Predictive Maintenance · Real-time Analytics
Chinalco needed to apply AI across a dispersed nonferrous-metal value chain where complex process control, safety risks, fragmented data, and plant-specific deployments made traditional optimization and inspection methods too slow and inconsistent.
Reported impact
Kun’an turned industrial AI into a reusable platform across eight business domains, helping Chinalco lower cost, improve scheduling and process control, and automate image-based materials analysis at industrial scale.
Chinalco Group used Huawei Cloud Stack to build the Kun'an large model as a group-wide AI platform for mining, smelting, processing, safety, supply chain, R&D, management, and services. The foundation is a unified computing center, data platform, and AI development platform built with cloud-edge collaboration, where the group trains L1 and L2 models centrally and then lets subsidiaries request compute and invoke models without moving data out of domain. On top of that shared base, Chinalco rolled out application models such as Zhisheng Alumina for evaporation, dissolution, and sedimentation control; Green Aluminum Cloud Wisdom for intelligent aluminum-electrolysis prediction and scheduling; Aluminum Intelligence Cloud Scheduling for cold-rolling plans; and Metal Intelligence Eye for metallographic image analysis. The pipeline spans data preprocessing, feature engineering, model prediction, and online production control.
Operational data from mining, smelting, and processing systems is collected into a centralized data and AI platform, where Chinalco trains L1/L2 models and scenario applications. The resulting models are then invoked by subsidiaries for process prediction, scheduling, inspection, and online production control without moving sensitive data outside each business domain.
"Chinalco built the Kun’an large model on Huawei Cloud Stack and has applied it across 8 business domains and 18 scenarios. (original: 中铝集团基于华为云Stack打造坤安大模型，应用于8大业务领域18个场景。)"
"Chinalco jointly built a unified computing center, data platform, and AI development platform, using cloud-edge collaboration so subsidiaries can request compute and call models after centralized training. (original: 中铝集团通过联合搭建统一算力中心、数据平台与AI开发平台，利用云边协同架构，集团负责L1模型和L2模型训练，模型训练完成后，各下属企业按需申请算力、调用数据模型，实现大模型在集团内各企业快速落地和复制推广，也满足了数据不出域的合规要求。)"
"Green Aluminum Cloud Wisdom brought more intelligent prediction and automatic scheduling to aluminum electrolysis and is estimated to save more than 85 million yuan per year. (original: 云铝股份基于坤安研发的“绿铝云慧鉴”，是铝电解行业首个大模型... 据测算，该模型在云铝文山500千安电解系列上推广应用后，每年可为企业降本8500余万元。)"
"Metal Intelligence Eye can quickly identify aluminum-alloy microstructures and reached 99.5% accuracy with more than 90% efficiency improvement. (original: 中铝材料院开发了金属材料显微组织AI大模型分析软件——“金属智眼”，可以批量快速地识别铝合金组织，并统计分析其面积和尺寸分布，准确率达到99.5%，效率提升90%以上。)"
"Zhisheng Alumina achieved end-to-end closed-loop management from data preprocessing and feature engineering to model prediction and online control, while reducing main-control workload by 85% and delivering annual cost savings above 10 million yuan. (original: 广西华昇基于坤安大模型发布了氧化铝行业首个AI大模型“智昇氧化铝”... 实现了从数据预处理、特征工程、模型预测到生产在线控制的全流程闭环管理。据初步估算，广西华昇生产主控操作工作量降低85%... 年降本超1000万元。)"
Review how this reference was scored and trace each normalized claim back to its documented source.
Scoring methodology
Customer Reference Value measures usefulness as a customer reference. Evidence Confidence measures the quality and traceability of available evidence, not independent verification.
Traceability
Field coverage shows which facts each source supports. Dates are shown only when documented by the source.
Related
Huawei Cloud Sample Archived · Jan 15, 2026 Conch Group Conch needed to scale AI across cement production, but traditional model development was slow, hard to generalize, and too weak to support real-time quality, energy, equipment, and safety optimization at industrial… 28 equipment-monitoring scenarios AI training center · Huawei Cloud Stack · NLP large model 
NVIDIA Sample Aug 25, 2026 Siemens Siemens needed to help industrial teams overcome fragmented engineering tools, labor shortages, and slow design and maintenance workflows so they could design, simulate, and operate complex products and factories faster… 10%–15% reductions in capital expenditure NVIDIA CUDA · NVIDIA DGX B200 · NVIDIA DGX H200 
In-house Sample Aug 13, 2026 Lockheed Martin Static flight-hour tables and usage metrics ignored real-time consumption, causing chronic under-resourcing, limited stock effectiveness, and a reactive sustainment supply chain that could not keep pace with modern… predict component failure up to 72 hours before it occurs LogC2 dashboard · deep-learning models · digital data fabric 