Simultaneous enhancement of the room temperature and high-temperature properties of Al-Cu-Mg-Ag alloys via machine learning on small datasets

Traditional alloy design based on trial-and-error approaches is often inefficient and costly. For specific alloy systems, the application of machine learning techniques is further challenged by limited dataset sizes and inconsistent high-temperature testing conditions. In this work, we proposed a machine learning-based strategy that incorporates prior normalization of high-temperature performance data. Key alloy factors were identified through a combination of correlation analysis, recursive elimination, and exhaustive selection. Bayesian optimization was subsequently employed to design alloy compositions with varying numbers of components with the aim of enhancing the mechanical performance at both room and elevated temperatures. Using the Al-Cu alloys as a model, the high-temperature tensile strength data were normalized to 300 °C. Through feature selection, six key alloy factors influencing high-temperature tensile strength, five key factors influencing room-temperature tensile strength, and six key factors influencing elongation were identified. Predictive machine learning models were then constructed, and Bayesian optimization was used to guide the design of a new alloy composition effectively. The resulting alloy, Al-5.8Cu-0.65Mg-0.4Ag-0.43Mn-0.28Ti-0.15Zr, demonstrated outstanding mechanical properties, with ultimate tensile strengths of 526 MPa at room temperature and 207 MPa at 300 °C. Microstructure characterization showed that the high thermal stability of T (Al₂₀Cu₂Mn₃) and Ω phases with high density, submicron and nano mixed distributions dominated the mechanical properties of the alloy over a wide temperature range. This study provides guidance for the efficient design of high-performance and wide-temperature adaptable Al-Cu alloys.