RM新时代有限公司

—At the Sino-German Smart Manufacturing Forum, multi-agent systems become a new benchmark for manufacturing transformation

10[Month]23,, at theSino-German Smart Manufacturing Technology Innovation Cooperation Forum and Alliance Annual Conference held in Chongqing, a discussion on"Industrial Large Models and Industrial Intelligence" attracted industry attention.

Zhang Li, Vice President of CloudWalk Technology, statedDelivered a speech on "Multi-Agent Collaboration in Complex Scenarios", where the top tenAIagent cases applied in Qingshan Industry became the focus of the event.

In the current wave of intelligent manufacturing, a core contradiction is becoming increasingly prominent: many seemingly"universal" general-purpose large models often suffer from "acclimatization issues" once they enter factory workshops. The absolute precision, deep specialization, and real-time response required by industrial scenarios precisely hit the weaknesses of general-purpose large models.

Multi-AgentThe systemBreaking the Dilemma of Industrial Large Models

The intelligent transformation of the manufacturing industry is facing severe challenges.

Industrial environments requireAIthe system to have absolutely accurate execution capabilities, but general-purpose large models may produce "hallucinations" and give irresponsible answers.

It lacks the deep industry knowledge accumulated by veteran workers over decades, and is more like a"generalist" rather than an "expert".

More critically, on high-speed production lines, the response speed of large models often cannot keep up with the pace of equipment, directly affecting production efficiency.

Yao Zhiqiang, co-founder of CloudWalk Technology, pointed out as early as2023that in the industrial field, artificial intelligence requires many long-tail algorithms. Whether it is detecting defects or other recognition tasks, it is necessary to train models separately, collect data separately, and use non-standard methods to solve problems. "Many industrial scenarios cannot useAIto solve problems, because to solve1%performance, you may pay100%the cost."

Facing this industry challenge, CloudWalk Technology has explored a pragmatic path—by building a multi-agent system to create a "digitalexpert team" that can collaborate.

In the practice of CloudWalk Technology and Qingshan Industry, agents have demonstrated two core advantages:"Fast Learning" and "Rich Experience".

The two parties created ten agents, each with its own responsibilities yet working collaboratively, forming a digital expert team covering key business areas of the enterprise:

Knowledge Management Expert:Two agents—the Super Knowledge Assistant and the Indicator Q&A—have become the enterprise's intelligent encyclopedia, responding to various business queries and data inquiries at any time.Among them,document query response timehas been significantly reduced from an average of15minutes to within3seconds, achieving a rapid transformation from"novice"0 to"to expert"at lightning speed.

Production Quality Expert:FTTThe Rework Analysis Officer can accurately analyze rework data and identify root causes of quality issues; the Equipment Maintenance Advisor assists in fault diagnosis and maintenance decisions, improving operational efficiency.These agents have analyzed over4000maintenance records, and can automatically diagnose faults based on problem descriptions and provide solutions based on historical data..

Security Monitoring Expert:The Server Security Guard continuously monitors and generates comprehensive threat analysis reports to ensure stable system operation.

Training and Office Expert:The Training Assistant can quickly convertPPTmaterials into video courseware, revolutionizing employee training; the Certificate and Seal Assistant handles general document writing and seal processes, enabling automated office work.

"These ten agents are not isolated 'tools' but a collaborative 'digital expert team'." Zhang Li, Vice President of CloudWalk Technology, introduced that the agents are linked through CloudWalk's hierarchical social network architecture. The decision-making layer (e.g., Indicator Q&A Agent) sets goals, the coordination layer (e.g., Super Knowledge Assistant) connects data chains, and the execution layer (e.g., Equipment Maintenance and Security Monitoringagents) implements operations, forming aperception-decision-executionPerception-Decision-Execution-Feedback" complete closed loop, improving the overall efficiency of relevant production processes at Qingshan Industry by.

This"digitalexpert team" means that experts in areas such as production scheduling, operations, and maintenance no longer work in isolation. Instead, they can share information and make collaborative decisions around common production goals, achieving a fundamental leap from point automation to global intelligence.

AI"AI infrastructureSupporting China's manufacturing fromfrom"following"0 to"to leading"

Where do such professional"digital employees"come from? Especially in industrial fields that require interaction with the physical world? The answer lies in a solidAI"AI infrastructure.

solution is to build anAIinfrastructure system that provides a safe and controllable"training base".

for the cultivation and growth of industrial agents. This system consists of three core components:

First is the controlled training ground.It is like a"central kitchen"and"or written exam room", providing basic data, algorithm training, and capability assessment for agents..

Second is the embodied intelligence training ground.This is the agent's"practical training base"and"or road test venue", allowing agents to undergo tens of thousands of operational training sessions under real physical rules and with real equipment, accumulating"muscle memory".

for handling complex situations. All trial and error is completed in this safe and controllable environment, and agents are deployed to real production lines only after they mature, greatly reducing the risk and cost of intelligent transformation in the physical industry.

Third is the pilot base.It provides a full-process toolchain and services to accelerateAIthe transformation of technology from the laboratory to the production line..

The practice of CloudWalk Technology and Qingshan Industry's"TenAIAgents" not only confirms that multi-agent is an important trend in industrial intelligence evolution but also demonstrates the implementation capability of Chinese enterprises in deeply integrating cutting-edge technology with complex industrial scenarios.

With the advancement of the State Council's "Opinions on Deeply Implementing the"Artificial Intelligence+Action", popularizingAIagents has become a national strategy. The cooperation paradigm between CloudWalk Technology and Qingshan Industry provides a highly valuable Chinese model for the intelligent transformation of the manufacturing industry.