Innovative data-driven design of titanium bimetal with superior strength-ductility synergy

To overcome the challenge of simultaneously achieving high-strength bonding and coordinated deformation in titanium bimetals, this study proposes an innovative data-driven framework integrating machine learning (ML) and genetic algorithm (GA). A design criterion was established based on (1) complementary room-temperature properties and (2) minimized high-temperature strength mismatch between the high-strength (HS) and high-ductility (HD) constituents. A comprehensive dataset of 139,822 samples was constructed, covering key alloying elements (V, Al, Cu, Mo, Cr, Fe, Nb, Sn, Zr, C, O, B). The multilayer perceptron (MLP) demonstrated superior predictive accuracy (R² = 0.9077, MSE = 0.0923) among four ML models. The MLP-GA coupling enabled the inverse design of optimal compositions and processing parameters. The fabricated HS/HD bimetal, produced via electron beam welding and hot forging, exhibited excellent deformation compatibility (evidenced by symmetrical “drum-shaped” morphology). After heat treatment (820 ℃/1 h/AC-510 ℃/6 h/AC), the interface achieved a bonding strength of 777.5 MPa while maintaining constituent properties, with the HS alloy reaching 1338.7 MPa UTS and the HD alloy showing 21.81 % elongation. This work establishes a reliable ML-GA co-design methodology for high-performance titanium bimetals, demonstrating a paradigm shift from trial-and-error to intelligent design in advanced materials development.