Resolving the strength-ductility trade-off in Al-Mg-Si alloys via physics-informed machine learning

Abstract

Composition-based machine learning for Al-Mg-Si alloys lacks generalization. A physics-informed strategy incorporating physicochemical descriptors was proposed, improving the coefficient of determination (R2) by 30% over composition-only models. Shapley additive explanations (SHAP) identified electronegativity difference and average shear modulus mismatch as dominant descriptors, providing a mechanistic basis for optimizing the concentrations of Mg-Si-Cu for precipitates and Mn-Cr-Zr for dispersoid formation. Guided by non-dominated sorting genetic algorithm (NSGA-II), a novel alloy achieved an ultimate tensile strength of 442.2 MPa with 14.6% elongation. Microstructural analysis revealed that a uniform grain structure and high-density β″, Q′, and L co-precipitates resolved the strength-ductility trade-off. GRAPHICAL ABSTRACTSix diagrams and charts outlining data collection, model construction, feature analysis, alloy design, extrusion process, and tensile results.The figure shows a workflow for designing an aluminum alloy using data driven analysis and experiments. On the left, a schematic diagram of stacked documents feeds into a table labeled as physics informed descriptors listing several symbols in rows. Below, a radar chart compares model performance metrics for descriptors and compositions, with multiple axes labeled with variations of coefficient of determination and mean absolute percentage error for training, cross validation, and test sets. In the center upper section, a feature analysis diagram includes horizontal bars labeled as Shapley additive explanation analysis and a list of top three descriptors for ultimate tensile strength and elongation. To the right of this list, two small scatter plots show Shapley values versus 2 different descriptor variables, each displaying curved point clouds across narrow value ranges. In the center lower section, a scatter plot labeled alloy design displays ultimate tensile strength in megapascal on the vertical axis from 280 to 480 with step labels of 20, and elongation in percent on the horizontal axis from about 14 to 18. Filled circles show many candidate design alloys, with a single highlighted point representing a selected composition noted below the chart as an aluminum magnesium silicon manganese copper chromium zirconium iron alloy. On the right upper section, a process diagram plots temperature against time for homogenization, extrusion, aging, and tensile testing, with stepped temperature profiles labeled 550 degrees Celsius for 12 hours, 480 degrees Celsius, extrusion at 1 degree Celsius per minute, room temperature water spray, and aging at 165 degrees Celsius for 24 hours. Next to this, a small tensile specimen icon is pulled vertically by arrows. In the right middle section, a perspective illustration of an extrusion die shows a cylindrical billet labeled diameter 178 millimeters and the resulting extruded profile emerging. At the bottom right, a scatter plot compares ultimate tensile strength in megapascal on the vertical axis from 300 to 450, with elongation in percent on the horizontal axis from 10 to 18. Multiple symbol sets correspond to different copper weight percent ranges, with one group labeled as this work indicating the performance of the newly designed alloy. All data are approximate.