Application of a multivariate feature-driven ACO-BPNN integrated model in predicting ultimate tensile strength and composition optimization of Ni-based alloys

• Published in: Materials science & engineering. A, Structural materials : properties, microstructure and processing Vol. 943; p. 148817
• Main Authors: Zhou, Jianxing, Jian, Wei, Shen, Lei, Yi, Xi, Wang, Hainan, Chan, Ka Chun, Chan, Yuk Lun, Wang, Haozheng, Zhang, Qinhao, Suo, Xinkun, Ren, Lu
• Format: Journal Article
• Language: English
• Published: Elsevier B.V 01.10.2025
• Subjects: Alloy design; Backpropagation neural network; Feature analysis; Ni-based alloy; Ultimate tensile strength; Ultimate tensile strength; Backpropagation neural network; Feature analysis; Alloy design; Ni-based alloy
• ISSN: 0921-5093
• DOI: 10.1016/j.msea.2025.148817

A deep understanding of the complex nonlinear synergistic and antagonistic interactions among multiple features governing the ultimate tensile strength (UTS) of Ni-based alloys is key to achieving composition optimization. In this study, an innovative integrated model combining ant colony optimization and backpropagation neural network (ACO-BPNN) was developed by fusing elemental features with physical parameters. This model effectively overcomes the limitations of traditional machine learning approaches that rely on single-feature categories and fixed algorithms. It achieves high prediction accuracy with an R2 of 0.97 and strong extrapolation ability. By applying the Maximal Information Coefficient (MIC) method for iterative screening, the original 21 elemental features were reduced to 7 key features: Al, Ti, Mo, Cr, Co, C, and Ni. The SHAP method was employed to quantitatively evaluate the positive and negative contributions of each key feature to UTS and to determine their respective effective ranges. A novel Ni-17.6Cr-12.7Co-based alloy was designed and fabricated, achieving a UTS of 977 MPa and 10 % elongation with less than 3 % prediction error. The strength enhancement of the new alloy is mainly attributed to the formation of fine dendritic structures on the surface and the precipitation of carbides. This study presents a novel paradigm for the intelligent design of Ni-based alloys. •Novel ACO-BPNN framework with multi-feature fusion.•High-accuracy ultimate tensile strength prediction in Ni-based alloys.•Data-driven design and experimental validation of novel Ni-based casting alloys.