Machine-learning-assisted prediction of the mechanical properties of Cu–Al alloy

The machine-learning approach was investigated to predict the mechanical properties of Cu–Al alloys manufactured using the powder metallurgy technique to increase the rate of fabrication and characterization of new materials and provide physical insights into their properties.Six algorithms were used to construct the prediction models, with chemical composition and porosity of the compacts chosen as the descriptors.The results show that the sequential minimal optimization algorithm for support vector regression with a puk kernel(SMOreg/puk) model demonstrated the best prediction ability. Specifically, its predictions exhibited the highest correlation coefficient and lowest error among the predictions of the six models. The SMOreg/puk model was subsequently applied to predict the tensile strength and hardness of Cu–Al alloys and provide guidance for composition design to achieve the expected values. With the guidance of the SMOreg/puk model, Cu–12Al–6Ni alloy with a tensile strength(390 MPa) and hardness(HB 139) that reached the expected values was developed.

A novel approach to predict green density by high-velocity compaction based on the materials informatics method

High-velocity compaction is an advanced compaction technique to obtain high-density compacts at a compaction velocity of ≤10 m/s. It was applied to various metallic powders and was verified to achieve a density greater than 7.5 g/cm^3 for the Fe-based powders. The ability to rapidly and accurately predict the green density of compacts is important, especially as an alternative to costly and time-consuming materials design by trial and error. In this paper, we propose a machine-learning approach based on materials informatics to predict the green density of compacts using relevant material descriptors, including chemical composition, powder properties, and compaction energy. We investigated four models using an experimental dataset for appropriate model selection and found the multilayer perceptron model worked well, providing distinguished prediction performance, with a high correlation coefficient and low error values. Applying this model, we predicted the green density of nine materials on the basis of specific processing parameters. The predicted green density agreed very well with the experimental results for each material, with an inaccuracy less than 2%. The prediction accuracy of the developed method was thus confirmed by comparison with experimental results.