A team of researchers at Lehigh University has developed a powerful machine learning model that can predict when materials might fail. Their system can spot abnormal grain growth in metals and ceramics long before it becomes visible. This discovery could help engineers design stronger, more reliable materials for demanding environments, such as engines, aerospace parts, and other high-performance products.
Materials like metals and ceramics are made up of grains, or crystals. When exposed to extreme heat, such as in a jet engine or rocket, these grains can grow or shrink. If some grains grow much larger than others, the material’s properties can change. For example, a material that was once flexible could become brittle and break. This failure often happens in high-stress, high-temperature environments, leading to safety risks and high costs.
Designers aim to prevent this by selecting materials that are less likely to fail. However, predicting abnormal grain growth has been a challenge. Testing materials in the lab takes a lot of time and money. Many materials must be tested before finding one that works well. The new AI model offers a smarter way to find the best materials faster.
The team combined two advanced machine learning techniques. First, they used a long short-term memory (LSTM) network to track how the properties of a grain change over time. Then, they used a graph-based convolutional network (GCRN) to map relationships between grains. This approach helps the model spot early signs of instability.
The AI was surprisingly effective. It predicted abnormal grain growth in 86% of cases, often during the first 20% of the material’s lifetime. This means designers could quickly rule out materials likely to fail and focus on those that are more stable.
This innovation is especially valuable for product designers, materials engineers, and automotive designers working with high-performance materials. By using AI, they can choose materials that are less likely to fail, reducing waste and improving product reliability.
The model also supports sustainability. By avoiding materials that might fail, designers can reduce waste and energy use in the production process. This helps lower the environmental impact of products.
The team plans to test the model on real-world materials, not just computer simulations. They hope it will help create materials that can withstand extreme conditions for longer. In the future, this method could also predict other rare events, such as chemical changes or even disease mutations.
This breakthrough shows how AI can help designers and engineers work smarter, faster, and more sustainably.
Source: Lehigh University
Photo: cottonbro studio