Jul 2026· SAE technical paper series· 0 citations· 11 references
Abstract
This study proposes a data-driven surrogate modeling framework for predicting
solidification time and mold thermal stress during low-pressure die casting
(LPDC) of aluminum alloy wheels. The methodology employed an optimal Latin
hypercube design (OLHD) to sample key parameters including cooling channel
geometry and process conditions. A sequential simulation methodology combining
ProCAST and Abaqus was implemented to generate a comprehensive dataset of
solidification times and thermal stress distributions. Based on this dataset,
surrogate models were developed using Support Vector Regression, Kriging, and
Polynomial Response Surface Methodology, with their hyperparameters
automatically tuned through Bayesian Optimization (BO). The optimized models
were rigorously evaluated using four statistical metrics: Coefficient of
Determination (R2), Mean Squared Error (MSE), Mean Absolute Error (MAE), and Root
Mean Squared Error (RMSE). The evaluation results show that the BO–SVR model
demonstrated superior prediction accuracy for both output responses and
exhibited exceptional nonlinear fitting capability. This work establishes an
effective modeling approach for simultaneous quality and efficiency optimization
in wheel manufacturing.
In low-pressure die casting (LPDC), there is a complicated relationship between the solidification behavior of the casting and the thermal stress of the mold. Conventional studies consider these two aspects separately, complicating global co-optimization. To address this, the present paper puts forward a methodology for the design of the process parameters and cooling system that is based on surrogate modelling and multi-objective inverse optimization. The LPDC of an aluminum alloy wheel hub was selected as a case study. Samples of process parameters and cooling structure parameters were obtained via optimal Latin hypercube design (OLHD). A thermo–mechanical coupled simulation was performed using ProCAST and Abaqus, resulting in a dataset for casting solidification time and maximum mold thermal stress. Within the framework of Bayesian optimization (BO), the predictive performance of three surrogate models, namely Support Vector Regression (SVR), Kriging, and Extreme Gradient Boosting (XGBoost), was compared. The study shows that BO-XGBoost exhibits lower predictive accuracy than the other two models, while BO-Kriging is marginally better for solidification time and BO-SVR is slightly superior for thermal stress. In addition, a comparative analysis was conducted on the performance of three multi-objective optimization algorithms, Non-dominated Sorting Genetic Algorithm II (NSGA-II), Multi-objective Particle Swarm Optimization (MOPSO), and Logistic Chaotic Mapping-based Sparrow Search Algorithm (LCSSA). The results show that LCSSA demonstrates optimal performance in terms of convergence, distribution, and stability. Consequently, it was selected to perform multi-objective optimization of solidification time and thermal stress. Combined with the surrogate models, an inverse optimization strategy was then employed to extract stable process parameter windows for various scenarios. Simulation verification demonstrates that the recommended parameter intervals satisfy target constraints. This study proposes a systematic methodology for the co-optimization of LPDC processes and mold design, enhancing the engineering applicability and stability of the process under variable working conditions.
Fan Fuhao, Yunlang Zhan, Zhenfei Zhan et al.· Proceedings of the Instituti...· 0 citations
In steel production plants, finite element (FE) simulations are widely used to develop plastic deformation processes such as rolling and forging. However, these simulations are large-scale and highly nonlinear, and therefore require long computation times, which makes systematic optimization of process conditions difficult. In this study, a POD-based non-intrusive reduced order model (NIROM), which is used here as a surrogate model, is applied to FE simulations of a sizing press in a hot rolling mill to address this challenge. The method combines Proper Orthogonal Decomposition (POD) with machine-learning-based regression to learn from existing simulation results and predict unknown process conditions. Gaussian process regression (GPR) is employed as the regression model. This approach accelerates computations while preserving the relevant physical characteristics of the solution. For slab displacement, POD-based NIROM with Gaussian process regression as developed in this study achieves a relative root mean square error (RMSE) of approximately 5 % across all nodes at each time step, although periodic error growth is observed during initial slab–die contact. Excluding the initial contact phase, relative RMSE values in all directions remain under 1 % relative to the characteristic lengths of the slab in the conveyance, thickness, and width directions. While one FE simulation requires about 20 hours on the target computational environment, the surrogate model produces predictions within a few seconds, resulting in a significant reduction in computation time. The results demonstrate that POD-based NIROM with Gaussian process regression is effective for predicting displacement even in simulations involving complex plastic deformation such as sizing press operations. Furthermore, the method provides sufficient accuracy and speed for process development in steel production, indicating its practical applicability. To the best of the authors’ knowledge, this is the first reported application of a POD-based NIROM to FE simulation models involving large-scale plastic deformation in industrial steel production processes.
Hiroto Goto, P. Eberhard· International Journal of Mat...· 0 citations
A novel, data-driven framework for predicting and optimizing the mechanical performance of 3D-printed polylactic acid (PLA) composites reinforced with date pit (DP) particles under controlled annealing conditions is presented, enabling simultaneous property prediction and design optimization from a minimal experimental dataset.
The residual stress is an important factor affecting the performance, durability and failure behaviour of High Velocity Oxy-Fuel (HVOF) sprayed based composite coating of carbon. Finite element analysis (FEA) and other conventional numerical methods are computationally demanding and constrained by assumptions, whereas machine learning (ML) models without physical interpretation can be trained entirely on less assumptions and purely using only data. This paper recommends a combined Physics-Informed Neural Network (PINN) model to predict HVOF thermal sprayed carbon-based composite coatings residual stress distributions. The model incorporates herein governing thermo-mechanical equations, such as heat transfer and elastic plastic deformation as a direct part of the neural network loss. Training and validation are done using experimental and simulated datasets. The suggested PINN model achieves a better accuracy (R2 > 0.96) and much lower cost of computation than traditional ANN and FEA-driven surrogate models. The model gives a generalizable, scalable method of predictive modelling of coating integrity and optimization of the parameters of HVOF process.
Ankit Tyagi, A. Dadhich, Sachin Sirohi et al.· Scientific Reports· 0 citations
Accurate prediction of the shear strength of reinforced concrete (RC) beams remains a challenging problem due to the complex nonlinear interactions among material properties, reinforcement characteristics, and beam geometry. This study presents a data-driven artificial neural network (ANN) framework for predicting the shear strength of RC beams using a systematically curated experimental database comprising 1,977 specimens collected from published literature. The database was preprocessed to remove incomplete and duplicate records, and the optimal ANN architecture was selected using the total goodness function. Model performance was evaluated using 10-fold cross-validation together with multiple statistical metrics, including the coefficient of determination (R²), mean absolute error, root mean squared error, bias, and prediction interval. The ANN achieved R² values of 0.9968 and 0.9686 for the representative training and testing datasets, respectively, and an overall R² of 0.993, with 1,668 predictions falling within the ± 30% error criterion. Comparative evaluation with the Canadian Standards Association (CSA), American Concrete Institute 318 (ACI 318), and Eurocode 2 (EC2) design-code models demonstrated that the proposed ANN consistently achieved superior predictive accuracy and reliability. Parametric analyses further confirmed that the predicted trends agree with established reinforced concrete shear mechanics, highlighting the dominant influence of the shear span-to-depth ratio, beam geometry, and reinforcement ratio on shear resistance. The proposed framework provides an accurate and robust decision-support tool that complements conventional design-code methods for predicting the shear strength of RC beams.
F. Khalid, Milad Razbin, S. Fareed et al.· Scientific Reports· 0 citations
This study presents a machine learning-driven framework for the constrained multi-criteria optimization of family-mold processes, where the simultaneous production of geometrically dissimilar parts creates complex flow imbalances. Dimensional deformation (warpage) and shear stress are identified as critical quality characteristics affecting the functional performance and structural integrity of molded products, while product weight is treated as a practical manufacturing constraint to ensure material efficiency. Preliminary experimental trials are conducted to verify process feasibility and to determine realistic operating ranges for key injection molding parameters. Based on these ranges, a face-centered central composite design (FCCCD) is constructed, and high-fidelity computational fluid dynamics (CFD) data are generated using Moldex3D analysis. A Kriging-based surrogate model is then developed to represent the complex and nonlinear relationships between process parameters and quality responses. The surrogate model is integrated with the Non-dominated Sorting Genetic Algorithm II (NSGA-II) to perform constrained multi-criteria optimization, where warpage and shear stress are simultaneously minimized under a prescribed weight constraint. The optimization process yields Pareto-optimal process parameter combinations that effectively balance dimensional accuracy and stress reduction while satisfying material usage requirements. Selected Pareto-optimal solutions are validated through detailed CFD simulations, yielding average relative errors of
1.56
%
and
3.58
%
for warpage and shear stress, respectively, demonstrating the predictive accuracy and robustness of the proposed framework. Overall, this study establishes an effective data-driven optimization strategy for quality-oriented design and decision support in intelligent family-mold processes.
Quoc-Nguyen Banh, Phat-Dat Truong, T. Nguyen et al.· Measurement and control (Lon...· 1 citation