Page 167 - AI for Good-Innovate for Impact

AI for Good-Innovate for Impact

(continued)

Domain Steel manufacturing industry
Key aspects • Workers previously used outdated methods to inventory steel bars one
of the solu- at a time which was slow and prone to error. By using the AI-based image 38-Huawei&XISC
tion detection capability, during the smelting stage, the steel bars can be auto-
matically identified, and the number of steel bars can be quickly calculated,
yielding higher levels of accuracy with a shorter turnaround.
• During the stage of steel rolling, workers can control the turning and speed
of the rolling mill using a trained cloud-based AI model and video data
from 5G-connected cameras. This method of rolling is smarter and more
efficient, as 1 more steel billet rolled per mill can be produced in an hour,
resulting 30,000 more tons rolled every year and annual revenue increased
by 100+ million CNY.
• With the help of AI data filtering, the Pangu steel model is also designed
to pre-train production, safety observation, and decision-making models
based on the principle of marking anything not normal as abnormal. These
models require only a small number of samples for training and boast
a 10% higher accuracy in detecting abnormalities compared to smaller
models. An AI-based analysis and automatic detection of abnormality is
adopted by the Xiangtan steel, have replaced the manual inspection that
workers had to walk around factory floors with over 1000 °C temperature.
• With images and videos collected for data training and real-time detection
ensure efficient operation and predictive maintenance, minimizing device
downtime for O&M reduced by 17% and maintenance costs of annual
motor damage rate from 5% to 2%, reallocating workers’ time and digital
skills to be better used (number of spot check personnel in slabs and bar
is reduced by 10 to 4, and O &M status can be checked on their mobile
phones).
• With the help of AI, Xiangtan Steel's workers have been able to operate
more efficiently, and in a much safer environment.

Technology Smart manufacturing, 5G+AI+Cloud+Industrial Internet, abnormality detec-
keywords tion, remote control, safety observation, AI-based image detection, computer
vision

Data avail- Data is privately available upon request
ability

Metadata Structured and unstructured data, descriptive data, administrative data
(type of
data)

Model • Predictive model with decision trees, regression and neural networks
Training and • Huawei OptVerse AI Solver with Large Network Linear Programming,
fine-tuning Convex Continuous QPLIB, Lpopt and MIPLIB2017 for model parameter
optimization
• Grid Search (gridsearchCV) for hyperparameter tuning.

Case Studies • https:// www .youtube .com/ watch ?v = KbuVEqj61TA
• http:// www .csteelnews .com/ qypd/ gl/ 202405/ t20240506 _87708 .html

Testbeds or • https:// www .analysysmason .com/ contentassets/ 3f8 280c2fe264
pilot deploy- a029bb5c96 8046b3a95/ analysys _mason _huawei _xisc _5g _may2021
ments _rma18 .pdf
• http:// www .csteelnews .com/ qypd/ gl/ 202405/ t20240506 _87708 .html
• https:// m .cnfin .com/ cy -lb/ zixun/ 20240429/ 4042612 _1 .ht

151