Page 518 - AI for Good Innovate for Impact

AI for Good Innovate for Impact

(continued)

Item Details
Key Aspects 1.Establish a unified architecture platform, we call it “1+3+3+N” architecture.
of Solution It can enable real-time data interaction between cloud and device terminals,
aggregate multi-source/heterogeneous industrial data, provide automobile
industry-specific large model development tools, and empower intelligent
scenarios across "R&D, production, supply, sales, and service”.
2. By leveraging Artificial General Intelligence (AGI)-powered text-to-image/
image-to-image generation technologies, the system shortens the concept
design phase in styling workflows while reducing product homogenization.
3.Apply Large Language Model (LLM) + Graph Retrieval Augmented Gener-
ation (RAG) technologies to enhance the efficiency of failure analysis in full
vehicle design processes.
4.Leverage Computer Vision (CV) large models to minimize repetitive manual
defect detection tasks and improve industrial quality inspection efficiency.
5.Implement predictive large models and large language models for 24-hour
monitoring and analysis of equipment health status, reducing unplanned
downtime.
6.Using solver technology to achieve intelligent production scheduling, lower
inventory costs, and respond to unexpected events proactively.

Technology Digital foundation, Internet of Things (IoT) cloud-edge collaboration plat-
Keywords form, Industrial big data platform, Industrial AI large model platform, AGI,
Large language model, Computer vision large model, Solver
Data Avail- Private data collected by the Guangzhou Automobile Group (GAC) Co., Ltd.,
ability such as equipment predictive maintenance database, intelligent production
scheduling dataset, visual quality inspection dataset

Metadata text, image, video
(Type of Data)

Model Train- Industrial cameras are used to collect image/video from the production line,
ing and using the labelled data to fine-tune the CV large model, and use the edge
Fine-Tuning deployment to implement efficient defect detection.
Testbeds or The deployment model can be found at [4]
Pilot Deploy-
ments

Code reposi- Not available
tories

2 Use Case Description

2�1 Description

• Traditional automobile manufacturing has a long design cycle. There are more than 1000
scenarios for vehicle design failure analysis, which is a heavy workload. The key quality
inspection procedures rely on manual inspection and the ability of quality inspectors.
The health status of the equipment is achieved through manual inspection, which has
problems such as insufficient timeliness and difficult operations in elevated or enclosed
areas. Production scheduling plans are passively received, with great plan fluctuation,
affecting the spare parts preparation of the downstream supply chain.

482