What question did this study set out to answer?

The aim is to enhance the prediction of the warm deformation behavior of 5A06 aluminum alloy using machine learning.

July 12, 2026Materials0 citationsOpen Access

Enhancing Constitutive Description of 5A06 Aluminum Alloy During Warm Deformation Using Machine Learning-Assisted Johnson–Cook Model

LZLiu ZKing University LDLei DengHuazhong University of Science and Technology JLJinchuan LongCentral South University

Key Points

The aim is to enhance the prediction of the warm deformation behavior of 5A06 aluminum alloy using machine learning.
Developed two machine learning-assisted Johnson–Cook frameworks: PD-ANN-JC and MOI-ANN-JC.
Validated predictive accuracy using an artificial neural network surrogate model.
Integrated the workflow into finite element simulation software for practical applications.
MOI-ANN-JC framework achieved an average absolute relative error (AARE) of 1.424% and R2 of 0.997 on the testing set.
Validations resulted in AARE of 3.302% and R2 of 0.987 for the MOI-ANN-JC framework.
Established a direct correlation between thermal softening and strain hardening during deformation processes.

Abstract

To accurately characterize the warm deformation behavior and workability of the 5A06 aluminum alloy, this study presents an innovative workflow that develops and systematically validates machine learning-assisted Johnson–Cook (ML-JC) frameworks based on artificial neural network (ANN) surrogate models. Two predictive frameworks—the parallel-decoupled PD-ANN-JC and the multi-objective integrated MOI-ANN-JC—were constructed. Quantitatively, both developed ML-JC frameworks achieve significantly higher stress prediction accuracy and superior generalization capability compared with the conventional JC model. Specifically, on the testing set, the MOI-ANN-JC framework yields an average absolute relative error (AARE) of 1.424% and an R2 of 0.997, outperforming the PD-ANN-JC framework (AARE of 3.246%, R2 of 0.988). On the validation set, the MOI-ANN-JC framework also demonstrates exceptional generalization, with an AARE of 3.302% and an R2 of 0.987. Scientifically, the superior performance of the MOI-ANN-JC framework stems from its ANN-mnδ surrogate model, which simultaneously predicts the strain hardening exponent n, thermal softening exponent m, and relative error δ directly from deformation parameters. This mutual coupling establishes an intrinsic correlation between m and n, successfully aligning with the physical reality wherein strain hardening and thermal softening are inherently linked during deformation. Qualitatively and practically, by integrating the MOI-ANN-JC framework into finite element (FE) simulation software, dynamic tracking and visualization of the thermal softening exponent m during warm deformation were achieved. Combined with FE simulations, Vickers hardness testing and EBSD observations, this study successfully establishes a direct qualitative spatial correspondence between low-m regions and macroscopic defects, which was further verified through the warm forging of a thin-walled dual-cavity component. Crucially, this approach for evaluating deformation stability bridges the gap caused by the inapplicability of conventional processing maps within this temperature regime, offering a robust and broadly applicable workflow for complex forming optimization.

Ask AI

Helpful

Bookmark

View Full Paper