Fig. 1: The active learning workflow for optimizing the γ′ volume fraction, γ′ size, and γ′ morphology in CoNiAlCr-based superalloys based on triple-objective optimization.
From: The γ/γ′ microstructure in CoNiAlCr-based superalloys using triple-objective optimization
a The iterative workflow includes multi-property screening, triple-objective optimization, and experimental feedback (this image does not contain any third party material). b The performance of the ML regressor (γ′ solvus temperature, solidus, liquidus, density, γ′ volume fraction, γ′ size, and cuboidal γ′ fraction) and classifier (precipitate of detrimental phases) used for predicting corresponding property in testing data. c Illustration of triple-objective optimization algorithm. The equiprobability distribution with the maximum confidence interval (95%) of a three-dimensional Gaussian probability density function. The 1/8 ellipsoid surface is discretized (yellow points). The weight function varies with the three target values (y1, y2, and y3) in the range (0, 100). The right mapping axis is the weight function value of the corresponding target values.
