Highlights

• Chinese researchers developed the first industrial-academic database of nearly 2,000 real NdFeB magnet samples.
• Use of machine learning and active learning optimizes composition faster and cheaper than traditional trial-and-error methods.
• The AI system creates a continuous learning loop that identifies uncertain areas, requests only the most useful data, and retrains.
• This innovation bridges the gap between industry's cost focus and academia's performance goals.
• This breakthrough indicates China is advancing beyond manufacturing scale to algorithmic advantage in rare-earth magnets.
• Rare-earth magnets are a critical chokepoint for U.S. and European electric vehicle and wind turbine supply chains.

---

Chinese researchers report a major step toward AI-driven manufacturing of high-performance rare-earth magnets, a critical component in electric vehicles and wind turbines, not to mention other major products. By combining nearly 2,000 real industrial magnet samples with advanced machine-learning and “active learning” techniques, the team claims it can optimize magnet composition and processing faster, cheaper, and more reliably than traditional trial-and-error methods. If validated and scaled, this approach could tighten China’s lead in NdFeB magnet manufacturing, a strategic chokepoint for the U.S. andEurope.

What’s New — and Why It Matters

According to a February 2, 2026 announcement from the Computer Network Information Center of the Chinese Academy of Sciences (opens in a new tab), working with the Ganjiang Innovation Institute of the Chinese Academy of Sciences (opens in a new tab), researchers have built what they describe as the first “industrial–academic dual-domain” database for sintered NdFeB magnets. The dataset contains nearly 2,000 samples drawn from real manufacturing environments rather than idealized lab conditions.

Using high-performance computing and multiple machine-learning models (random forests, gradient boosting, and classical and quantum-inspired support vector regression), the team tested how data selection strategies affect prediction accuracy and manufacturing outcomes in a virtual experiment setting.

Reviewing the Materials

The Chinese Society of Rare Earths included an infographic that illustrates an active learning production loop:

• Data preparation: Raw magnet data are cleaned, standardized, and splitinto training, validation, and test sets.
• Model training: Multiple AI models predict magnet performance based on composition and processing parameters.
• Active learning loop: Instead of training once, the system repeatedly selects the most informative new samples, retrains the model, and stops early when gains level off.
• Outcome: Faster convergence on optimal magnet recipes with fewer costly physical experiments.

In simple terms: the AI learns where it is uncertain, asks for only the most useful new data, and improves continuously—cutting time and cost.

Strategic Signal for the West

The study explicitly highlights a tension: industry prioritizes cost and stability, while academia pushes performance limits. The Chinese team claims its framework bridges this gap, offering a scalable path to factory-ready AI optimization. For the U.S. and EU—still struggling to localize NdFeB magnet supply—this signals that China is moving beyond scale alone toward algorithmic manufacturing advantage, potentially widening the competitive moat.

The peer-reviewed results were published in npj Computational Materials and funded by major Chinese state research programs.

Disclaimer: This news item originates from media affiliated with a state-backed Chinese research ecosystem. While technically plausible and published in a reputable journal, all claims should be independently verified before being used for investment or policy decisions.