Article
Construction & Building Technology
Qiushi He, Ziwei Li, Wen Gao, Hongzhong Chen, Xiaoying Wu, Xiaoxi Cheng, Borong Lin
Summary: This study proposes proxy models for daylight simulation for general floorplans based on convolutional neural network and generative adversarial network, which show promise in predicting daylight metrics and illuminance distribution and providing real-time feedback for designers. The experiments demonstrate the potential of using deep neural networks for feature extraction from general building forms and building predictive models to aid in automatic form-finding and design optimization.
BUILDING AND ENVIRONMENT
(2021)
Article
Computer Science, Information Systems
Xu Kang, Liang Liu, Huadong Ma
Summary: The study introduces a new framework for reconstructing urban environmental signals utilizing a generative adversarial network based on sensory data. Experimental results demonstrate its superior performance in signal recovery accuracy.
IEEE INTERNET OF THINGS JOURNAL
(2021)
Article
Engineering, Electrical & Electronic
Borui Hou, Ruqiang Yan
Summary: In this article, a new generative adversarial network (GAN) called the triplet-classifier GAN is designed for finger-vein verification. Unlike traditional GAN methods, this model uses generated fake data to improve the learning ability of the classifier. Experiments prove that this model has superior performance in finger-vein verification and shows promise in finger-vein-based biometric verification.
IEEE TRANSACTIONS ON INSTRUMENTATION AND MEASUREMENT
(2022)
Article
Engineering, Industrial
Clayton Cooper, Jianjing Zhang, Y. B. Guo, Robert X. Gao
Summary: Predicting machined surface roughness is essential for estimating part performance characteristics, but there is a lack of quantitative association between machining power and surface roughness. This paper presents a method using a conditional generative adversarial network (CGAN) to synthesize power signals and augment measured signals for predicting surface roughness. The experiments show that data augmentation by CGAN significantly improves the accuracy of surface roughness prediction.
JOURNAL OF MANUFACTURING SYSTEMS
(2023)
Article
Computer Science, Artificial Intelligence
Zezheng Lv, Xiaoci Huang, Wenguan Cao
Summary: The paper introduces a novel Generative Adversarial Network model for predicting future pedestrian trajectories, capturing path uncertainty and generating more reasonable results. The method includes a generator with convolutional self-attention and Mish Feed-Forward Network, as well as a discriminator for classifying predicted and ground truth paths as socially acceptable.
INTERNATIONAL JOURNAL OF INTELLIGENT SYSTEMS
(2022)
Article
Computer Science, Information Systems
Subba Rao Polamuri, Kudipudi Srinivas, A. Krishna Mohan
Summary: Deep learning has achieved significant success in the financial domain, particularly in stock market prediction. This paper addresses the limited use of generative adversarial networks (GANs) in stock market prediction by proposing a GAN-based hybrid prediction algorithm. The algorithm overcomes the difficulty in setting hyperparameters using reinforcement learning and Bayesian optimization. Empirical results demonstrate the promising performance of the GAN-based deep learning framework (Stock-GAN) compared to the state-of-the-art model (MM-HPA) in stock price prediction.
JOURNAL OF KING SAUD UNIVERSITY-COMPUTER AND INFORMATION SCIENCES
(2022)
Article
Computer Science, Information Systems
Haiyan Jin, Guodong Xu, Kangda Cheng, Jinlong Liu, Zhilu Wu
Summary: Link prediction is an important research direction in complex network analysis with broad application prospects. Traditional link prediction algorithms based on sparse expression of the adjacency matrix suffer from high computational cost, inefficiency, and inability to run on large-scale networks while preserving their higher order structural features. To address this, we propose a GAN-based link prediction algorithm that layers the network graph, preserves local and higher-level structural features, and uses a generative adversarial model to obtain low-dimensional vector forms. Our method outperforms many state-of-the-art algorithms.
Article
Computer Science, Artificial Intelligence
Yuchen Wei, Shuxiang Xu, Byeong Kang, Sabera Hoque
Summary: This paper proposes a multi-angle Generative Adversarial Network (MAGAN) for data augmentation in grocery product recognition tasks. By generating realistic training images from different angles, the deep learning model can achieve improved accuracy in recognizing grocery products.
Article
Environmental Sciences
Ali Jamali, Masoud Mahdianpari, Fariba Mohammadimanesh, Brian Brisco, Bahram Salehi
Summary: Developed in response to the rapid loss or change of natural ecosystems, especially wetlands, due to human activities and climate change, this study presents a Deep Convolutional Neural Network (DCNN) model using a modified architecture of AlexNet and a Generative Adversarial Network (GAN) for wetland classification and generation of Sentinel-1 and Sentinel-2 data. Tested in a 370 sq. km area in Newfoundland, the proposed model achieved an average accuracy of 92.30% with improved F-1 scores for various wetland classes compared to the original CNN network of AlexNet. These results demonstrate the high capability of the proposed model for large-scale wetland classification tasks.
Article
Computer Science, Artificial Intelligence
Fang Fang, Pengpeng Zhang, Bo Zhou, Kun Qian, Yahui Gan
Summary: This paper proposes a method based on generative adversarial network model and attention mechanism to predict pedestrian trajectories in dynamic scenarios. The method can extract the interaction information between pedestrians and allocate the influence weight of pedestrians using an attention pooling module. Experimental results demonstrate the superiority of this method in prediction accuracy compared to existing methods.
COGNITIVE COMPUTATION
(2022)
Article
Engineering, Mechanical
Xiao Wang, Bo Xu, Tailin Han, Yan Wang
Summary: Limited by the dynamic characteristics of the sensor, the high-frequency signal passing through the sensor will be distorted by dynamic error, affecting the accuracy of the real value. This paper proposes a compensation model based on deep learning to reduce the dynamic error. By using deep convolutional generative adversarial networks, the limited sensor dynamic data is augmented, and a sensor compensation model is obtained through speech enhancement generative adversarial networks. This method has demonstrated better results than traditional methods in the example of a pressure sensor.
MECHANICAL SYSTEMS AND SIGNAL PROCESSING
(2023)
Article
Computer Science, Hardware & Architecture
Syed M. Raza, Boyun Jang, Huigyu Yang, Moonseong Kim, Hyunseung Choo
Summary: This paper proposes an Improved Generative Adversarial Network with Fact Forcing (iGAN-FF) for mobility prediction, which reduces the need for training data and achieves higher accuracy through adversarial learning. Experimental results confirm the superiority of the iGAN-FF method.
JOURNAL OF NETWORK AND COMPUTER APPLICATIONS
(2022)
Article
Construction & Building Technology
Gaowei Zhang, Yue Pan, Limao Zhang
Summary: This research introduces a semisupervised generative adversarial network (SSGAN) with two sub-networks for automatic defect detection, utilizing attention mechanism and dual loss functions to enhance segmentation performance and reduce data labeling efforts.
AUTOMATION IN CONSTRUCTION
(2021)
Article
Optics
Ju Liu, Jin Duan, Youfei Hao, Guangqiu Chen, Hao Zhang
Summary: Polarization image fusion is the process of fusing an intensity image and a polarization parameter image into a more detailed image. Conventional fusion strategies lack targeting and robustness due to their failure to account for differences in polarization properties and manually designed fusion rules. To address this, a novel network model called SGPF-GAN is proposed, which employs a polarization image information quality discriminator to guide the fusion process.
Article
Chemistry, Multidisciplinary
Hyung Yong Kim, Ji Won Yoon, Sung Jun Cheon, Woo Hyun Kang, Nam Soo Kim
Summary: This study proposed a progressive generator and multi-scale discriminator to address two issues in speech enhancement. Experimental results demonstrated that the proposed approach can improve training speed, stability, and performance on various metrics for speech enhancement.
APPLIED SCIENCES-BASEL
(2021)
Article
Mechanics
Jorge S. Salinas, Santiago Zuniga, M. Cantero, M. Shringarpure, J. Fedele, D. Hoyal, S. Balachandar
Summary: This study investigates gravity currents in the subcritical to supercritical range through seven direct and large-eddy simulations. The focus is on the near-self-similar state reached by the currents downstream, where certain factors reach a constant value while others continue to increase linearly. The study also examines their dependence on the slope.
JOURNAL OF FLUID MECHANICS
(2022)
Article
Mechanics
Jacob Behrendt, S. Balachandar, Joshua Garno, Thomas P. McGrath
Summary: Predicting the force on an isolated particle when a shock wave passes over it is an important problem in natural and industrial applications. The force on the particle has been observed to exhibit a nonmonotonic behavior with a sharp peak when the shock wave is located halfway across the particle. This nonmonotonic behavior is due to the unsteady nature of the compression and rarefaction waves that radiate as the shock wave diffracts around the particle, making it unpredictable with a quasi-steady model. An accurate force model must consider the unsteady nature of the flow and the sharp discontinuities in flow properties caused by the shock wave.
Article
Mechanics
J. St Clair, T. McGrath, S. Balachandar
Summary: This study describes particle-resolved simulations of a single aluminum particle interacting with a strong nitromethane shock within a layer of particles. The results show that the presence of neighboring particles influences the shape, plastic deformation, and strength of shock reflection of the particles.
Article
Thermodynamics
Sudhanshu Pandey, Sang Youl Yoon, S. Balachandar, Man Yeong Ha
Summary: The thermal and flow characteristics of carboxymethyl cellulose (CMC) were studied using experimental and numerical methods to assess shear-thinning behavior in a non-Newtonian fluid. A novel particle image velocimetry technique was utilized to analyze natural convective flow. The Carreau model was found to be the most effective in predicting the thermal and flow performance within the studied thermal system.
INTERNATIONAL JOURNAL OF HEAT AND MASS TRANSFER
(2022)
Article
Environmental Sciences
Minglan Yu, Xiao Yu, S. Balachandar
Summary: A two-phase Euler-Lagrangian framework was used to study the flocculation dynamics of cohesive sediment in isotropic turbulence. The research found that the ratio of turbulent shear to floc strength strongly influenced the floc size distribution and settling velocity.
WATER RESOURCES RESEARCH
(2022)
Article
Computer Science, Interdisciplinary Applications
M. Allahyari, K. Liu, J. Salinas, N. Zgheib, S. Balachandar
Summary: Using large eddy simulations, the dynamic behavior of puffs and droplets generated by coughs and sneezes was investigated. The effects of ejection volume, momentum, orientation, and mouth shape on the dynamics were explored. It was found that the ejection angle and mouth shape have minimal impact on the puff and droplet dynamics.
COMPUTERS & FLUIDS
(2023)
Article
Mechanics
S. Balachandar, Kai Liu
Summary: This study improves upon prior correction procedures for particle self-induced perturbation velocity in a two-way coupled Euler-Lagrange simulation. It introduces a vector correction procedure that allows for different directions of feedback force and relative velocity, and incorporates the effect of a nearby wall. Analytical and numerical methods are presented for obtaining the regularization Oseen solutions necessary for the correction procedure. The study demonstrates the rapid convergence of the correction procedure within three or four iterations.
INTERNATIONAL JOURNAL OF MULTIPHASE FLOW
(2023)
Article
Mechanics
K. A. Krishnaprasad, J. S. Salinas, N. Zgheib, S. Balachandar
Summary: We propose a statistical framework to estimate airborne droplet nuclei concentration in indoor spaces by accounting for the effects of recycling and filtration in ventilation systems. The framework is demonstrated in a typical room with a four-way cassette air-conditioning system, where the flow field is computed using large eddy simulations. Our approach breaks down the path of virus-laden droplet nuclei into four separate processes, enabling us to provide turbulence-informed and statistically relevant pathogen concentration at any location in the room. The analysis shows that proper filtration can significantly increase the cumulative exposure time and allow longer visitations in nursing homes.
Article
Mechanics
Sam Briney, S. Balachandar
Summary: Supersonic aircraft flying in bad weather can be damaged by impacts from water droplets and other airborne particles. The particles encounter a bow shock before impact, causing them to break up into smaller droplets. These smaller droplets are less likely to collide with the aircraft due to their reduced inertia.
Article
Mechanics
Jorge S. Salinas, S. Balachandar, Santiago L. Zuniga, M. Shringarpure, J. Fedele, D. Hoyal, M. Cantero
Summary: Gravity currents are studied in this work, focusing on the flow of a heavier fluid along the bottom of a sloping bed, beneath a stagnant lighter ambient fluid. The thickness of the current increases due to entrainment of ambient fluid. The rate of penetration of mean momentum, mean concentration, and turbulence-related quantities into the ambient fluid are analyzed, with a comparison to wall-bounded turbulent flows.
Article
Multidisciplinary Sciences
Minglan Yu, Xiao Yu, Ashish J. Mehta, Andrew J. Manning, Faisal Khan, S. Balachandar
Summary: The study investigates the impact of turbulence on flocculation by analyzing the time-evolution of individual flocs within an Eulerian-Lagrangian framework. It identifies two mechanisms of floc reshaping: breakage-regrowth and restructuring by hydrodynamic drag. Surface erosion is the primary breakup mechanism for strong flocs, while fragile flocs tend to split into fragments of similar sizes. Turbulence lowers the aggregation efficiency of flocs comparable to or greater than the Kolmogorov scale. The findings emphasize the restrictive effects of turbulence on both floc size and structure.
SCIENTIFIC REPORTS
(2023)
Article
Physics, Fluids & Plasmas
B. Siddani, S. Balachandar
Summary: Developing deterministic neighborhood-informed point-particle closure models using machine learning has attracted interest in the dispersed multiphase flow community recently. However, the robustness of neural models for this complex multibody problem is hindered by the lack of particle-resolved data. In this study, two strategies are implemented to address this limitation: the use of a rotation and reflection equivariant neural network, and a physics-based hierarchical machine learning approach. The resulting machine-learned models demonstrate high accuracy in predicting neighbor-induced force and torque fluctuations under various conditions.
PHYSICAL REVIEW FLUIDS
(2023)
Article
Computer Science, Interdisciplinary Applications
K. Choudhary, K. A. Krishnaprasad, S. Pandey, N. Zgheib, J. S. Salinas, M. Y. Ha, S. Balachandar
Summary: We investigate the dispersal of droplet nuclei in a room with a four-way cassette air-conditioning unit using RANS simulations. The simulations compare well with reference LES simulations, suggesting that the computationally cheaper RANS model can be used to predict pathogen concentration in confined spaces. However, there may be an increased statistical discrepancy.
COMPUTERS & FLUIDS
(2023)
Article
Mechanics
Andreas Nygard Osnes, Magnus Vartdal, Mehdi Khalloufi, Jesse Capecelatro, S. Balachandar
Summary: This article presents models for quasi-steady drag, quasi-steady drag variation, and transverse forces in compressible flows through random, fixed particle suspensions. The correlations are formulated based on drag force measurements obtained from particle-resolved simulation data covering a range of flow conditions from subsonic to supersonic. The newly proposed drag models extend existing models for incompressible dense gas-solid flows to finite Mach numbers, while the transverse particle force model is novel and incorporates the covariation between streamwise and transverse forces.
INTERNATIONAL JOURNAL OF MULTIPHASE FLOW
(2023)
