Article
Meteorology & Atmospheric Sciences
Jan Ackmann, Peter D. Dueben, Tim Palmer, Piotr K. Smolarkiewicz
Summary: This paper investigates the potential computational savings of using mixed precision arithmetic in elliptic solvers for atmosphere and ocean models. The study shows that for certain key components, using half precision can achieve a speed-up of 4 times compared to double precision.
JOURNAL OF ADVANCES IN MODELING EARTH SYSTEMS
(2022)
Article
Computer Science, Interdisciplinary Applications
Michael Weiss, Maya Neytcheva, Thomas Kalscheuer
Summary: We develop an efficient and robust iterative framework for solving linear system of equations in geophysical controlled-source electromagnetic applications, and demonstrate its robustness and computational advantage through numerical experiments.
COMPUTATIONAL GEOSCIENCES
(2023)
Article
Engineering, Multidisciplinary
Rihui Lan, Wei Leng, Zhu Wang, Lili Ju, Max Gunzburger
Summary: This study investigates the parallel performance of two ocean models using exponential time differencing (ETD) methods, which show potential for improving computational efficiency in numerical simulations. Benchmark tests demonstrate the effectiveness of ETD methods for simulating real-world geophysical flows in ocean modeling.
COMPUTER METHODS IN APPLIED MECHANICS AND ENGINEERING
(2021)
Article
Computer Science, Interdisciplinary Applications
Ki-Ha Kim, Ji-Hoon Kang, Xiaomin Pan, Jung-Il Choi
Summary: This study aims to solve many tridiagonal systems in multi-dimensional partial differential equations efficiently and with reduced communication overhead. A modified Thomas algorithm and a newly designed communication scheme were used for this purpose. Benchmark tests showed significant reduction in communication time compared to global all-to-all communication methods, especially for larger problem sizes and more cores. The computational procedures were implemented in an open-source library called PaScaL_TDMA, and demonstrated good scalability on a cluster system.
COMPUTER PHYSICS COMMUNICATIONS
(2021)
Article
Computer Science, Theory & Methods
Federico Ciccozzi, Lorenzo Addazi, Sara Abbaspour Asadollah, Bjorn Lisper, Abu Naser Masud, Saad Mubeen
Summary: This article provides a comprehensive overview of research on languages for parallel computing in software-intensive systems. By analyzing 225 studies, it identifies trends, technical characteristics, challenges, and research directions. The systematic review is valuable for researchers and practitioners in this field.
ACM COMPUTING SURVEYS
(2023)
Article
Geography, Physical
H. Nienhaus, P. Yogeshwar, W. Morbe, B. Tezkan, C. Buettner, M. Legler, S. Buske, B. Lushetile, V. Wennrich, M. Melles
Summary: This study investigates the sedimentary record in the Aurus clay pan in the southern Namib Desert for long-term paleoclimate research. Geophysical surveys and sedimentological investigations were conducted to understand the subsurface structure and sediment composition. The findings suggest a relatively humid Holocene period and a drier late Pleistocene period. Compared to similar studies in the Atacama Desert, the sediment record in Aurus is longer but provides limited information on sediment composition.
GLOBAL AND PLANETARY CHANGE
(2023)
Article
Computer Science, Information Systems
Arturo Fernandez
Summary: IaaS from the public cloud is emerging as a new option for organizations seeking HPC capabilities. It offers greater flexibility and cost-effectiveness, enabling even small organizations to access resources previously limited to larger ones. This article evaluates the performance of IaaS from five cloud vendors and two architectures using benchmark tests. The results demonstrate the benefits of higher network bandwidth when scaling up clusters and show promising scalability compared to on-premises supercomputers.
IEEE TRANSACTIONS ON CLOUD COMPUTING
(2022)
Article
Computer Science, Software Engineering
Arne Rempke
Summary: Line-implicit preconditioners, commonly used in computational fluid dynamics solvers, have been shown to have advantages beyond just solving partial differential equations for anisotropic cells. They can also be applied to problems like node-based mesh deformation with linear elasticity. A new algorithm for identifying lines for line-implicit preconditioners is presented, which improves parallel processing and leads to more homogeneous lines. Using this new algorithm, faster convergence is achieved for the mesh deformation problem based on linear elasticity.
BIT NUMERICAL MATHEMATICS
(2023)
Article
Engineering, Electrical & Electronic
Kasia Swirydowicz, Nicholson Koukpaizan, Tobias Ribizel, Fritz Goebel, Shrirang Abhyankar, Hartwig Anzt, Slaven Peles
Summary: This paper investigates the economic dispatch challenges of integrating renewable resources into the transmission grid and proposes a sparse linear solver that accelerates computation using GPUs. The authors treat the problems as sparse ones and demonstrate significant performance improvements by executing the entire computation on GPU-based hardware. They also identify research opportunities for better utilization of heterogeneous systems.
INTERNATIONAL JOURNAL OF ELECTRICAL POWER & ENERGY SYSTEMS
(2024)
Article
Computer Science, Theory & Methods
Massimo Bernaschi, Alessandro Celestini, Flavio Vella, Pasqua D'Ambra
Summary: We present a sparse linear solver that efficiently utilizes heterogeneous parallel computers and can be easily integrated into scientific applications on modern parallel computers with Nvidia GPU accelerators. Based on the hybrid MPI-CUDA software environment, our solver extends previous efforts in using a single GPU accelerator and proposes a Krylov-type linear solver with an efficient AMG preconditioner. The hybrid implementation minimizes data communication overhead when multiple GPUs are employed, while preserving the efficiency of GPU kernels. Results show a speedup of up to 2.0x compared to Nvidia AmgX solution in the solve phase.
IEEE TRANSACTIONS ON PARALLEL AND DISTRIBUTED SYSTEMS
(2023)
Article
Computer Science, Hardware & Architecture
Oleg Bessonov, Sofiane Meradji
Summary: This paper focuses on developing efficient parallel methods for the FireStar3D wildfire modeling code, with discussions on the MILU-preconditioned conjugate gradient method and algebraic multigrid. Solutions for parallelizing these methods are presented, with a novel quasi-geometric interpolation technique introduced. Performance comparison demonstrates the superiority of the new methods over traditional conjugate gradient methods.
JOURNAL OF SUPERCOMPUTING
(2021)
Article
Chemistry, Multidisciplinary
Yongmeng Qi, Qiang Li, Zhigang Zhao, Jiahua Zhang, Lingyun Gao, Wu Yuan, Zhonghua Lu, Ningming Nie, Xiaomin Shang, Shunan Tao
Summary: This study uses high-performance computing to model the behavior of floods, specifically using the two-dimensional Saint-Venant equations as an example. The researchers applied large-scale parallel computing and various optimization techniques, finding the method to be efficient and recommending its use in solving similar problems.
APPLIED SCIENCES-BASEL
(2022)
Article
Computer Science, Hardware & Architecture
A. J. Munoz-Montoro, D. Suarez-Dou, J. J. Carabias-Orti, F. J. Canadas-Quesada, J. Ranilla
Summary: This paper introduces a parallel low-latency multichannel source separation system suitable for applications such as interactive live broadcast classical music, achieving real-time processing in tested scenarios through a combination of multi-core architectures and high-performance parallel techniques.
JOURNAL OF SUPERCOMPUTING
(2021)
Article
Computer Science, Interdisciplinary Applications
Raphael Egan, Arthur Guittet, Fernando Temprano-Coleto, Tobin Isaac, Francois J. Peaudecerf, Julien R. Landel, Paolo Luzzatto-Fegiz, Carsten Burstedde, Frederic Gibou
Summary: The study proposes a parallel approach for solving the Navier-Stokes equations on Octree grids, demonstrating strong scalability and dynamic adaptive capabilities through additional parallel algorithms and performance analyses.
JOURNAL OF COMPUTATIONAL PHYSICS
(2021)
Article
Computer Science, Interdisciplinary Applications
Yujie Wang, Shengquan Wang, Xuerui Zhang, Guangyao Li, Yong Cai
Summary: This paper introduces a direct linear solver based on heterogeneous hybrid parallel computing, which can efficiently utilize computing resources of multiple devices to achieve performance improvement. By task decomposition and numerical decomposition strategies, parallelism of nodes and trees can be achieved on CPUs, and efficient numerical decomposition can be achieved on GPUs through batch processing and maximizing the overlap between computations and data transfers. Numerical experiments show that compared to MKL PARDISO, the performance of numerical factorization can be improved by up to 10 times by using CPU and dual-path GPU hybrid calculations, and the computation time for multi-condition analysis of Body In White can be reduced by one-third, and for large-scale nonlinear finite element deformation analysis can be reduced by 20%.
COMPUTER PHYSICS COMMUNICATIONS
(2023)
Article
Engineering, Mechanical
Diab W. Abueidda, Seid Koric, Nahil A. Sobh, Huseyin Sehitoglu
Summary: This study applied sequence learning models to predict the history-dependent responses of materials, showing that gated recurrent unit and temporal convolutional network can accurately learn and instantly predict such phenomena, with TCN being more computationally efficient during the training process.
INTERNATIONAL JOURNAL OF PLASTICITY
(2021)
Article
Computer Science, Interdisciplinary Applications
Fereshteh A. Sabet, Seid Koric, Ashraf Idkaidek, Iwona Jasiuk
Summary: This study compared implicit and explicit methods in investigating the mechanical properties of trabecular bone using finite element analysis. The results indicated that the two methods gave comparable results, with the explicit method performing faster and consuming less memory.
COMPUTER METHODS AND PROGRAMS IN BIOMEDICINE
(2021)
Article
Materials Science, Multidisciplinary
Seid Koric, Diab W. Abueidda
Summary: This study utilizes advanced numerical modeling techniques and deep learning methods to accurately capture and predict the nonlinear thermo-mechanical behavior of solidifying steel, even in unseen test data samples.
Article
Engineering, Multidisciplinary
Diab W. Abueidda, Qiyue Lu, Seid Koric
Summary: Deep learning and the collocation method are merged to solve partial differential equations describing structures' deformation, offering a meshfree approach that avoids spatial discretization and data generation bottlenecks.
INTERNATIONAL JOURNAL FOR NUMERICAL METHODS IN ENGINEERING
(2021)
Article
Computer Science, Interdisciplinary Applications
T. Scharf, C. L. Kirkland, M. L. Daggitt, M. Barham, V. Puzyrev
Summary: This study presents an open-source graphical application called AnalyZr for analyzing zircon grain shape. The application utilizes a new segmentation algorithm to improve the separation of touching zircon grains. Two case studies demonstrate the application of AnalyZr in resolving geologically relevant information in zircon grains.
COMPUTERS & GEOSCIENCES
(2022)
Article
Engineering, Multidisciplinary
Mehdi Ghommem, Vladimir Puzyrev, Rana Sabouni, Fehmi Najar
Summary: Gas sensors have been increasingly used for various applications, and this study proposes a novel MEMS gas sensor that utilizes mechanically-coupled microbeams and metal organic frameworks to detect the presence and estimate the concentrations of carbon dioxide and methane through deep learning methods. The results show high prediction accuracy and outperform classical statistical approaches.
APPLIED MATHEMATICAL MODELLING
(2022)
Article
Geochemistry & Geophysics
Vladimir Puzyrev, Chris Elders
Summary: With the increased size and complexity of seismic surveys, manual labeling of seismic fades has become a significant challenge. Application of automatic methods for seismic facies interpretation could significantly reduce the manual labor and subjectivity of a particular interpreter present in conventional methods. Our developed deep convolutional autoencoder for unsupervised seismic facies classification can generate accurate facies maps instantaneously without human intervention.
Article
Multidisciplinary Sciences
Patricia M. Gregg, Yan Zhan, Falk Amelung, Dennis Geist, Patricia Mothes, Seid Koric, Zhang Yunjun
Summary: By combining satellite InSAR data with numerical models using high-performance computing data assimilation, the prolonged unrest and eruption timing of the Sierra Negra volcano in the Galapagos were successfully predicted. The evolution of the stress state in the surrounding rock and a faulting event were found to be key factors in the eruption.
Article
Computer Science, Interdisciplinary Applications
Shantanu Shahane, Erman Guleryuz, Diab W. Abueidda, Allen Lee, Joe Liu, Xin Yu, Raymond Chiu, Seid Koric, Narayana R. Aluru, Placid M. Ferreira
Summary: Surrogate neural network models are used in cell phone camera systems to accurately evaluate lens configurations and analyze optical properties. They provide efficient handling of large amounts of data for sensitivity and uncertainty analysis, and are valuable tools for optimizing tolerance design and component matching.
COMPUTERS & STRUCTURES
(2022)
Article
Engineering, Multidisciplinary
Junyan He, Diab Abueidda, Seid Koric, Iwona Jasiuk
Summary: This paper investigates the application of graph convolutional networks in the deep energy method model for solving the momentum balance equation of linear elastic and hyperelastic materials in three-dimensional space. Numerical examples demonstrate that the proposed method achieves similar accuracy with shorter run time compared to traditional methods. The study also discusses two different spatial gradient computation techniques.
INTERNATIONAL JOURNAL FOR NUMERICAL METHODS IN ENGINEERING
(2023)
Article
Geosciences, Multidisciplinary
Vladimir Puzyrev, Tristan Salles, Greg Surma, Chris Elders
Summary: This paper proposes a generator of 2D subsurface models based on deep generative adversarial networks. By training separate networks on realistic density and stratigraphy models, the generator is able to produce highly detailed and varied models in real-time. This approach enables the creation of large synthetic training datasets at a lower cost, facilitating the development of deep learning algorithms for real-time inversion and interpretation.
GEOSCIENCE LETTERS
(2022)
Article
Engineering, Multidisciplinary
Diab W. Abueidda, Seid Koric, Erman Guleryuz, Nahil A. Sobh
Summary: Physics-informed neural networks are used to solve equations governing physical phenomena, but they have issues that can be resolved using techniques like Fourier transform. This paper proposes a physics-informed neural network model with multiple loss terms and weight assignment using the coefficient of variation scheme. The model is standalone and meshfree, accurately capturing mechanical response. The study focuses on 3D hyperelasticity and demonstrates the model's performance by solving various problems.
INTERNATIONAL JOURNAL FOR NUMERICAL METHODS IN ENGINEERING
(2023)
Article
Engineering, Manufacturing
V. Perumal, D. Abueidda, S. Koric, A. Kontsos
Summary: Metal additive manufacturing (AM) involves complex multiscale and multiphysics processes. Deep learning-based approaches, specifically temporal convolutional networks (TCNs), have been proposed as a solution to the challenges faced by physics-based modeling methods in predicting thermal histories in AM. This study presents the use of TCNs for fast inferencing in directed energy deposition (DED) processes, achieving comparable accuracy to other deep learning methods with significantly reduced compute and training times.
JOURNAL OF MANUFACTURING PROCESSES
(2023)
Article
Thermodynamics
Seid Koric, Diab W. Abueidda
Summary: DeepONet approximates linear and nonlinear PDE solution operators by using parametric functions as inputs and mapping them to different PDE solution function output spaces. Unlike PINN, DeepONet models can predict parametric solutions in real-time without the need for retraining or transfer learning. It shows good performance in solving the heat conduction equation and is orders of magnitude faster than classical numerical solvers.
INTERNATIONAL JOURNAL OF HEAT AND MASS TRANSFER
(2023)
Article
Engineering, Mechanical
Daegun You, Orcun Koray Celebi, Ahmed Sameer Khan Mohammed, Diab W. Abueidda, Seid Koric, Huseyin Sehitoglu
Summary: A predictive model is developed to accurately predict the dislocation glide stress in FCC materials, considering the anisotropic continuum energy, the atomistic misfit energy, and the minimum energy path for the intermittent motion of Shockley partials. By generating a large material dataset and using machine learning, the model achieves a 94% accuracy in predicting the critical resolved shear stress for 1033 materials, revealing the sensitivity of material parameters to the predicted stress.
INTERNATIONAL JOURNAL OF PLASTICITY
(2023)
Article
Computer Science, Interdisciplinary Applications
Yapo Abole Serge Innocent Oboue, Yunfeng Chen, Sergey Fomel, Wei Zhong, Yangkang Chen
Summary: Strong noise can disrupt the recorded seismic waves and negatively impact subsequent seismological processes. To improve the signal-to-noise ratio (S/N) of seismological data, we introduce MATamf, an open-source MATLAB code package based on an advanced median filter (AMF) that simultaneously attenuates various types of noise and improves S/N. Experimental results demonstrate the usefulness and advantages of the proposed AMF workflow in enhancing the S/N of a wide range of seismological applications.
COMPUTERS & GEOSCIENCES
(2024)
Article
Computer Science, Interdisciplinary Applications
Upkar Singh, P. N. Vinayachandran, Vijay Natarajan
Summary: The Bay of Bengal maintains its salinity distribution due to the cyclic flow of high salinity water and the mixing with freshwater. This paper introduces an advection-based feature definition and algorithms to track the movement of high salinity water, validated through comparison with observed data.
COMPUTERS & GEOSCIENCES
(2024)
Article
Computer Science, Interdisciplinary Applications
Bijal Chudasama, Nikolas Ovaskainen, Jonne Tamminen, Nicklas Nordback, Jon Engstro, Ismo Aaltonen
Summary: This contribution presents a novel U-Net convolutional neural network (CNN)-based workflow for automated mapping of bedrock fracture traces from aerial photographs acquired by unmanned aerial vehicles (UAV). The workflow includes training a U-Net CNN using a small subset of photographs with manually traced fractures, semantic segmentation of input images, pixel-wise identification of fracture traces, ridge detection algorithm and vectorization. The results show the effectiveness and accuracy of the workflow in automated mapping of bedrock fracture traces.
COMPUTERS & GEOSCIENCES
(2024)
Article
Computer Science, Interdisciplinary Applications
Ruizhen Wang, Siyang Wan, Weitao Chen, Xuwen Qin, Guo Zhang, Lizhe Wang
Summary: This paper proposes a novel framework to generate a finer soil strength map based on RCI, which uses ensemble learning models to obtain USCS soil classification and predict soil moisture, in order to improve the resolution and reliability of existing soil strength maps.
COMPUTERS & GEOSCIENCES
(2024)
Article
Computer Science, Interdisciplinary Applications
Zhanlong Chen, Xiaochuan Ma, Houpu Li, Xuwei Xu, Xiaoyi Han
Summary: Simulated terrains are important for landform and terrain research, disaster prediction, rescue and disaster relief, and national security. This study proposes a deep learning method, IGPN, that integrates global information and pattern features of the local terrain to generate accurate simulated terrains quickly.
COMPUTERS & GEOSCIENCES
(2024)
Article
Computer Science, Interdisciplinary Applications
Daniele Secci, Vanessa A. Godoy, J. Jaime Gomez-Hernandez
Summary: Neural networks excel in various machine learning applications, but lack physical interpretability and constraints, limiting their accuracy and reliability in predicting complex physical systems' behavior. Physics-Informed Neural Networks (PINNs) integrate neural networks with physical laws, providing an effective tool for solving physical problems. This article explores recent developments in PINNs, emphasizing their application in solving unconfined groundwater flow, and discusses challenges and opportunities in this field.
COMPUTERS & GEOSCIENCES
(2024)
Article
Computer Science, Interdisciplinary Applications
Renguang Zuo, Ying Xu
Summary: This study proposes a hybrid deep learning model consisting of a one-dimensional convolutional neural network (1DCNN) and a graph convolutional network (GCN) to extract joint spectrum-spatial features from geochemical survey data for mineral exploration. The physically constrained hybrid model performs better in geochemical anomaly recognition compared to other models.
COMPUTERS & GEOSCIENCES
(2024)