Article
Computer Science, Artificial Intelligence
Esther Andres-Perez, Carlos Paulete-Perianez
Summary: Computational fluid dynamics (CFD) simulations are widely used in aeronautical industries to analyze aerodynamic performance, with surrogate models being considered as a substitute for reducing time and cost. This paper reviews surrogate regression models for aerodynamic coefficient prediction and compares them using three different aeronautical configurations.
COMPLEX & INTELLIGENT SYSTEMS
(2021)
Article
Automation & Control Systems
Fatemeh Akhoni Pourhosseini, Kumars Ebrahimi, Mohammad Hosein Omid
Summary: Accurate monitoring of water quality is crucial in arid and semi-arid countries like Iran, with Total Dissolved Solids (TDS) playing a significant role. This study developed several hybrid models to predict TDS in Babolrood River, Iran, using monthly data from 1968 to 2016. The results showed that the SVM-TLBO5 model improved predictions compared to the LS-SVR1 model, with improved MAE and SI values at different stations.
ENGINEERING APPLICATIONS OF ARTIFICIAL INTELLIGENCE
(2023)
Article
Materials Science, Multidisciplinary
Rui Li, Mingzhou Jin, Vincent C. Paquit
Summary: This study introduced a Machine Learning-based scheme for detecting geometric defects in additively manufactured objects, using synthetic 3D point clouds for training and outperforming the existing Z-difference method. Bagging and Random Forest were identified as the best prediction models, suitable for various conditions of point cloud densities and defect sizes.
MATERIALS & DESIGN
(2021)
Article
Engineering, Civil
Hailong Cao, Xianjun Xie, Jianbo Shi, Yanxin Wang
Summary: Data-driven machine learning models have been used to predict levels of hazardous substances in groundwater. However, class-imbalanced data leads to low sensitivity in these models, despite high overall accuracy. To improve sensitivity, four algorithms were tested. The results showed that all four algorithms produced more accurate predictions with an average increase in sensitivity of 53.8%. ADASYN performed the best, increasing the model's G-means by over 40% on average. Furthermore, ADASYN-optimized models predicted higher groundwater exposure risk in Ghana compared to Ethiopia.
JOURNAL OF HYDROLOGY
(2022)
Article
Engineering, Electrical & Electronic
Jungsik Kim, Sun Jin Kim, Jin-Woo Han, M. Meyyappan
Summary: As FinFET scales aggressively, even a single point defect can cause performance variability. A machine learning algorithm is tested to replicate TCAD results, with the impact of a point defect in bulk FinFET used for validation. The trained model shows high accuracy test results, indicating potential for expediting failure analysis cycle through machine learning.
IEEE TRANSACTIONS ON DEVICE AND MATERIALS RELIABILITY
(2021)
Article
Computer Science, Artificial Intelligence
Meetesh Nevendra, Pradeep Singh
Summary: The study found that optimizing hyperparameters for learning techniques significantly improves defect count prediction performance, changes the ranking of classifiers, and grid search optimization generally outperforms random search optimization for default parameters.
EXPERT SYSTEMS WITH APPLICATIONS
(2022)
Article
Thermodynamics
Ante Sikirica, Luka Grbcic, Lado Kranjcevic
Summary: In this paper, microchannel designs with secondary channels and with ribs were investigated using computational fluid dynamics and were optimized using a multi-objective optimization algorithm. The proposed framework, which combines Latin hypercube sampling, machine learning-based surrogate modeling, and multi-objective optimization, showed promising results. The optimized solutions demonstrated improved performance with lower temperatures and reduced pressure drops compared to conventional microchannel designs. The proposed methodology can be used as an efficient approach for microchannel heat sink design optimization.
APPLIED THERMAL ENGINEERING
(2023)
Article
Environmental Sciences
Ozan Nadirgil
Summary: This paper develops and compares 48 hybrid machine learning models for accurate carbon price prediction. By using CEEMDAN, VMD, PE, and multiple types of ML models optimized by GA, the study shows that the CEEMDAN-VMD-BPNN-GA optimized double decomposition hybrid model outperforms the others with a striking R2 value of 0.993, RMSE of 0.0103, MAE of 0.0097, and MAPE of 1.61%.
JOURNAL OF ENVIRONMENTAL MANAGEMENT
(2023)
Article
Computer Science, Artificial Intelligence
Xiubin Zhu, Dan Wang, Witold Pedrycz, Zhiwu Li
Summary: Understanding the rationale behind machine learning predictions is crucial for building confidence and trust in intelligent systems. This study proposes a fuzzy local surrogate model to provide explanations for predictions and enhance interpretability of machine learning results. The model is composed of readable rules, making it highly interpretable for prediction interpretation. The proposed methodology offers a significant contribution to the interpretation of machine learning models and demonstrates high estimation accuracy in experimental studies.
IEEE TRANSACTIONS ON FUZZY SYSTEMS
(2023)
Article
Quantum Science & Technology
Alistair W. R. Smith, A. J. Paige, M. S. Kim
Summary: We propose a novel optimization strategy for small-to-intermediate scale variational quantum algorithms (VQAs) on noisy near-term quantum processors. This strategy utilizes a Gaussian process surrogate model equipped with a classically-evaluated quantum kernel. Our approach shifts the computational burden onto the classical optimizer component of hybrid algorithms, significantly reducing the number of quantum circuit evaluations required. We demonstrate that our approach achieves higher accuracy and convergence speed compared to classical kernels in noiseless and noisy VQE simulations.
QUANTUM SCIENCE AND TECHNOLOGY
(2023)
Article
Computer Science, Interdisciplinary Applications
Arsenii Uglov, Sergei Nikolaev, Sergei Belov, Daniil Padalitsa, Tatiana Greenkina, Marco San Biagio, Fabio Massimo Cacciatori
Summary: Injection molding is a popular method for manufacturing complex plastic objects. This study proposes a data processing pipeline with machine learning models to predict fill time and deflection distribution. The solution outperforms Moldflow in execution time and has been approved for use by automotive companies.
STRUCTURAL AND MULTIDISCIPLINARY OPTIMIZATION
(2022)
Article
Biochemical Research Methods
Talal Ahmed, Mark A. Carty, Stephane Wenric, Jonathan R. Dry, Ameen A. Salahudeen, Aly A. Khan, Eric Lefkofsky, Martin C. Stumpe, Raphael Pelossof
Summary: Reproducibility of results obtained using RNA data in cancer research remains challenging. Current RNA correction algorithms require access to patient-level data, but SpinAdapt computes corrections using aggregate statistics, preserving patient data privacy. SpinAdapt outperforms other methods on publicly available cancer studies and can correct new samples for unbiased evaluation.
BRIEFINGS IN BIOINFORMATICS
(2022)
Article
Water Resources
Yongen Lin, Dagang Wang, Yue Meng, Wei Sun, Jianxiu Qiu, Wei Shangguan, Jingheng Cai, Yeonjoo Kim, Yongjiu Dai
Summary: This study investigates the incorporation of bias learning components into data driven models for streamflow prediction. Experiments are conducted in the Andun river basin of China and 273 watersheds in the United States to validate the effectiveness of the mapping-bias-learning models. The results show that these models outperform mapping-learning-alone models and machine learning methods are superior to traditional statistical methods in terms of bias learning ability.
JOURNAL OF HYDROLOGY-REGIONAL STUDIES
(2023)
Article
Construction & Building Technology
Manish Kumar, Rahul Biswas, Divesh Ranjan Kumar, Pijush Samui, Mosbeh R. Kaloop, Mohamed Eldessouki
Summary: Traditional methods for predicting the compressive strength of concrete are difficult, while modern softcomputing models have emerged as a reliable solution for accurate forecasting.
CASE STUDIES IN CONSTRUCTION MATERIALS
(2023)
Article
Nuclear Science & Technology
Stefan Dasbach, Sven Wiesen
Summary: One major limitation for tokamak fusion reactors is the heat load on the divertor target. Machine learning models trained on simulation data can provide fast predictions for various scenarios in the design parameter space, and thus be used in system codes for rapid design studies of fusion power plants. This work aims to develop surrogate models for plasma exhaust using a preliminary dataset and train machine learning models to test their performance.
NUCLEAR MATERIALS AND ENERGY
(2023)
Article
Materials Science, Multidisciplinary
Yuji Sato, Thomas Swinburne, Shigenobu Ogata, David Rodney
Summary: The research demonstrates that anharmonic vibrational effects can impact the kinetics of thermally activated processes, leading to an unexpected decrease in the nucleation rate even at low temperatures.
MATERIALS RESEARCH LETTERS
(2021)
Article
Materials Science, Multidisciplinary
Thomas D. Swinburne
Summary: This review summarizes recent efforts towards quantifying the uncertainty of atomic trajectories in condensed phase systems. Bayesian methods are shown to measure sampling incompleteness rigorously, manage parallel simulations autonomously, and evaluate activation free energy with full treatment of anharmonic thermal vibrations. These freely available methods have been demonstrated on a wide range of challenging materials science problems.
COMPUTATIONAL MATERIALS SCIENCE
(2021)
Article
Materials Science, Multidisciplinary
Lauren T. W. Fey, Anne Marie Z. Tan, Thomas D. Swinburne, Danny Perez, Dallas R. Trinkle
Summary: This study used the Parallel Trajectory Splicing (ParSplice) method to simulate dislocation climb in nickel and focused on investigating the dominant mechanism for vacancy absorption by jogs. The results suggest that the main mechanism for vacancy absorption by jogs is biased diffusion to the dislocation core followed by fast pipe diffusion to the jog.
PHYSICAL REVIEW MATERIALS
(2021)
Article
Nuclear Science & Technology
L. Malerba, M. J. Caturla, E. Gaganidze, C. Kaden, M. J. Konstantinovic, P. Olsson, C. Robertson, D. Rodney, A. M. Ruiz-Moreno, M. Serrano, J. Aktaa, N. Anento, S. Austin, A. Bakaev, J. P. Balbuena, F. Bergner, F. Boioli, M. Boleininger, G. Bonny, N. Castin, J. B. J. Chapman, P. Chekhonin, M. Clozel, B. Devincre, L. Dupuy, G. Diego, S. L. Dudarev, C-C Fu, R. Gatti, L. Gelebart, B. Gomez-Ferrer, D. Goncalves, C. Guerrero, P. M. Gueye, P. Hahner, S. P. Hannula, Q. Hayat, M. Hernandez-Mayoral, J. Jagielski, N. Jennett, F. Jimenez, G. Kapoor, A. Kraych, T. Khvan, L. Kurpaska, A. Kuronen, N. Kvashin, O. Libera, P-W Ma, T. Manninen, M-C Marinica, S. Merino, E. Meslin, F. Mompiou, F. Mota, H. Namburi, C. J. Ortiz, C. Pareige, M. Prester, R. R. Rajakrishnan, M. Sauzay, A. Serra, I Simonovski, F. Soisson, P. Spatig, D. Tanguy, D. Terentyev, M. Trebala, M. Trochet, A. Ulbricht, M. Vallet, K. Vogel, T. Yalcinkaya, J. Zhao
Summary: The M4F project aims to study the effects of radiation on F/M steels, integrating various experimental and computational materials science tools to understand and simulate the complex phenomena of radiation-induced defect formation and evolution. The project focuses on local deformation and ion irradiation effects, developing models and best practices, and is entering its fourth year, close to delivering high-quality results.
NUCLEAR MATERIALS AND ENERGY
(2021)
Review
Materials Science, Multidisciplinary
M. R. Gilbert, K. Arakawa, Z. Bergstrom, M. J. Caturla, S. L. Dudarev, F. Gao, A. M. Goryaeva, S. Y. Hu, X. Hu, R. J. Kurtz, A. Litnovsky, J. Marian, M-C Marinica, E. Martinez, E. A. Marquis, D. R. Mason, B. N. Nguyen, P. Olsson, Y. Osetskiy, D. Senor, W. Setyawan, M. P. Short, T. Suzudo, J. R. Trelewicz, T. Tsuru, G. S. Was, B. D. Wirth, L. Yang, Y. Zhang, S. J. Zinkle
Summary: Predicting material performance in fusion reactor environments relies on computational modeling supported by experiments. Leading experts discussed current positions and ongoing challenges in modeling fusion materials, highlighting topics such as irradiation-induced defect production, gas behavior, clustering, and defect evolution.
JOURNAL OF NUCLEAR MATERIALS
(2021)
Article
Chemistry, Physical
Mouad Ramil, Caroline Boudier, Alexandra M. Goryaeva, Mihai-Cosmin Marinica, Jean-Bernard Maillet
Summary: Sampling the minimum energy path (MEP) between two minima of a system is often hindered by energy barriers. To overcome this limitation, augmented sampling methods based on collective variables or reaction coordinates can be used. By coupling statistical sampling techniques with autoencoders, it is possible to gradually learn the MEP and collective variables.
JOURNAL OF CHEMICAL THEORY AND COMPUTATION
(2022)
Article
Materials Science, Multidisciplinary
D. Da Fonseca, F. Onimus, F. Mompiou, M. -C. Marinica, E. de Sonis, E. Clouet, T. Jourdan
Summary: This study investigates the influence of elastic properties of point defects on dislocation climb under stress and irradiation. The elastic dipole tensors and diaelastic polarizabilities of vacancies and self-interstitial atoms in aluminum are evaluated using density functional theory calculations. These parameters are then incorporated into a Monte Carlo code and a diffusion model to estimate the stress dependence of dislocation climb. The results show that both parameters have an influence on point defect absorption under stress, with the dipole tensor effect only being 5 times larger than the polarizability effect. Additionally, considering polarizability is necessary for simulations under applied stress.
Article
Materials Science, Multidisciplinary
Petr Grigorev, Alexandra M. Goryaeva, Mihai-Cosmin Marinica, James R. Kermode, Thomas D. Swinburne
Summary: Calculations of dislocation-defect interactions are difficult due to the limitations of ab initio simulations. Hybrid methods, such as linear-in-descriptor machine learning potentials, offer a solution by embedding ab initio simulations in an empirical medium. This allows for more accurate modeling of dislocation migration pathways and defect geometries.
Article
Materials Science, Multidisciplinary
Clovis Lapointe, Thomas D. Swinburne, Laurent Proville, Charlotte S. Becquart, Normand Mousseau, Mihai-Cosmin Marinica
Summary: Machine learning surrogate models using atomic environment descriptors have wide applicability in materials science. This study investigates the accuracy and applicability limits of data driven surrogate models for vibrational properties. By increasing the dimension of the descriptor space and incorporating physical relevant information, the accuracy of the model is improved. The linear surrogate models are used to study the correlation between migration entropy and energy, providing new avenues for correlation analysis.
PHYSICAL REVIEW MATERIALS
(2022)
Article
Multidisciplinary Sciences
Alexandra M. Goryaeva, Christophe Domain, Alain Chartier, Alexandre Dezaphie, Thomas D. Swinburne, Kan Ma, Marie Loyer-Prost, Jerome Creuze, Mihai-Cosmin Marinica
Summary: It has been commonly believed that defects in face-centred cubic metals form larger dislocation loops through the coalescence of interstitial dumbbells. However, this study reveals that interstitial atoms in these metals actually cluster into compact 3D inclusions of A15 Frank-Kasper phase before forming dislocation loops. These A15 nano-phase inclusions then act as a source for prismatic or faulted dislocation loops. This discovery provides a better understanding of the complex mechanisms behind interstitial defect formation in metals.
NATURE COMMUNICATIONS
(2023)
Article
Materials Science, Multidisciplinary
Anruo Zhong, Clovis Lapointe, Alexandra M. Goryaeva, Jacopo Baima, Manuel Athenes, Mihai-Cosmin Marinica
Summary: The elastic properties of tungsten, an important material in future energy systems, are investigated up to its melting temperature using a data-driven approach. A machine learning force field is combined with enhanced sampling techniques to achieve accurate predictions of the material's behavior. A Bayesian sampling scheme is proposed to overcome the computational limitations of the machine learning force field, resulting in improved convergence speed and overall accuracy. The proposed method allows for the prediction of tungsten's elastic properties in temperature ranges that cannot be explored experimentally, opening up new possibilities for studying finite-temperature material properties.
PHYSICAL REVIEW MATERIALS
(2023)
Article
Materials Science, Multidisciplinary
Paul Lafourcade, Jean-Bernard Maillet, Christophe Denoual, Eleonore Duval, Arnaud Allera, Alexandra M. Goryaeva, Mihai-Cosmin Marinica
Summary: This study utilizes spectral descriptors to encode local atomic environments and build crystal structure classification models. The proposed simple classification model is effective in training with small databases and demonstrates inherent transferability. The method shows good accuracy in extreme conditions and is applied to Zirconium and Aluminum with specific cases.
COMPUTATIONAL MATERIALS SCIENCE
(2023)
Article
Materials Science, Multidisciplinary
Arnaud Allera, Alexandra M. Goryaeva, Paul Lafourcade, Jean-Bernard Maillet, Mihai-Cosmin Marinica
Summary: Accurate structural analysis is crucial for understanding atomic-scale processes in materials, but traditional methods often face limitations when applied to systems with thermal fluctuations or defect-induced distortions. To address this, the authors propose a novel descriptor for encoding atomic environments into 2D images, which enables accurate analysis using Convolutional Neural Networks at a low computational cost.
COMPUTATIONAL MATERIALS SCIENCE
(2024)
Article
Chemistry, Physical
Jacopo Baima, Alexandra M. Goryaeva, Thomas D. Swinburne, Jean-Bernard Maillet, Maylise Nastar, Mihai-Cosmin Marinica
Summary: This study applies deep learning techniques to free energy calculations in materials science, addressing the challenge of partitioning atomic configuration space by constructing appropriate collective variables. By using autoencoder neural networks and the adaptive biasing force method, the study successfully discovers reaction coordinates and performs free energy sampling in crystalline systems with localized defects simultaneously.
PHYSICAL CHEMISTRY CHEMICAL PHYSICS
(2022)
Article
Materials Science, Multidisciplinary
Daniel J. Antonio, Joel T. Weiss, Katherine S. Shanks, Jacob P. C. Ruff, Marcelo Jaime, Andres Saul, Thomas Swinburne, Myron Salamon, Keshav Shrestha, Barbara Lavina, Daniel Koury, Sol M. Gruner, David A. Andersson, Christopher R. Stanek, Tomasz Durakiewicz, James L. Smith, Zahirul Islam, Krzysztof Gofryk
Summary: Research on uranium dioxide crystals in the antiferromagnetic state under strong magnetic fields reveals the presence of piezomagnetism and magneto-elastic memory effect. The unexpected splitting of the [888] Bragg diffraction peak indicates the simultaneous existence of time-reversed magnetic domains and structural distortions. This study provides insights into the microscopic understanding of piezomagnetism and magnetic coupling in actinide materials.
COMMUNICATIONS MATERIALS
(2021)