Editor’s Note
Great Wall Strategic Consulting has been compiling and publishing the “Enterprise Research Report” since 1995, conducting in-depth analysis and development assessments in multiple fields such as technological innovation, industrial upgrading, regional development, government governance, and innovative concepts. This article is edited based on the content from the 396th issue of the Enterprise Research Report titled “Characteristics and Typical Formats of New Quality Service Industry” to provide readers with insights.
Industrial Large Models refer to artificial intelligence models established to meet the needs of the industrial sector, aimed at predicting and optimizing the operation of industrial systems. They are widely used in various scenarios and stages of research and development, production, and management, enhancing industrial efficiency, reducing operational costs, and improving product quality.
Industrial Large Models possess three characteristics: multi-dimensional modeling, high integration, and intelligent optimization, making them an important engine for promoting the intelligent transformation of industries.
First, Multi-Dimensional Modeling. Its multi-dimensional modeling capability allows for a comprehensive and detailed representation of industrial systems, covering not only the operational status of physical equipment but also analyzing multi-dimensional factors such as process flows, human resource allocation, and supply chain management, constructing a three-dimensional, dynamic mirror of the industrial system. This comprehensiveness ensures that the model can accurately capture every subtle change in the industrial system, providing a solid foundation for precise analysis and decision-making.
Second, High Integration. Industrial Large Models exhibit a high degree of integration. Within the model, various sub-modules and components collaborate closely to form an organic whole, achieving cross-domain and cross-system data sharing and collaborative optimization. This integration not only enhances the operational efficiency of the model but also promotes the tight linkage of various links within the industrial system, significantly boosting overall effectiveness.
Third, Intelligent Optimization. The deep integration of AI technology endows Industrial Large Models with powerful intelligent optimization capabilities. Leveraging big data, deep learning, and machine learning technologies, the model can autonomously analyze massive amounts of data, simulating the decision-making process of human experts, providing efficient and precise forecasting and decision-making support for industrial production. Additionally, its self-learning and continuous optimization characteristics ensure that the model can keep pace with industrial development trends, continuously adapting to new demands and challenges.
In the 1990s, the development of Industrial Large Models abroad underwent profound changes from theoretical exploration to technological breakthroughs and widespread applications. In the early days, the field of artificial intelligence was still in its infancy, with researchers focusing on rule-based expert systems and knowledge representation systems, laying the theoretical foundation for later developments. Entering the 21st century, especially from 2006 to 2010, the rise of deep learning technology became a key turning point for the development of Industrial Large Models. By constructing deep neural networks, models could automatically learn feature representations of data, significantly enhancing the performance of large models. A landmark event was the research results from Geoffrey Hinton’s team in 2006 and the victory of AlexNet in the ImageNet competition in 2010, marking a solid step forward for the application of deep learning technology in the industrial field.
Since 2018, with the continuous enhancement of data resources and computational power, large-scale pre-trained models have become a new trend in the development of Industrial Large Models. By the end of 2022, landmark achievements such as OpenAI’s GPT series and Google’s BERT model not only demonstrated exceptional performance in multiple tasks but also pushed the boundaries of AI technology. The application of these models has not only improved the intelligence level of industrial production but also injected strong momentum into the transformation and upgrading of the global industrial system.
Currently, China’s large model technology is in a period of rapid development. In January 2024, the State Council’s executive meeting studied and deployed work to promote the empowerment of artificial intelligence in new industrialization, emphasizing the deep integration of artificial intelligence with manufacturing as the main line, focusing on intelligent manufacturing, and accelerating the intelligent upgrade of key industries by leveraging scenario applications to develop intelligent products and empower the industrial manufacturing system at a high level.
In recent years, China has shown strong catching-up and independent innovation capabilities in the field of large model technology, with domestic companies such as Baidu, iFlytek, Tsinghua Zhizhu, and SenseTime rapidly narrowing the gap with international frontiers. At the same time, significant breakthroughs have also been made in the fields of open-source models and multi-modal technologies, further consolidating the technological foundation.
On June 4, 2023, China’s first independently controllable industrial large model similar to ChatGPT—Zhigong Industrial Large Model—was officially launched, marking an important step for China in the field of industrial intelligence. This model has been continuously iterated to versions 2.0 and 3.0, successfully landing in several key areas such as energy management and comprehensive management of electric energy equipment. In addition, Zhigong Industrial Large Model has won multiple awards and has been listed in important rankings such as “AIGC Industrial Chain Beijing Specialized and Innovative Enterprises Map” and “2023 AI Large Model Pioneer List TOP 50,” fully demonstrating China’s technological strength and market recognition in this field.
At this stage, the main entities laying out industrial large models can be divided into four categories: AI manufacturers, industrial technology service providers, innovative growth enterprises, and research institutions/technology giants.
Industrial Large Models, as key tools for intelligent upgrades, are crucial for industrial enterprises and possess vast application scenarios. Unlike previous applications of artificial intelligence technology in single industrial scenarios, Industrial Large Models are based on general large model technology and trained through learning large amounts of industrial data and knowledge. They can not only analyze and judge based on existing data but also generate new knowledge and solutions through training, serving as the foundation for promoting intelligent transformation in industrial enterprises. For example, the iFlytek Spark’s Antelope Industrial Large Model can achieve various functions such as industrial understanding calculations, industrial knowledge Q&A, industrial code generation, and industrial text generation. China’s number of industrial enterprises has reached 500,000, and the demand for intelligent transformation of industrial enterprises provides a vast application market for Industrial Large Models. It is predicted that by 2026, China’s industrial large model market size will exceed $500 million, with a five-year compound growth rate of 116%. Currently, many participants are involved in the industrial large model market, including internet platforms (such as Baidu, Tencent, Alibaba), technology companies (such as iFlytek, Huawei), AI startups (such as Innovation Intelligence), and manufacturing enterprises (such as Haier).
Industrial Large Model application scenarios span the entire lifecycle of industrial enterprises, enabling “cross-scenario” empowerment. In the research and design phase, Industrial Large Models can be used in scenarios such as drug research and industrial product design. For example, Innovation Intelligence’s “ChatCAD” can automatically generate compliant industrial design drawings through a dialogue Q&A format and can be exported to traditional software for fine-tuning. In the production and manufacturing phase, they are mainly applied in scenarios such as industrial code generation, safety supervision, equipment management, and product testing. For example, Feite Testing’s “Feifan Industrial Large Model” can achieve intelligent Q&A for process experts, interconnectivity of equipment data, and generation of quality reports. In the daily operation phase, they can improve work efficiency by integrating with management software and enhance enterprise knowledge management levels through Q&A interactions as the main form.[2]
The successful application of Industrial Large Models relies on the support of computing power infrastructure. The construction of Industrial Large Models requires a large amount of industrial data and knowledge, with parameter counts reaching hundreds of millions or more. For instance, the parameter count of the “Qizhi Kongming Industrial Large Model” released by Innovation Intelligence has reached 75 billion. New types of computing power infrastructure, such as intelligent computing centers, can not only provide powerful intelligent computing resources for the training and inference of Industrial Large Models but also offer ample storage space to meet the data storage needs of large models, forming the cornerstone for the landing application of Industrial Large Models.
Huawei Pangu Large Model: Empowering Thousands of Industries with a Multi-Level Large Model
The Huawei Pangu Large Model includes a “5+N+X” three-layer architecture: “5” represents five basic general large models that provide knowledge Q&A, code generation, and other services available for direct use by enterprises; “N” represents N industry large models that can train industry-specific large models focusing on multiple sectors such as mining, finance, and meteorology based on the general large model; “X” includes specific business scenarios such as lead drug screening, foreign object detection on conveyor belts, and typhoon path prediction. With the help of the Pangu mining large model, Shandong Energy Group has achieved intelligent leaps in multiple links, such as using the Pangu predictive large model to intelligently analyze coal quality data and process parameters of raw coal from different coal seams, different seasons, and different ash contents, promoting an increase in the recovery rate of premium coal by 0.1%-0.2%; the AI main operation intelligent monitoring system of the Pangu mining large model can accurately identify abnormal situations such as large coal blocks and anchor rods, achieving 24/7 inspections with an accuracy rate of 98% for foreign object identification.
Simu Technology IndustryGPT: A Multi-Modal Large Model Focused on Industrial Manufacturing
Simu Technology, founded in 2019 by experts in the field of machine vision and a lifelong professor at the Chinese University of Hong Kong, IEEE Fellow Jia Jiaya, developed the world’s first industrial multi-modal large model—IndustryGPT V1.0. This model can understand and process complex problems in the industrial field through deep learning and big data analysis techniques. The training data for the model covers over 200 different industrial scenarios, more than 3 million industrial images, and over 50 billion tokens (the basic unit for processing text). In the electronics manufacturing industry, IndustryGPT V1.0 can predict potential equipment failures and conduct maintenance in advance, enhancing the stability and reliability of production lines. In the steel industry, IndustryGPT V1.0 is applied in intelligent quality inspection systems, using image recognition technology to automatically identify minor defects on product surfaces, significantly improving detection speed and accuracy.
Although progress has been made in the development of Industrial Large Models in China, there are still certain gaps compared to international top levels in terms of AI penetration rates in manufacturing, implementation in industrial verticals, core production applications, and data integrity and connectivity.In the future, it is necessary to continue increasing R&D investment, strengthen technological innovation and talent cultivation, and promote the landing and deepening of Industrial Large Models in more application scenarios, with a more open and cooperative stance to jointly promote the prosperous development of global industrial large model technology.
[2] Reference material: China Communication Industry Association “Industrial Large Model Application Report”
This article is excerpted from the Great Wall Strategic Consulting Enterprise Research Report No. 396
“Characteristics and Typical Formats of New Quality Service Industry”
(Report Authors: Xu Huan, Shen Wenqi, Cao Xing, Zhou Xin)
