2025 Volume 66 Issue 1 Pages 144-150
“The 17th International Symposium on Novel and Nano Materials (ISNNM)” was held in Jeju, Korea from 14th to 18th November, 2022, and the proceedings for the session of “Integrated Computer-Aided Process Engineering (ICAPE)” were published in Feb, 2023, as a special issue of Materials Transactions (Vol. 64, No. 9). Following the first special issue, which covered the content of the ICAPE session at the International Symposium on Innovation in Materials Processing (ISIMP), this second special issue also presents various topics, including computational materials science, data-driven optimization, as well as experimental validation of optimized process. This article offers a concise overview of several key topics presented at the second special issue, including: macro-scale numerical analysis through finite element methods (FEM), and microstructure simulations using phase-field modelling (PFM), as well as various optimization methods such as machine learning (ML), artificial intelligence (AI), and design of experiments (DoE).
Optimizing materials and processes to achieve desired microstructures and properties through experimental trial-and-error is both time-consuming and labor-intensive. As the time required for migration of a process from the laboratory to the pilot plant or commercial scale needs to be shorten, there is a growing demand for new methodologies that can accelerate this optimization process. Despite extensive efforts over the years to clarify material behavior during manufacturing processes [1–21], the simulation of “microstructural evolution” under diverse processing conditions has often been overshadowed by continuum mechanics modeling, which focuses on describing the mechanical behavior of materials. This imbalance is primarily due to the intricate nature of microstructural evolution, involving numerous simultaneous physical phenomena such as heat transfer, fluid dynamics, phase transformations, and elastic/plastic deformations. Consequently, it is essential to integrate material simulation models across various length scales to accurately capture the process-structure-property (PSP) relationships during manufacturing, which include finite element methods (FEM) [22–25], phase-field modeling (PFM) [26–28] and molecular dynamics (MD) [29–33].
Along with scale-bridging simulation skills mentioned above, optimization techniques such as machine learning (ML) [34–36], artificial intelligence (AI) [37–40], and design of experiments (DoE) [41–43] has the potential to greatly speed up process development by swiftly analyzing and determining the best process conditions using computational and/or experimental data. By binding the extensive data produced from simulations and experiments, the parameters of materials and processes can be optimized effectively. This approach minimizes the reliance on expensive and time-consuming trial-and-error methods. Consequently, combining these advanced optimization techniques with state-of-the-art simulation methods can significantly improve the precision and efficiency of predicting material properties and refining process optimizations.
The Integrated Computer-Aided Process Engineering (ICAPE) session at the 17th International Symposium on Novel and Nano Materials (ISNNM), held in Jeju, Korea from 14th to 18th November, 2022, intended to unite specialists in scale-bridging simulations and characterization techniques in order to understand and predict the process-structure-property (PSP) relationships of newly developed materials at the laboratory and industrial process levels. Furthermore, the presentation on optimization techniques such as ML, AI, and DoE, which made up more than half of the session, highlighted the strong interest these methods have generated in this community. Among those, nine selected papers [44–50] were published in a special issue of Materials Transactions (Vol. 64, No. 9), and this article will review the three main topics covered in the session.
Computational materials simulations include various techniques, providing insights into materials behavior at different scales. FEM and crystal plasticity FEM (CPFEM) are commonly employed for macro-scale and meso-scale simulations, respectively, enabling detailed analysis of material responses under various thermal mechanical conditions. Phase Field Modeling (PFM) and Molecular Dynamics (MD) simulations offer fine-grained views of microstructural evolution and atomic interactions, respectively. Additionally, Density Functional Theory (DFT) provides a quantum mechanical perspective on material properties at the atomic level, while Discrete Dislocation Dynamics (DD) focuses on the behavior of dislocations within materials. Together, these computational techniques facilitate a comprehensive understanding of the behavior of materials, from fundamental atomic interactions to large-scale structural responses.
Cho et al. [46] employed FEM to investigate the martensite transformation and transformation-induced plasticity (TRIP) to predict the quench distortion of a cut-cylinder 4340 steel. The effect of the martensite transformation and TRIP on the quench distortion was effectively simulated as shown in Fig. 1, which demonstrates that TRIP has a very critical impact on the quench distortion of cut-cylinder 4340 steel. In addition, the chemical composition and quenching condition were confirmed to be important factors.
Simulation result of distorted shape after quenching, with initial shape overlapped [46].
Kahhal et al. [52] also employed the finite element analysis using a MATLAB code and an ABAQUS python script to generate observations for the neural network training. A neural network-based approach is proposed to minimize the maximum axial stress in the powder forming process. Powders of three different particle size distributions were mixed with a various mixing fractions. The genetic algorithm could effectively determine the optima and the proposed method had strong prediction capability and accuracy. As a result, Fig. 2 shows the predicted surface of axial stress at a relative density of 90% for the independent design variables. It provides insights into the relationship between these variables and the resulting axial stress.
ANN predicted surface of the axial stress objective based on the independent design variables [52]. (online color)
Combining the strengths of DoE and ML results in powerful tools for materials science. DoE facilitates cost-effective experiments by systematically designing them to investigate multiple factors affecting material properties simultaneously. ML leverages data to create models that can predict or classify outcomes. By integrating data collection through DoE with modeling through ML, researchers can achieve systematic experimental design and data-driven predictions simultaneously, leading to a deep understanding and optimization of material properties. This integrated approach is instrumental in enabling researchers to explore and develop complex material systems more effectively.
Jung et al. [44] employed a method using convolutional neural networks (CNNs) to classify cracks on plastic surfaces. The research team collected various images of plastic surface cracks and trained the CNN model using them. As shown in Fig. 3, the performance indices of the four selected models were compared for the training data, test data, and the entire data. It was demonstrated that the CNN model could classify cracks faster and more accurately than traditional methods.
Performance metrics for the four selected models using different data sets: (a) training data, (b) test data, and (c) total data [44].
Jeon et al. [45] developed a machine learning model to predict martensite start temperature (Ms) temperature using various alloy steel data sets. They trained the model using several algorithms such as random forest, support vector machine (SVM), and artificial neural network (ANN). The prediction mechanism and feature importance of the model were analyzed using the Shapley additive explanations (SHAP) technique. Figure 4 shows the results of (a) the average impact and (b) the influence of each variable on Ms using the SHAP technique. The prediction of the model showed that it has the potential to improve the performance and optimization of alloy steel.
SHAP analysis results of the ANN model, showing (a) the average impact of each alloying element and AGS on Ms, and (b) a scatter plot of variables affecting Ms [45].
Park et al. [47] attempted to improve the surface hydrophilicity of nonwoven fabrics by atmospheric pressure plasma treatment using the DoE method. The experimental variables included plasma treatment time, power, and gas flow. Figure 5(A) shows the standardized effect on the distribution fit line, and (B) is a Pareto chart with a baseline indicating whether the effect is statistically significant. In this way, the effect of each variable was systematically analyzed and optimized using the design of experiments methodology to maximize the efficiency of plasma treatment.
(A) Normal plot of the standardized effects under the initial condition, and (B) Pareto chart with a baseline for the effects [47].
In addition, Jeon et al. [48] employed random forest regression (RFR) model proposed to predict the bainite start temperature (Bs) of alloy steel. 97 data sets were trained and tested using RFR, an artificial neural network (ANN), and k-nearest neighbor (kNN) models. RFR model demonstrated a performance improvement of approximately 1.2% over the empirical equation. Alloying elements and an average grain size (AGS) were assigned importance, in that order, in the RFR using SHAP analysis. As shown in Fig. 6, SHAP analysis allowed us to specifically explain the prediction mechanism.
SHAP analysis results of RFR model, showing (a) the average impact of each alloying element and AGS on the Bs, and (b) a scatter plot of variables affecting Bs [48].
Validating the process/materials optimization conditions derived from previous theoretical or simulation results through actual experiments is an important step to confirm whether the theoretical optimization results are actually valid. This validation can secure the reliability of the theoretical predictions or simulation results and increase the likelihood of applying the verified optimization results in practice. If necessary, model improvements can be made, allowing the development of processes and materials to be performed quickly and effectively.
Seo et al. [49] studied the deformation and fracture behavior of heterostructures of STS316L (stainless steel) and Inconel718 (nickel alloy) manufactured using Laser Powder Bed Fusion (LPBF) technology. As shown in Fig. 7, the heterostructures of STS316L and Inconel718 manufactured using LPBF technology exhibited excellent mechanical properties due to the formation of a strong metallic bond at the interface between the two materials, thereby enhancing deformation and fracture resistance. The heterostructures manufactured under appropriate manufacturing conditions can combine the advantages of each material to increase the potential applications in the aerospace, energy, and marine industries.
(a) Inverse pole figure (IPF), (b) image quality (IQ), and (c) energy dispersive spectroscopy (EDS) maps of the STS316L and Inconel718 manufactured using LPBF technique [49]. (online color)
Joo et al. [50] studied the effect of Mg (magnesium) content on the precipitation hardening behavior in Al-Mg-Si-(Cu) alloys. Through the analysis of Al-Mg-Si-(Cu) alloys with various Mg contents, it was confirmed that the degree of precipitation hardening changed with increasing Mg content. As shown in Fig. 8, it was found that increasing Mg content enhanced the nucleation and growth rate of precipitation phases, leading to stronger hardening. In addition, it suggested the possibility of finding the optimal Mg content to maximize the performance of the alloy.
Relationship between predicted precipitate-phases and mechanical properties of (a) 0.3Mg, (b) 0.45Mg alloys [50]. (online color)
Kim et al. [51] studied the microstructure and mechanical properties of dissimilar joints obtained by friction welding of pure titanium (CP-Ti) and titanium alloy (Ti-6Al-4V). Figure 9 shows the hardness distribution of dissimilar titanium alloys friction-welded under various conditions, the relationship between the microhardness of the weld joint and the upset length on both sides. In conclusion, it was proven that friction welding is an effective method to form a joint showing excellent mechanical properties by combining the strengths of two materials through optimized welding conditions.
(a) Hardness distribution of dissimilar titanium alloys friction welded under various conditions, and the relationship between microhardness in the weld zone and upset length on both sides for (b) CP-Ti-2 and (c) Ti-6Al-4V alloys [51]. (online color)
This article briefly reviews the papers published in the special issue on ICAPE (Vol. 64, No. 9) in Materials Transactions. It provides an overview of key topics presented in this second special issue, including materials optimization techniques such as macroscopic numerical analysis using the finite element method (FEM), microstructure simulation using phase field modeling (PFM), as well as machine learning (ML), artificial intelligence (AI), and design of experiments (DoE). It was confirmed that the availability of data has increased. In conclusion, there is no doubt that computer-aided engineering of materials and processes holds great promise for advancing materials science and engineering and accelerating innovation across a wide range of industries.
This study was supported by the National Research Foundation of Korea (NRF) funded by the Ministry of Science and ICT (No. NRF-2020M3H4A3106736, and No. NRF-2021M3H4A6A01045764).
