question and answer scenarios; Second, "ask the number", by opening up the business data, so that the business data can be answered immediately, and the production situation can be comprehensively and quickly analyzed; Third, task scheduling, by opening up the API of the previous information system, using a large model to judge how to schedule.
As for the future application trend of large models and small models, the guests at the meeting all said that the selection of models should be considered in combination with the needs of the scene. Li Yuanbo, head of big data AI at LONGi Green Energy, said that in the selection of large models and small models, enterprises need to combine the needs of actual industrial scenarios, and will not and do not need to pursue the use of large models to solve problems. In addition, in addition to ROI, information security is also a core factor for enterprises to consider in the use of AI in any scenario.
Now log on to the official website of Tiger Sniff Think Tank to learn more about AI innovation practice insights and research.
About Tiger Sniff Think Tank:
Tiger Sniff Think Tank is a new research service organization focusing on enterprise digitalization and AI innovation practice, providing insightful research reports, case selection, online meetings, offline activities and visits for both parties in the process of industrial intelligence, so as to support the wise decision-making of enterprise executives in intelligence and digitalization.
The core values we offer:
Timely and high-quality insights, understanding technology, understanding the industry, understanding peers and competitors;
It provides an important reference for decision-makers in technology and product strategic decision-making, industrial planning, and solution selection;
Help the market fully understand the development status of cutting-edge technologies and the industries they affect, as well as future trends.
People who are changing and want to change the world are all in the Tiger Sniff APP