Materials informatics approach to understand aluminum alloys

Abstract

The relations between the mechanical properties, heat treatment, and compositions of elements in aluminum alloys are extracted by a materials informatics technique. In our strategy, a machine learning model is first trained by a prepared database to predict the properties of materials. The dependence of the predicted properties on explanatory variables, that is, the type of heat treatment and element composition, is searched using a Markov chain Monte Carlo method. From the dependencies, a factor to obtain the desired properties is investigated. Using targets of 5000, 6000, and 7000 series aluminum alloys, we extracted relations that are difficult to find via simple correlation analysis. Our method is also used to design an experimental plan to optimize the materials properties while promoting the understanding of target materials.

Keywords: 106 Metallic materials; 404 Materials informatics / Genomics; Markov chain Monte Carlo; Materials informatics; aluminum alloys.

Graphical abstract

Figure 1.

Flow of our strategy to extract the relations by combining a regression model and MCMC.

Figure 2.

Dependence of the mechanical properties of (a) 0.2% proof stress, (b) tensile strength, and (c) elongation on the temper designations X and n, and the compositions of nine types of elements in the 5000 series. Values of r denote the correlation coefficient. Correlation coefficients in red, blue, and black indicate positive, negative, and no relations, respectively. Each type of aluminum alloy is distinguished by the color of the points.

Figure 3.

Prediction results by machine learning models for the 0.2% proof stress, tensile strength, and elongation in the 5000 series aluminum alloys. These points are predictions for the test data when the leave-one-out cross validation is performed, that is, for the prediction of each point, target data is not included in the training of the machine learning model. Root mean square error (RMSE) for the test data by the leave-one out method is also denoted. As highlighted in red, the elastic net regression provides a relatively higher prediction accuracy for the three mechanical properties.

Figure 4.

Distributions of temper designations X and n and compositions of elements to obtain high (red) or low (blue) mechanical properties by MCMC sampling for the 5000 series aluminum alloys. Elastic net regression is used as a machine learning prediction model. Temper designations X and n have a discrete value, while others have continuous values.

Figure 5.

Distributions of temper designation X and compositions of elements to obtain high (red) or low (blue) mechanical properties by MCMC sampling in the 6000 series aluminum alloys. Random forest regression is used as a machine learning prediction model. Temper designation X has a discrete value, while others have continuous values.

Figure 6.

Distributions of temper designation X and compositions of elements to obtain high (red) or low (blue) mechanical properties by MCMC sampling in the 7000 series. Support vector regression is used as a machine learning prediction model. Temper designation X has a discrete value, while others have continuous values.