An intelligent design for Ni-based superalloy based on machine learning and multi-objective optimization
Yuedan Deng, Yu Zhang, Xiufang Gong, Hu Wang, Yucheng Wang, Ying Liu, Lixian Lian
Abstract
The simultaneous consideration of multiple conflicting properties in alloy optimization design is necessary yet remains challenging. A comprehensive strategy of machine learning (ML) and multi-objective optimization based on thermodynamic simulation data was proposed to accelerate the composition design of Ni-based superalloys. The microscopic parameters were determined by Pearson correlation analysis and domain knowledge as the key affect factors of tensile strength and elongation. The Multi-objectives Evolutionary Algorithm (MOEA) was adopted to search the well-built surrogate by ML meta-heuristically for the Pareto front of three objectives and its responding Pareto optimal solution set of composition. Furthermore, nine high-performance superalloy samples selected from the obtained Pareto front were well verified by fabricating and testing in the laboratory. Specially, a new composition among the nine as-fabricated samples was the best one according to the pre-defined design preferences with the γ′ solvus temperature, γ′ volume fraction, and TCP phases content approximating to 1210 °C, 65%, and 0.01%, respectively. This intelligent cooperation strategy based on ML and MOEA extends the methodology for multi-composition and multi-property design materials, which can optimize multiple conflicting objectives simultaneously rather than do one by one.