Noise-Aware Active Learning to Develop High-Temperature Shape Memory Alloys with Large Latent Heat.

Area of Science:

• Materials Science
• Thermodynamics
• Data Science

Background:

• Shape memory alloys (SMAs) are promising for high-temperature thermal energy storage (TES) due to their large latent heat during phase transformations.
• Designing SMAs with desired properties, like high latent heat and transformation temperatures, is challenging due to the vast compositional space and noisy experimental data.

Purpose of the Study:

• To develop a data-driven method for accelerating the design of NiTi-based SMAs with high latent heat for elevated temperature TES applications.
• To address the challenge of noisy experimental data in materials design through a novel noise-aware active learning strategy.

Main Methods:

• A noise-aware active learning strategy was proposed, incorporating a noise-aware Kriging model.
• The optimal noise level was estimated by minimizing model error across a range of noise hyper-parameters.
• This approach was used to search the component space of doped NiTi-based SMAs.

Main Results:

• The strategy led to the discovery of an SMA with a latent heat of -36.08 J g^-1, exceeding the best previous value by 9.2% within four additional experiments.
• The newly discovered alloy exhibits a high austenite finish temperature of 481.71°C and relatively small hysteresis.
• This represents a significant advancement in identifying SMAs for latent heat TES in high-temperature environments.

Conclusions:

• The noise-aware active learning approach effectively accelerates materials design using noisy data.
• This method facilitates the discovery of SMAs with superior performance for high-temperature thermal energy storage.
• The developed strategy is adaptable for data-driven materials design in various fields facing noisy data challenges.