Key Engineering Materials Vol. 1050

Abstract: This work presents a numerical study of the friction surfacing process using a GPU-accelerated Smoothed Particle Hydrodynamics (SPH) framework previously validated against experimental observations. The model is employed to examine how thermal boundary conditions, rod diameter, and rod bending angle influence material deposition efficiency and the resulting deposit geometry. Variations in rod diameter are shown to influence both the thermal response and the contact pressure, with smaller rods producing higher efficiency but exhibiting greater process fluctuations. The findings highlight the critical roles of thermal management and geometric configuration in optimizing friction surfacing performance and provide actionable insight for experimental design and process control in solid-state deposition technologies.

Abstract: Previous research shows that predicting width deviation inherits a central importance in hot rolling processes, so that the pass planning in the hot strip mill (HSM) can be optimized. These predictions can be enabled using machine learning, complementing analytical formulations of width spread. For reliable production, it is important for the plant operator to be able to control the geometry with high accuracy across the entire plant. Therefore, the width must be accurately known throughout the entire HSM. This paper aims on the prediction of width deviation in early product stages during the roughing mill processes, where the major deformation takes place, and thus also has the most significant influence on the width spread. Therefore, this paper takes industrial data into account which is also used for the roll passing planning. To achieve a prediction during rough rolling for the width after exiting the mill (future state of the strip width), various machine learning algorithms were implemented and tested. The prediction results are evaluated against an inline width measurement, where the XGB model performs best with a Root Mean Squared Error (RMSE) of 1.11 mm. Subsequently, feature importance analyses are used to examine which features are relevant for the prediction result and to elaborate which significance process-and geometry data has on the same strip.

Abstract: Gaining a better understanding of the structure-property relationship in materials is a vital step in optimizing forming processes in order to minimize the induced damage and thereby maximizing the materials’ performance.Dual phase (DP) steels are comprised out of hard martensite surrounded by a soft and ductile ferrite matrix. Due to the complex microstructure of DP-steels, different mechanisms of damage initiation can occur, such as martensite cracking or ferrite-martensite phase boundary decohesion. A key problem with computational microstructure optimization focusing on one specific damage mechanism is, that this can lead to virtual microstructures, which are good against one mechanism, but vice-versa problematic for another mechanism. This is why all optimization strategies have to consider more than one mechanism. In this study, a multi-objective Bayesian optimization (moBo) approach is developed for the design of damage-tolerant DP-microstructures. It combines full-field crystal plasticity simulations on 3D representative volume elements with computational optimization. By employing the moBo, the sets of microstructure parameters are determined, where the combined minimum of both damage indicators is located. The proposed algorithm was applied to identify pareto-optimal microstructure configuration for DP800, considering both prevalent damage mechanisms It also provides an estimate of the variance associated with each parameter, which defines how critical the correct regulation of that aspect is. The results are in line with prevailing knowledge about DP steel, thus showing that the proposed approach is a promising tool for computational microstructure design

Abstract: Stamp forming of fiber-reinforced thermoplastic composite materials is governed by large deformations, anisotropic and rate-dependent material behavior, and frictional multi-body contact, making high-fidelity finite element simulations expensive and often impractical for rapid design studies and process optimization. We leverage recent advancements in Machine Learning-based simulations and tailor Algebraic-hierarchical Message Passing Networks (AMPNs) to stamp forming simulation of composite materials. To efficiently handle multi body contact during forming, we model the laminates by a multi-layer graph with explicit ply–ply and tool–ply contact and extend AMPNs by local component-wise contact edges. Using a multiscale graph hierarchy, the method captures local wrinkling effects, global material draw-in, and contact-driven deformation across the full laminate. Trained on high-fidelity data from state-of-the-art Finite Element Method (FEM) simulations, the surrogate accurately simulates the stamp forming process for unseen process settings, while reducing simulation times from hours to seconds, enabling approximately real time simulation of large, complex geometries.

Abstract: When it comes to predicting quality-relevant outcomes of rotary draw bending procedures, like springback and geometric errors, machine learning algorithms have demonstrated encouraging results. However, the challenges associated with understanding these models’ predictions still restrict their actual application in industrial contexts. A web-based 3D visualization designed to help with interactive exploration and explainability of machine learning predictions in rotary draw bending is evaluated through an expert-centered study. Based on an earlier Random Forest regression model, the visualization lets users change important process parameters and view the projected tube geometry and springback in real time. Sixteen experts participated in a structured online survey that combined openended comments, subjective agreement scores, and interactive parameter modification tasks. Results show that while multi-objective optimization remained difficult, participants with different degrees of machine learning knowledge and tube-bending experience were generally able to identify appropriate parameter settings in single-objective problems. Subjective assessment and qualitative feedback from the participants also highlight that the visualization could be used to assist in understanding model behavior and in early process design and training situations. Overall, our study suggests that experts in tube bending applications find benefit from the interactive 3D visualization of the predicted geometry and as a useful interface for exploring machine learning models’ predictions.

Abstract: Understanding and predicting static recrystallization (SRX) behavior is crucial for controlling the microstructure and mechanical properties of metals during thermomechanical processing. Among various numerical modelling approaches that can be used to support experimental studies on this topic is the cellular automata (CA) method. This approach gained significant attention due to its ability to simulate microstructural evolution at the mesoscale with high spatial resolution. However, the main limitation of CA models is their significant simulation time, especially for the 3D computational domains. Therefore, the paper focuses on enhancing the efficiency of CA SRX simulations to deliver results within an acceptable time frame. The goal is to minimize computation time and memory usage through code-level optimization, without altering the hardware or compiler settings. Optimization is performed on the sequential version of the validated CA SRX code. Initially, the source code was analyzed using a profiler tool to identify performance bottlenecks. The most inefficient parts of the code were then reimplemented to eliminate these bottlenecks. Optimization methods included eliminating redundant functions, modifying neighbor assignments in the automata space, reducing class data structures, enabling direct access to attributes, simplifying mathematical formulas, and removing unused objects. The obtained results are also validated against the output from the sequential version to ensure the model's predictive capabilities. The work clearly demonstrates that the optimization improved simulation efficiency across all tested variants, with only minor increases in memory usage.

Abstract: Bobbin-Tool Friction Stir Welding (BT-FSW) is a solid-state joining process in which axial forces are internally balanced by the tool, eliminating the need for a backing plate and enabling the joining of hollow aerospace structures. Owing to the coupled thermo-mechanical nature of the process, weld stability and quality are governed by the interaction between process parameters and the resulting torque response, which is difficult to assess in situ using conventional sensing alone. BT-FSW experiments were performed on AA2024-T351 sheets with thicknesses of 2.4, 2.8 and 3.6 mm using a structured Design of Experiments (DoE). The 3.6 mm joints achieved approximately 90 % of the base-material strength, while the 2.4 and 2.8 mm joints reached about 80 % and 85 %, respectively. These mechanical results were used as ground truth to train machine learning regression models for steady-state torque predictions. By augmenting nominal process parameters with force-derived features, the proposed soft-sensing framework achieved strong agreement between predicted and measured torque, demonstrating that compact, physics-based feature engineering enables reliable prediction under limited experimental data conditions.

Abstract: Accurate prediction of quality-relevant material parameters, such as thickness and grain size, ensures product quality in hot forming processes. This task becomes especially complex in hot rolling, where the sequential and time-dependent nature of the process results in pass schedules with varying numbers of passes and grain size evolution that depends on the deformation history. To address this complexity, this study aims to develop and train deep learning models based on Long Short-Term Memory (LSTM) networks, which are well-suited for modelling sequential data. As input features for the model, real-time process parameters such as rolling force and rolling temperature are used, which can be captured by sensors during operation. Simulation data for both material and process parameters are acquired using Simulation as a Service (SaaS) through the Fast Rolling Model (FRM) called Rolling Calculation Tool (RoCaT), focusing on steel grade S355. The performance of the LSTM model is evaluated by analysing loss curves over training epochs and comparing predicted values to reference data. The maximum relative percentage error for thickness between the LSTM predicted value and the RoCaT value is 18.818% and 16.56259%, respectively, for pass schedules of 15 and 17 passes, with a starting thickness of 205mm. The percentage of relative error values for grain size is more pronounced during the initial passes as compared to later passes for both pass schedules. The statistical validation is performed on the denormalized data using metrics such as Root Mean Squared Error (RMSE), Mean Absolute Error (MAE), and R-Squared (R²), which demonstrate the model’s ability to predict key material properties in the hot rolling process reliably. The RMSE, MAE, and R² values for thickness are obtained as 45.2175mm, 18.7555mm, and 0.8115, respectively. For grain size, the corresponding values are 25.8287µm, 13.4319µm, and 0.9192.

Abstract: In the early concept phase of vehicle development, crash simulations must provide reliable statements on energy absorption and failure behavior, although detailed forming simulations and process data are typically not yet available. Manufacturing-induced material states, such as plastic pre-strains and local sheet thickness distributions, are therefore often neglected or approximated using simplified low-fidelity approaches with high uncertainty. This contribution introduces UmMatCraML, a data-driven medium-fidelity method for rapid initialization of typical crash shell meshes (element edge lengths 3–5 mm) with forming-induced field variables. Starting from the final component geometry, a purely geometric unfolding is performed to approximate the blank, from which a mesh-and coordinate-independent areal strain (A_r) is determined. A monotonicity-preserving gradient boosting regressor subsequently compensates for systematic deviations of this geometric surrogate model compared to high-fidelity deep drawing simulations, with four Hockett-Sherby parameters consistently parameterizing the material description. The training data are generated from approximately 5,000 LS-DYNA forming simulations and cover a broadly varied, physically consistent parameter space. In validation on a demonstrator part, UmMatCraML reduces the computation time for determining forming-induced component properties from about 60 min for High-Fidelity-Simualtion (HFS) or 10–15 min for Low-Fidelity-Simulation (LFS) to under 10 s, with simultaneously improved prediction quality. Demonstrations on components of a Toyota Yaris full-vehicle model show robust predictions even with trimming and perforations. Limitations arise from the model assumptions made (e.g., isotropic hardening, limited mapping of multi-stage process paths). Overall, UmMatCraML enables real-time, reproducible provision of manufacturing-induced field variables for concept crash simulations without explicit tool modeling.