The goal of this project is to develop a data‐science approach for alloy design and apply it to Magnesium (Mg) alloys. Researchers will build an artificial neural network (ANN) model to establish the relationships between material composition, microstructural features and mechanical properties. They will train the ANN model first for commercial and other well‐characterized Mg alloys, and then extend it to a wider range of composition and processing histories. To this end, researchers will generate physics‐based simulation data to inform the model regarding the effects of various alloying elements and processing conditions, along with their prediction uncertainties. Finally, a few new Mg alloys with superior mechanical properties will be identified for potential commercialization using the model.
Objectives: As a lightweight metal, Magnesium (Mg) alloys possess higher specific strength, i.e., strength‐to‐ weight ratio, than Aluminum (Al) alloys and steels, making Mg alloys very attractive as structural materials for lightweighting automobiles and aircraft structures. To increase the usage of Mg alloys in structural applications, however, further improvements are required for both their absolute strength and ductility. Current alloy development efforts mostly adopt the “trial and error” approach. A large number of Mg alloys with different chemical compositions have been developed over the past few decades, and a few that show good mechanical properties (e.g., yield strength, ultimate strength, strain to failure) and low cost have been identified and commercialized, including AZ91 (Mg‐9Al‐1Zn, wt%), AM60 (Mg‐6Al‐0.1Mn, wt%), and ZK61 (Mg‐ 6Zn‐1Zr, wt%). While these alloys are acceptable for current applications, improvements in strength are required for new applications such as automobile side‐impact beams, A‐pillars and aircraft floor supports. Compared to most metal systems, Mg alloy development is at a relatively immature state. For example, most common Mg alloys have only 1‐2 alloying additions, namely Al, Zn or Mn. In contrast, advanced aluminum alloys often have 6‐9 alloying elements and advanced nickel‐based superalloys can have up to 10 alloying elements.
A major drawback of the “trial and error” approach of alloy development is its low efficiency. It relies heavily on the “knowhow” of the metallurgist to introduce certain microstructure into the material that may lead to improved properties. The problem with this approach, in which new elements are added to change specific microstructural features, is that in many cases a certain alloying element may have multiple effects on the final microstructure and different alloying elements may react with each other. Therefore, it is difficult to expect what microstructures and material properties would ultimately results when alloy compositions are changed. Due to these complications, existing theories are unable to predict material properties of Mg alloys with arbitrary compositions and it can take 10‐20 years to develop and implement a new alloy.
In this project, researchers will develop a data‐science approach for alloy design and apply it for Mg. The team will build an artificial neural network (ANN) model to explore the relationships between material composition, microstructural features, and mechanical properties. Because precipitate strengthening is the main strengthening mechanism for advanced cast Mg alloys, we simplify the problem by only considering the precipitate characteristics (phase compositions, volume fractions, size, aspect ratio, etc.) as the relevant microstructure features. The model will be first trained by the data available for commercial and other well‐characterized Mg alloys to achieve a good confidence. Simultaneously, the team will generate physics‐based simulation data to inform the model regarding the effects of various alloying elements and processing conditions, along with their prediction uncertainties. Combined with such data, the ANN model will be used to predict the precipitate characteristics and the resulting material properties of Mg alloys with a range of chemical compositions and heat treatment history. Finally, a few new Mg alloys with superior mechanical properties will be identified using the model for potential commercialization.