Key Engineering Materials Vol. 1049

Abstract: Sheet metal components with complex geometries are typically recycled by remelting. Direct remanufacturing necessitates the flattening of parts, which requires the implementation of cuts to facilitate unwinding. The exact positioning of these cuts is a complex planning task, because several influencing factors can be considered, such as material usage, ease of flattening, or minimal forming required. This study presents a geometry-based concept addressing this challenge and demonstrates its use for a test geometry. The finite element method is applied to simulate the flattening process of the resulting sections, and the results are evaluated in terms of planarity and induced plastic strain. The findings of the present work indicate a discernible dependency of results on the selection of the flattening directions. In particular, curved areas impact the induced plastic deformation and springback of flattened sections. This is a crucial consideration when planarity is prioritised over material utilisation.

Abstract: This paper revisits the long-standing question of how to fully characterise the in-plane plastic anisotropy of sheet metals without assembling evidence from multiple standardised tests. The central idea is pragmatic: a single, well-designed heterogeneous biaxial experiment can replace the conventional combination of uniaxial and equibiaxial tests if the specimen and the inverse identification method are co-designed to (i) activate informative stress states and (ii) maintain low strain gradients for accurate digital image correlation measurements. The proposed cruciform specimen is deliberately conceived as a benchmark configuration for full-field inverse identification, with known locations and stress-strain states at which relevant material information is embedded. The approach is coupled with a Finite Element Model Updating framework, enabling all anisotropy parameters of the YLD2000-2d model to be identified from a single full-field dataset. Sensitivity and identifiability analyses demonstrate that a physically based parameter formulation significantly improves the conditioning of the inverse problem. Virtual experimentation confirms the robustness and accuracy of the proposed “one-test” identification strategy.

Abstract: Machine learning (ML) algorithms have been studied in literature as an inverse method to predict material constitutive parameters. However, these approaches are often dependent on the mesh discretisation settings applied during numerical simulations, and then difficulty model adaptation to experimental digital image correlation (DIC) subsets. Although a recent study explores the use of an interpolation-based approach to achieve experimental adaptation from numerically-based trained ML models, the proposed methodology lacks evaluation using experimental data. As a follow-up, this study proposes a new evaluation approach. Numerical data is DIC-levelled via MatchID software and then submitted to interpolation. An XGBoost algorithm is then trained on interpolated DIC data and evaluated for parameter prediction, comparing the obtained results with those obtained from the model trained on interpolated numerical data. Overall, the proposed DIC-levelling and interpolation pipeline yields an excellent predictive performance, with results comparable to those obtained when training on interpolated numerical data. The largest deviations are observed for the hardening exponent, while the remaining parameters are predicted with consistently high accuracy. These findings validate the practical applicability of the interpolation-based strategy to reduce the subset scheme dependency of ML models trained on real experimental data.

Abstract: In the framework of sustainable manufacturing and circular economy, the reuse of metallic components at the end of their first life (EoL) is a promising strategy to reduce energy consumption and material waste, but it requires an accurate assessment of residual plastic deformation, which strongly affects structural integrity and remaining formability. Conventional full-field techniques such as Digital Image Correlation (DIC) require a reference image of the undeformed state, which is generally unavailable for EoL components. To address this limitation, this work investigates a deep learning–based, reference-free approach for strain estimation directly from a single image of the deformed surface. The method relies on a convolutional neural network architecture derived from VGG-16, trained to regress the in-plane principal strains and their orientation from local image subsets of the surface texture using a synthetically generated dataset based on real material textures and realistic imaging conditions. The trained model is applied to high-resolution optical images of pre-deformed steel and aluminum components from regions subjected to different deformation histories, with partial validation provided by finite element simulations and conventional DIC measurements. Preliminary results show that the proposed approach can distinguish regions with different levels of plastic deformation and provide strain maps consistent with independent mechanical assessments, demonstrating its potential as a rapid, non-destructive tool for deformation mapping and classification of EoL components to support remanufacturing and reuse decisions.

Abstract: Inverse identification of material parameters from experimental data is a long-standing challenge, especially when calibrating complex constitutive models characterized by a large number of parameters. Heterogeneous mechanical tests combined with full-field measurements provide a large amount of information for material parameter identification but lead to high computational costs when used within a Finite Element Model Updating (FEMU) framework. This work presents an exploratory study on the use of surrogate-assisted Bayesian optimization to assess its potential for reducing the number of simulations required for FEMU-based calibration using data from a single notched tensile test. FEMU cost function is applied based on the discrepancy between experimental and numerical strain fields. A Gaussian Process surrogate model is iteratively constructed, and new sets of material parameters are selected using an Expected Improvement criterion. The results are discussed in terms of convergence behaviour and optimization efficiency, providing insight into the suitability of Bayesian optimization for solving inverse identification problems.

Abstract: This work deals with Robust design optimization (RDO) under interval uncertainty and the resolution of such problems using a Bayesian optimization algorithm. In metal forming, process parameters such as tool radius, step size, or forming toolpath introduce variability that directly affects the final geometry and quality ofthe formed parts. In this context we aim at finding a design minimizing the amplitude of the performance interval but such a formulation does not account for the nominal performance. In this work, we introduce a scalarized objective adapted to the proposed algorithm allowing it to identify a Pareto optimum of both stability and nominal behavior. We propose an efficient expected improvement (EI) estimator for this objective based on an extreme-value approximation of surrogate extrema. The approach is illustrated on an analytical test problem and on a forming simulation with spring-back, where the new objective yields more practically relevant solutions than a variation-only robustness criterion.