Process (computing)

Magnesium alloy

Metallurgy

Materials science

Operating system

Magnesium

Particle swarm optimization

Computer science

Artificial intelligence

Artificial neural network

Machine learning

Computational Materials Science

Developing high-performance, low-cost magnesium (Mg) alloys using conventional plastic forming processes is a tremendous challenge with great potential for commercial application. However, the current research and development for Mg alloys are still based on "trial and error" methods, which are inefficient, unpredictable, and time-consuming. Recently, machine learning (ML) technology has shown great potential in materials, which has provided new ideas and approaches to alloy design. In this work, a Reverse Machine Learning Design Model (RMLDM) has been created to design high-performance and low-cost Mg-Mn wrought Mg alloys. In RMLDM, five relatively inexpensive alloying elements and three conventional extrusion process parameters were selected as features to ensure the "low cost" of all designed alloys. The particle swarm optimization (PSO) algorithm was innovatively used to optimize the inputs of the artificial neural network (ANN), thus achieving the "reverse design" from "target performance" to "composition and process". Four alloys with higher performance were proposed through the RMLDM, which were determined to be close to the targets after experimental verification, and the best accuracy can reach 90%. The calculation errors demonstrate that the three ANN models' prediction accuracies are >94%. Furthermore, the RMLDM is generally a practical approach in developing new Mg alloys. The proposed reverse design strategy can be improved using additional data and easily applied to other alloys by changing the dataset.

A reverse design model for high-performance and low-cost magnesium alloys by machine learning