Article
Chemistry, Physical
Yuting Weng, Dezhi Zhou
Summary: In this paper, a multiscale physics-informed neural network (MPINN) approach is proposed for solving stiff chemical kinetic problems. MPINNs group chemical species with different time scales and train them using multiple neural networks. By introducing a small number of ground truth data points and adding data loss terms, MPINNs achieve high-precision prediction of stiff ODE solutions. The validation results show that MPINNs effectively avoid the influence of stiffness on neural network optimization.
JOURNAL OF PHYSICAL CHEMISTRY A
(2022)
Article
Thermodynamics
Xu Han, Ming Jia, Yachao Chang, Yaopeng Li
Summary: The study successfully developed a deep learning model as a fast alternative to numerical integration for ignition simulations, proposing methodological improvements to extend the model's applicable range. These techniques can unlock further potentials of artificial neural networks in making reliable predictions in practical applications. The comparison between these methods and conventional ones also provides a deeper understanding of the impact of data distribution and network architecture on the performance of ANN models.
COMBUSTION AND FLAME
(2022)
Article
Computer Science, Theory & Methods
Nhut-Minh Ho, Weng-Fai Wong
Summary: This article proposes Tensorox, a framework that utilizes the half-precision tensor cores on recent GPUs to accelerate non deep learning applications. By training shallow neural networks and running multiple instances in parallel using tensor operations on Nvidia GPUs, our approximation achieves higher accuracy than running the original single precision programs, while allowing for runtime adjustment of the degree of approximation.
IEEE TRANSACTIONS ON PARALLEL AND DISTRIBUTED SYSTEMS
(2022)
Article
Energy & Fuels
Majid Haghshenas, Peetak Mitra, Niccolo Dal Santo, David P. Schmidt
Summary: The LIT methodology utilizes machine learning algorithms to accelerate combustion kinetics modeling in high-dimensional composition spaces, achieving good results through data clustering and localized DNN training. Clustering is performed using SOM, fully connected layer DNN models are optimized with Bayesian optimization, and a nonlinear transformation improves sensitivity to minor species, reducing prediction errors for ignition delay.
Article
Computer Science, Interdisciplinary Applications
Magnus Furst, Andrea Bertolino, Alberto Cuoci, Tiziano Faravelli, Alessio Frassoldati, Alessandro Parente
Summary: An adaptable chemical kinetics optimization toolbox has been developed in this work, which can handle a large number of uncertain parameters and utilize various optimization methodologies. The toolbox is capable of handling experimental targets from different sources, showcasing its versatility in predicting chemical reactions.
COMPUTER PHYSICS COMMUNICATIONS
(2021)
Article
Computer Science, Artificial Intelligence
Yuxuan Liang, Kun Ouyang, Yiwei Wang, Zheyi Pan, Yifang Yin, Hongyang Chen, Junbo Zhang, Yu Zheng, David S. Rosenblum, Roger Zimmermann
Summary: Spatio-temporal forecasting has various applications in smart cities, but the state-of-the-art method, GCRNN, fails to consider higher-order spatial relations and underlying physics in real-world systems. Therefore, we propose MixRNN+, a general model that captures complex spatial relations and addresses underlying physics, for spatio-temporal forecasting. Experimental results on three forecasting tasks demonstrate the superiority of MixRNN+ against existing methods, and a cloud-based system using MixRNN+ as the bedrock model showcases its practicality.
IEEE TRANSACTIONS ON KNOWLEDGE AND DATA ENGINEERING
(2023)
Article
Computer Science, Information Systems
Seog Chung Seo
Summary: Researchers proposed an efficient implementation of the SIKE mechanism on GPUs, optimizing underlying field arithmetic and taking full advantage of the GPU architecture. Experimental results showed that the GPU software outperformed the SIKE CPU software on Intel i9-10900K CPU by a factor of 140.64-146.81.
Article
Mechanics
Bohao Zhou, Xudong Huang, Ke Zhang, Dianfang Bi, Ming Zhou
Summary: This paper proposes a block data-parallel lower-upper relaxation (BDPLUR) scheme based on Jacobi iteration and Roe's flux scheme and implements it on a GPU. Numerical experiments show that the BDPLUR scheme, especially when implemented on a GPU, is more than ten times faster than the original data-parallel lower-upper relaxation scheme.
Article
Engineering, Aerospace
V. B. Betelin, B. Kryzhanovsky, N. N. Smirnov, V. F. Nikitin, I. M. Karandashev, M. Yu Malsagov, E. Mikhalchenko
Summary: This study focuses on using artificial neural networks to solve chemical kinetics problems, developing a simple model capable of predicting system behavior in a recursive mode.
Article
Computer Science, Information Systems
Cihangir Tezcan
Summary: This work focuses on the performance of the AES algorithm on GPUs, achieving significant breakthroughs in optimization to provide higher encryption throughput and surpassing CPU performance with hardware instructions and traditional FPGA clusters. Transitioning from AES-128 to AES-256 on GPUs has been proven to offer increased security without sacrificing performance.
Article
Thermodynamics
Songlin Liu, Weidong Fan, Xin Wang, Jun Chen, Hao Guo
Summary: This study re-evaluates the rate constant and activation energy of N2O elementary reaction and improves the elementary reaction involving N2O based on previous research. The kinetic parameters of N2O reacting with other substances are also updated, providing important references for the research on N2O and nitrogen oxides.
Article
Computer Science, Software Engineering
A. Salmi, Sz Csefalvay, J. Imber
Summary: The use of realism enhancement methods in real-time and resource-constrained settings has been hindered by the high cost of existing methods. These methods achieve high quality results at the expense of long runtimes and high requirements for bandwidth, memory, and power. A more efficient alternative has been proposed: a high-performance, generative shader-based approach that applies machine learning techniques to real-time applications, even in resource-constrained environments such as embedded systems and mobile GPUs. The proposed learnable shader pipeline, composed of differentiable functions, can be trained end-to-end using an adversarial objective to accurately reproduce the appearance of a target image set without manual tuning. The shader pipeline is optimized for efficient execution on the target device, delivering temporally stable, faster-than-real time results that rival the quality of many neural network-based methods.
COMPUTER GRAPHICS FORUM
(2023)
Article
Computer Science, Theory & Methods
Johannes Pekkila, Miikka S. Vaisala, Maarit J. Kapyla, Matthias Rheinhardt, Oskar Lappi
Summary: Modern compute nodes provide high parallelism and processing power. Optimization of data movement is critical for achieving strong scaling in communication-heavy applications. This study explores the computational aspects of iterative stencil loops and implements a communication scheme using CUDA-aware MPI to accelerate magnetohydrodynamics simulations.
PARALLEL COMPUTING
(2022)
Article
Neurosciences
Rin Kuriyama, Claudia Casellato, Egidio D'Angelo, Tadashi Yamazaki
Summary: This study focuses on large-scale simulation of detailed computational models of neuronal microcircuits using a scaffolding approach, which involves replacing simulation modules to improve computational speed. The research demonstrates that the scaffolding method can accelerate real-time simulation significantly, reducing computational time without affecting experimental results.
FRONTIERS IN CELLULAR NEUROSCIENCE
(2021)
Article
Computer Science, Theory & Methods
Fei Li, Yuzhu Wang, Jinrong Jiang, He Zhang, Xiaocong Wang, Xuebin Chi
Summary: The physical process of atmospheric cumulus convection is crucial in climate modeling, but its computational complexity hinders the development of high-resolution models. This paper proposes parallel algorithms for the University of Washington shallow cumulus (UWshcu) model, suitable for large-scale, heterogeneous computing systems. The experimental results demonstrate the efficiency and scalability of these algorithms.
FUTURE GENERATION COMPUTER SYSTEMS-THE INTERNATIONAL JOURNAL OF ESCIENCE
(2023)
Article
Thermodynamics
Shivam Barwey, Malik Hassanaly, Venkat Raman, Adam Steinberg
Summary: This work utilizes data-driven methods to transform experimental OH-PLIF images into corresponding three-component planar PIV fields using a fully convolutional network. The performance of global CNN and local CNNs is compared, and the inclusion of time history in the PLIF input is investigated. The study also reveals the ability of local CNNs to utilize symmetry and antisymmetry in unseen domain regions.
COMBUSTION SCIENCE AND TECHNOLOGY
(2022)
Article
Thermodynamics
Shivam Barwey, Supraj Prakash, Malik Hassanaly, Venkat Raman
Summary: This research implemented a data-driven approach to classify combustion regimes in detonation waves and demonstrated a procedure for domain-localized source term modeling based on these classifications. Through clustering analysis, distinctions between different combustion regimes within the detonation wave structure were illuminated. By using artificial neural networks, localized source term modeling was guided, leading to improved estimations of source terms.
FLOW TURBULENCE AND COMBUSTION
(2021)
Article
Thermodynamics
Yihao Tang, Venkat Raman
Summary: The study introduces a novel strained, non-adiabatic flamelet generated manifold model based on a counterflow premixed flame, which shows improved accuracy in predicting flame behavior. By varying strain and heat loss effects, different combustion modes are distinguished, enhancing the understanding of turbulent flame evolution.
COMBUSTION AND FLAME
(2021)
Review
Thermodynamics
Malik Hassanaly, Venkat Raman
Summary: In practical combustion system design, ensuring safety and reliability is crucial, but analyzing failure events and extending it to failure events faces challenges. Specifically, in data-poor problems with high cost data generation, developing predictions for extreme events is critical.
PROGRESS IN ENERGY AND COMBUSTION SCIENCE
(2021)
Review
Mechanics
Venkat Raman, Supraj Prakash, Mirko Gamba
Summary: A rotating detonation engine (RDE) uses a traveling detonation wave to compress and release heat, providing high efficiency in small volumes. However, nonidealities such as unsteady mixing and multiple competing waves need to be managed for optimal performance. This review discusses the understanding of these nonidealities and the techniques used to study them in RDEs.
ANNUAL REVIEW OF FLUID MECHANICS
(2023)
Article
Thermodynamics
Harshavardhana A. Uranakara, Shivam Barwey, Francisco E. Hernandez Perez, Vijayamanikandan Vijayarangan, Venkat Raman, Hong G. Im
Summary: This study evaluates the impact of transferring computationally intensive tasks to the GPU in solving compressible reacting flow problems. The workflow and data transfer penalties of plugging in a GPU-based chemistry library into CPU-based and GPU-based solvers are compared. The results show that offloading the source term computation to the GPU leads to a significant reduction in solution time and faster computation compared to CPU-based methods.
PROCEEDINGS OF THE COMBUSTION INSTITUTE
(2023)
Article
Thermodynamics
Vidhan Malik, Sheikh Salauddin, Rachel Hytovick, Ral Bielawski, Venkat Raman, John Bennewitz, Jason Burr, Eric Paulson, William Hargus, Kareem Ahmed
Summary: This study explores the dynamics of detonation waves driven by aerosolized liquid fuel sprays. Evidence of RP-2 driving the detonation phenomenon is quantified using dynamic pressure measurements and four simultaneous induced fluorescence (PLIF), and particle Mie scatter. The investigation provides supporting information on liquid fuel droplet burning and heat release driving the detonation wave.
PROCEEDINGS OF THE COMBUSTION INSTITUTE
(2023)
Article
Computer Science, Interdisciplinary Applications
Ral Bielawski, Shivam Barwey, Supraj Prakash, Venkat Raman
Summary: Emerging supercomputing systems combine CPUs and GPUs to reach exascale capabilities with minimal energy footprint. This presents challenges for fluid solvers due to the differences in hardware architecture and operation between GPUs and CPUs. This work presents a general approach for efficient implementation of finite-volume based reacting flow solvers on heterogeneous systems. Specific algorithms are developed to handle compressible reacting flows, including chemical reactions, convection terms, and turbulence, ensuring GPU-based efficiency. The approach is demonstrated using the OpenFOAM software and shows excellent scalability on a large number of GPUs (> 3000) with constant throughput for a large number of control volumes.
COMPUTERS & FLUIDS
(2023)
Article
Multidisciplinary Sciences
S. Barwey, V. Raman
Summary: A new approach called time-axis clustering is developed in this work for modal decomposition through re-interpretation of unsteady dynamics, demonstrated on an experimental turbulent reacting flow dataset. The method uses K-Means clustering algorithm to interpret the dataset as one-dimensional time series, identifying spatial modes and temporal coefficients that represent average trajectories of flow quantity of interest conditioned on the regions in physical space. The non-overlapping nature of K-Means clusters allows for visualization of modes, providing a unique pathway for flow feature extraction based on dynamical similarity.
PROCEEDINGS OF THE ROYAL SOCIETY A-MATHEMATICAL PHYSICAL AND ENGINEERING SCIENCES
(2023)
Article
Thermodynamics
Michael Ullman, Supraj Prakash, Deborah Jackson, Venkat Raman, Carson Slabaugh, John Bennewitz
Summary: Detonative combustors, such as rotating detonation engines, offer higher theoretical thermal efficiencies and are viewed as next-generation propulsion and power generation systems. This study simulates a methane-oxygen reflective shuttling detonation combustor to better understand the impacts of various conditions and phenomena. The simulations successfully predict the wave velocity trends and reveal significant reactant stratification and parasitic combustion ahead of the waves, affecting their speeds and pressures.
COMBUSTION AND FLAME
(2023)
Article
Thermodynamics
Michael Ullman, Venkat Raman
Summary: This study develops a calibration procedure for a one-dimensional model of the wall pressure in a scramjet flowpath. Using wall pressure measurements from three-dimensional simulations, the six model parameters are tuned using a Bayesian methodology. The results show that the calibrated model can capture the mean wall pressure in both the isolator and combustor. The procedure can converge for both uniform and Gaussian priors, and physically-consistent correlations between parameters are obtained.
COMBUSTION SCIENCE AND TECHNOLOGY
(2023)
Article
Thermodynamics
Michael Ullman, Shivam Barwey, Gyu Sub Lee, Venkat Raman
Summary: The advent of data-based modeling has provided new methods and algorithms for analyzing complex flow fields in high-speed combustion applications, which can provide insight into the underlying physics of the system and identify avenues for the development and application of new models.
APPLICATIONS IN ENERGY AND COMBUSTION SCIENCE
(2023)
Article
Engineering, Chemical
Hengrui Liu, Fatemeh Salehi, Rouzbeh Abbassi, Tim Lau, Guan Heng Yeoh, Fiona Mitchell-Corbett, Venkat Raman
Summary: This paper investigates the dispersion of leaked hydrogen and concludes that ventilation is a critical safety measure to mitigate the risk of fire and explosion.
JOURNAL OF LOSS PREVENTION IN THE PROCESS INDUSTRIES
(2023)
Article
Thermodynamics
Supraj Prakash, Venkat Raman
Summary: Pressure gain combustion through continuous detonations, particularly in rotating detonation engines (RDEs), offers significant efficiency improvements in propulsion and energy conversion devices. However, non-premixed fuel injection in practical RDEs faces challenges such as turbulence-induced shock-front variations and incomplete fuel-air mixing. Preburning was found to weaken shock fronts and lead to delayed combustion of partially-oxidized fuel-air mixtures, contributing to combustion efficiency losses. This parasitic combustion process hinders detonation efficiency by diverting heat release away from the shock wave.
PROCEEDINGS OF THE COMBUSTION INSTITUTE
(2021)
Article
Thermodynamics
Supraj Prakash, Venkat Raman, Christopher F. Lietz, William A. Hargus, Stephen A. Schumaker
Summary: The rotating detonation engine is an important realization of pressure gain combustion for rocket applications, characterized by a highly unsteady flow field and entrainment of partially-burnt gases post-detonation. Numerical simulations show spatially fluctuating detonation wave strengths and parasitic combustion of fresh reactants in the flow field.
PROCEEDINGS OF THE COMBUSTION INSTITUTE
(2021)