Article
Engineering, Multidisciplinary
Mengwu Guo, Shane A. McQuarrie, Karen E. Willcox
Summary: This work proposes a Bayesian inference method for the reduced-order modeling of time-dependent systems. The study formulates the task of learning a reduced-order model as a Bayesian inverse problem, with a Gaussian prior and likelihood. The resulting posterior distribution characterizes the operators defining the reduced-order model, enabling predictions with uncertainty. The method estimates statistical moments of the predictions through efficient Monte Carlo sampling.
COMPUTER METHODS IN APPLIED MECHANICS AND ENGINEERING
(2022)
Article
Computer Science, Interdisciplinary Applications
Dongwei Ye, Valeria Krzhizhanovskaya, Alfons G. Hoekstra
Summary: This work presents a data-driven surrogate model for efficient prediction of blood flow simulations on similar but distinct domains. The proposed model utilizes group surface registration to parameterize shapes and uses geometry information for hemodynamics prediction. The results demonstrate that the surrogate model has accuracy and efficiency in hemodynamics prediction and can be applied to real-time simulation and uncertainty quantification for complex patient-specific scenarios.
JOURNAL OF COMPUTATIONAL PHYSICS
(2024)
Article
Mechanics
Zhengxiao Ma, Jian Yu, Ruoye Xiao
Summary: This paper proposes a nonintrusive reduced basis (RB) method based on dynamic mode decomposition (DMD) for parameterized time-dependent flows. The offline stage involves extracting the reduced basis functions and introducing a novel hybrid DMD regression model for the temporal evolution of the RB coefficients. To enhance stability for complex nonlinear problems, a threshold value is used to modify the DMD eigenvalues and eigenvectors. Additionally, interpolation of the coefficients in parameter space is performed using a feedforward neural network or random forest algorithm. The online stage enables the prediction of the RB solution at a new time/parameter value with low computational cost and complete decoupling from the high-fidelity dimension. The proposed model is demonstrated with two cases, showing reasonable efficiency and robustness.
Article
Mechanics
Xinshuai Zhang, Tingwei Ji, Fangfang Xie, Changdong Zheng, Yao Zheng
Summary: A novel data-driven nonlinear reduced-order modeling framework is proposed for predicting the flow field variations in unsteady fluid-structure interactions (FSIs). By using a convolutional variational autoencoder model and the sparse identification of nonlinear dynamics (SINDy) algorithm, the framework efficiently extracts the nonlinear low-dimensional manifolds and the dynamical equations of the vibration responses, revealing the underlying FSI mechanism.
Article
Engineering, Multidisciplinary
Anthony Gruber, Max Gunzburger, Lili Ju, Zhu Wang
Summary: This study investigates the impact of autoencoder architecture on reduced-order models, comparing deep convolutional autoencoders with other autoencoder alternatives. The results demonstrate that the proposed architecture shows superior performance when applied to data with irregular connectivity and a sufficiently large latent space.
COMPUTER METHODS IN APPLIED MECHANICS AND ENGINEERING
(2022)
Article
Computer Science, Interdisciplinary Applications
Junming Duan, Jan S. Hesthaven
Summary: This paper proposes a non-intrusive reduced-order modeling approach for time-dependent parametrized problems. It uses a convolutional autoencoder for dimensionality reduction and high-order dynamic mode decomposition for modeling time-dependent problems. Numerical tests show that the approach can accurately predict unseen full-order solutions.
JOURNAL OF COMPUTATIONAL PHYSICS
(2024)
Article
Computer Science, Interdisciplinary Applications
Mengnan Li, Lijian Jiang
Summary: In this article, the authors introduce data-driven reduced-order modeling for nonautonomous dynamical systems in multiscale media. They explain the use of the Koopman operator and its dependence on time pair in nonautonomous systems. To estimate time-dependent Koopman operators, a moving time window is used for data decomposition and the extended dynamic mode decomposition method is used. The authors propose reduced-order modeling as a strategy to solve the challenge of high-dimensional data and introduce offline and online stages. They also develop three methods for online reduced-order modeling: fully online, semi-online, and adaptive online. Numerical examples are provided to demonstrate the performance of different methods.
JOURNAL OF COMPUTATIONAL PHYSICS
(2023)
Article
Engineering, Multidisciplinary
Changhong Mou, Birgul Koc, Omer San, Leo G. Rebholz, Traian Iliescu
Summary: A new data-driven reduced order model (ROM) framework is proposed in this study, which is based on the hierarchical structure of the variational multiscale (VMS) methodology and utilizes data to enhance ROM accuracy. By separating scales, identifying closure terms, and modeling them with available data, the new framework shows significantly higher accuracy compared to standard ROMs.
COMPUTER METHODS IN APPLIED MECHANICS AND ENGINEERING
(2021)
Article
Computer Science, Interdisciplinary Applications
Changhong Mou, Nan Chen, Traian Iliescu
Summary: This paper develops a systematic multiscale stochastic reduced order model (ROM) framework for complex systems with strong chaotic or turbulent behavior. The new ROM focuses on recovering the large-scale dynamics and also captures the statistical features of medium-scale variables. It incorporates physics constraints and facilitates an efficient and accurate scheme for nonlinear data assimilation.
JOURNAL OF COMPUTATIONAL PHYSICS
(2023)
Article
Mathematics, Applied
Alec J. Linot, Michael D. Graham
Summary: In this study, we propose a data-driven reduced-order modeling method for chaotic dynamics, which involves finding the coordinate representation of the manifold and describing the dynamics using a system of ordinary differential equations (ODEs) in this coordinate system. We apply this method to a specific system and find that dimension reduction improves performance compared to predictions in the ambient space. Furthermore, we demonstrate that the low-dimensional model is capable of accurately recreating the true dynamics using widely spaced data.
Article
Engineering, Multidisciplinary
C. Liu, R. Fu, D. Xiao, R. Stefanescu, P. Sharma, C. Zhu, S. Sun, C. Wang
Summary: This work presents a new predictive data assimilation framework based on a data-driven reduced order model (DDROM), and demonstrates its capabilities through two test cases.
ENGINEERING ANALYSIS WITH BOUNDARY ELEMENTS
(2022)
Article
Computer Science, Interdisciplinary Applications
Fleurianne Bertrand, Daniele Boffi, Abdul Halim
Summary: In this article, a data-driven reduced basis method is proposed for approximating parametric eigenvalue problems. The method utilizes offline and online stages, where snapshots are generated and a basis for the reduced space is constructed using a POD approach in the offline stage. Gaussian process regressions are employed to approximate the eigenvalues and projection coefficients in the reduced space. The trained regressions can be used in the online stage to obtain outputs corresponding to new parameters.
JOURNAL OF COMPUTATIONAL PHYSICS
(2023)
Article
Engineering, Multidisciplinary
Jacob Fish, Yang Yu
Summary: A hybrid data-physics driven reduced-order homogenization (dpROH) approach has been developed to improve the accuracy of physics-based reduced order homogenization (pROH) while preserving its interpretability and extrapolation. The dpROH utilizes data generated by a high-fidelity model to enhance the accuracy of the physics-based model reduction. The dpROH consists of an offline stage employing a Bayesian inference (BI) strategy with a gated recurrent unit (GRU) neural network surrogate, and an online stage utilizing dpROH for component level predictions. Numerical examples demonstrate improved accuracy compared to pROH and reference solution.
INTERNATIONAL JOURNAL FOR NUMERICAL METHODS IN ENGINEERING
(2023)
Article
Computer Science, Interdisciplinary Applications
Anna Ivagnes, Giovanni Stabile, Andrea Mola, Traian Iliescu, Gianluigi Rozza
Summary: In this paper, data-driven closure/correction terms are developed to improve the accuracy of pressure and velocity in reduced order models (ROMs) for fluid flows. The proposed pressure-based data-driven variational multiscale ROM uses available data to construct closure/correction terms for the momentum equation and continuity equation. Numerical investigation shows that the novel pressure data-driven variational multiscale ROM yields significantly more accurate velocity and pressure approximations compared to the standard ROM and the original data-driven variational multiscale ROM without pressure components.
JOURNAL OF COMPUTATIONAL PHYSICS
(2023)
Article
Materials Science, Multidisciplinary
Jialu Song, Hujin Xie, Yongmin Zhong, Jiankun Li, Chengfan Gu, Kup-Sze Choi
Summary: The paper introduces a new reduced-order nonlinear Kalman filter to emulate nonlinear behaviors of biological deformable tissues for accurate simulation of tissue physical deformation in real time. The approach reduces the order of the nonlinear state-space equation to decrease computational cost, constructing an extended Kalman filter to calculate tissue physical deformation behaviors online. Simulation results and comparison analysis verify the effectiveness of the proposed method.
JOURNAL OF THE MECHANICS AND PHYSICS OF SOLIDS
(2022)
Article
Biophysics
Mostafa Mahmoudi, Ali Farghadan, Daniel R. McConnell, Alex J. Barker, Jolanda J. Wentzel, Matthew J. Budoff, Amirhossein Arzani
Summary: This study explores the impact of wall shear stress on coronary artery atherosclerosis, highlighting how high shear stress can inhibit the development of atherosclerosis while low shear stress can contribute to its formation. Additionally, the role of wall shear stress on different biochemicals is complex, potentially promoting or preventing atherosclerosis.
JOURNAL OF BIOMECHANICAL ENGINEERING-TRANSACTIONS OF THE ASME
(2021)
Article
Biophysics
Mohammadreza Soltany Sadrabadi, Mohammadali Hedayat, Iman Borazjani, Amirhossein Arzani
Summary: This study utilized fluid-structure interaction simulations and mass transport models to investigate the biological transport processes related to calcification and thrombosis, revealing a close connection between vortex structures and biochemical concentration patterns, as well as the relationship between leaflet concentration and wall shear stress.
JOURNAL OF BIOMECHANICS
(2021)
Article
Engineering, Aerospace
Katherine J. Asztalos, Scott T. M. Dawson, David R. Williams
Summary: The study shows that lift response of a NACA0009 airfoil with leading-edge momentum injection is highly dependent on the state of the wake and the phase chosen for actuation. By developing a reduced-order model, it is possible to capture this phase dependency and achieve multiple control objectives with the same actuator.
Article
Physics, Fluids & Plasmas
Maysam Shamai, Scott T. M. Dawson, Igor Mezic, Beverley J. McKeon
Summary: The study investigates the wake of a streamwise-oscillating cylinder using particle image velocimetry and dynamic mode decomposition, examining the interaction of different frequencies and developing a scaling parameter to relate forced dynamics to unforced dynamics. The study shows that the transformation leads to a dynamic mode decomposition similar to that of the unforced system for the streamwise-oscillating cylinder.
PHYSICAL REVIEW FLUIDS
(2021)
Article
Engineering, Aerospace
Scott T. M. Dawson, Steven L. Brunton
Summary: The Wagner function plays a crucial role in unsteady aerodynamic methods, although explicit expressions are difficult to obtain. Approximations are commonly used for efficient computations, but they often exhibit noticeable differences from the true Wagner function, especially in long-time asymptotic behavior. A proposed alternative approximation methodology models the Wagner function as the solution of a nonlinear scalar ordinary differential equation, accurately capturing various accuracy measures.
Review
Engineering, Biomedical
Amirhossein Arzani, Jian-Xun Wang, Michael S. Sacks, Shawn C. Shadden
Summary: Recent progress in machine learning has opened up new research opportunities in cardiovascular modeling. While ML has shown promising results in patient outcomes and medical image segmentation, its application in predicting biomechanics is still in the early stages. This article discusses the challenges and potential applications of ML in cardiovascular biomechanics, emphasizing the need for clearly defined goals and careful interpretation of accuracy and speed tradeoffs.
ANNALS OF BIOMEDICAL ENGINEERING
(2022)
Article
Thermodynamics
Maryam Aliakbari, Mostafa Mahmoudi, Peter Vadasz, Amirhossein Arzani
Summary: In this paper, a multi-fidelity approach is proposed to improve the accuracy of low-fidelity computational fluid dynamics (CFD) data by combining it with physics-informed neural networks (PINN). The results show that this approach not only enhances the accuracy of the low-fidelity CFD data but also improves the convergence speed and accuracy of PINN.
INTERNATIONAL JOURNAL OF HEAT AND FLUID FLOW
(2022)
Article
Engineering, Mechanical
Xiaowei He, David R. Williams, Scott T. M. Dawson
Summary: This article presents the modification made to the Andrew Fejer Unsteady Wind Tunnel by adding a suction duct on top of the test section to generate a vertical velocity component. The results show that the transverse gusts generated in the test section are essentially irrotational, and a potential flow model provides accurate predictions of the flow field. The suction-driven approach significantly reduces the turbulence level and demonstrates high repeatability.
EXPERIMENTS IN FLUIDS
(2022)
Article
Chemistry, Physical
Mohamed Ali Saafi, Shiqi Ou, Yilan Jiang, Hailin Li, Xin He, Zhenhong Lin, Yu Gan, Zifeng Lu, Scott T. M. Dawson
Summary: China aims to achieve carbon neutrality by 2060 and plans to use hydrogen vehicle technologies to reduce greenhouse gas emissions. According to the Transportation Energy Analysis Model, hydrogen demand could reach 25% of the total transportation energy demand by 2050.
INTERNATIONAL JOURNAL OF HYDROGEN ENERGY
(2022)
Article
Computer Science, Interdisciplinary Applications
Amirhossein Arzani, Kevin W. Cassel, Roshan M. D'Souza
Summary: Physics-informed neural networks (PINNs) have become popular for scientific machine learning and differential equation modeling. This study introduces boundary-layer PINN (BL-PINN), which treats thin boundary layers as singular perturbation problems. By incorporating classical perturbation theory, BL-PINN employs different parallel PINN networks to approximate the boundary layer problem in both inner and outer regions. BL-PINN outperforms traditional PINN and other extensions such as XPINN in various benchmark problems, providing accurate solutions.
JOURNAL OF COMPUTATIONAL PHYSICS
(2023)
Article
Mechanics
Hunor Csala, Scott T. M. Dawson, Amirhossein Arzani
Summary: Computational fluid dynamics (CFD) produces high-dimensional data, and recent advances in machine learning (ML) have introduced various techniques for extracting physical information from it. This study investigates four manifold learning and two deep learning methods for nonlinear dimensionality reduction (NDR), and compares them to principal component analysis (PCA). The results suggest that using NDR methods can help build more efficient reduced-order models of fluid flows.
Article
Mechanics
Tarcisio Deda, William R. Wolf, Scott T. M. Dawson
Summary: Backpropagation of neural network models is used for controlling nonlinear dynamical systems through different approaches. Two novel approaches, neural network control (NNC) and linear control design, are presented and compared to gradient-based model predictive control (MPC). The feasibility of building surrogate models with control inputs that can learn important features is demonstrated. The proposed control approaches are tested on low-dimensional systems with stable and unstable limit cycles, chaos, as well as higher-dimensional chaotic systems and compressible Navier-Stokes equations.
THEORETICAL AND COMPUTATIONAL FLUID DYNAMICS
(2023)
Article
Engineering, Multidisciplinary
Seungik Baek, Amirhossein Arzani
Summary: Ultrasound imaging is important in detecting abdominal aortic aneurysms, but it is only recommended for men aged 65-75 with a smoking history and not recommended for women. New technologies and methods like personalized medicine and data-driven approaches have the potential to make breakthroughs in the detection of small AAAs, monitoring patients during follow-ups, predicting AAA growth, assessing rupture risk, and post-surgical prognosis for AAA patient management.
APPLICATIONS IN ENGINEERING SCIENCE
(2022)
Article
Biology
Seyyed Bahram Borgheai, Alyssa Hillary Zisk, John McLinden, James Mcintyre, Reza Sadjadi, Yalda Shahriari
Summary: This study proposed a novel personalized scheme using fNIRS and EEG as the main tools to predict and compensate for the variability in BCI systems, especially for individuals with severe motor deficits. By establishing predictive models, it was found that there were significant associations between the predicted performances and the actual performances.
COMPUTERS IN BIOLOGY AND MEDICINE
(2024)
Article
Biology
Hongliang Guo, Hanbo Liu, Ahong Zhu, Mingyang Li, Helong Yu, Yun Zhu, Xiaoxiao Chen, Yujia Xu, Lianxing Gao, Qiongying Zhang, Yangping Shentu
Summary: In this paper, a BDSMA-based image segmentation method is proposed, which improves the limitations of the original algorithm by combining SMA with DE and introducing a cooperative mixing model. The experimental results demonstrate the superiority of this method in terms of convergence speed and precision compared to other methods, and its successful application to brain tumor medical images.
COMPUTERS IN BIOLOGY AND MEDICINE
(2024)
Article
Biology
Jingfei Hu, Linwei Qiu, Hua Wang, Jicong Zhang
Summary: This study proposes a novel semi-supervised point consistency network (SPC-Net) for retinal artery/vein (A/V) classification, addressing the challenges of specific tubular structures and limited well-labeled data in CNN-based approaches. The SPC-Net combines an AVC module and an MPC module, and introduces point set representations and consistency regularization to improve the accuracy of A/V classification.
COMPUTERS IN BIOLOGY AND MEDICINE
(2024)
Article
Biology
Omair Ali, Muhammad Saif-ur-Rehman, Tobias Glasmachers, Ioannis Iossifidis, Christian Klaes
Summary: This study introduces a novel hybrid model called ConTraNet, which combines the strengths of CNN and Transformer neural networks, and achieves significant improvement in classification performance with limited training data.
COMPUTERS IN BIOLOGY AND MEDICINE
(2024)
Article
Biology
Juan Antonio Valera-Calero, Dario Lopez-Zanoni, Sandra Sanchez-Jorge, Cesar Fernandez-de-las-Penas, Marcos Jose Navarro-Santana, Sofia Olivia Calvo-Moreno, Gustavo Plaza-Manzano
Summary: This study developed an easy-to-use application for assessing the diagnostic accuracy of digital pain drawings (PDs) compared to the classic paper-and-pencil method. The results demonstrated that digital PDs have higher reliability and accuracy compared to paper-and-pencil PDs, and there were no significant differences in assessing pain extent between the two methods. The PAIN EXTENT app showed good convergent validity.
COMPUTERS IN BIOLOGY AND MEDICINE
(2024)
Article
Biology
Biao Qu, Jialue Zhang, Taishan Kang, Jianzhong Lin, Meijin Lin, Huajun She, Qingxia Wu, Meiyun Wang, Gaofeng Zheng
Summary: This study proposes a deep unrolled neural network, pFISTA-DR, for radial MRI image reconstruction, which successfully preserves image details using a preprocessing module, learnable convolution filters, and adaptive threshold.
COMPUTERS IN BIOLOGY AND MEDICINE
(2024)
Article
Biology
Alireza Rafiei, Milad Ghiasi Rad, Andrea Sikora, Rishikesan Kamaleswaran
Summary: This study aimed to improve machine learning model prediction of fluid overload by integrating synthetic data, which could be translated to other clinical outcomes.
COMPUTERS IN BIOLOGY AND MEDICINE
(2024)
Article
Biology
Jinlian Ma, Dexing Kong, Fa Wu, Lingyun Bao, Jing Yuan, Yusheng Liu
Summary: In this study, a new method based on MDenseNet is proposed for automatic segmentation of nodular lesions from ultrasound images. Experimental results demonstrate that the proposed method can accurately extract multiple nodules from thyroid and breast ultrasound images, with good accuracy and reproducibility, and it shows great potential in other clinical segmentation tasks.
COMPUTERS IN BIOLOGY AND MEDICINE
(2024)
Article
Biology
Jiabao Sheng, SaiKit Lam, Jiang Zhang, Yuanpeng Zhang, Jing Cai
Summary: Omics fusion is an important preprocessing approach in medical image processing that assists in various studies. This study aims to develop a fusion methodology for predicting distant metastasis in nasopharyngeal carcinoma by mitigating the disparities in omics data and utilizing a label-softening technique and a multi-kernel-based neural network.
COMPUTERS IN BIOLOGY AND MEDICINE
(2024)
Article
Biology
Zhenxiang Xiao, Liang He, Boyu Zhao, Mingxin Jiang, Wei Mao, Yuzhong Chen, Tuo Zhang, Xintao Hu, Tianming Liu, Xi Jiang
Summary: This study systematically investigates the functional connectivity characteristics between gyri and sulci in the human brain under naturalistic stimulus, and identifies unique features in these connections. This research provides novel insights into the functional brain mechanism under naturalistic stimulus and lays a solid foundation for accurately mapping the brain anatomy-function relationship.
COMPUTERS IN BIOLOGY AND MEDICINE
(2024)
Article
Biology
Qianqian Wang, Mingyu Zhang, Aohan Li, Xiaojun Yao, Yingqing Chen
Summary: The development of PARP-1 inhibitors is crucial for the treatment of various cancers. This study investigates the structural regulation of PARP-1 by different allosteric inhibitors, revealing the basis of allosteric inhibition and providing guidance for the discovery of more innovative PARP-1 inhibitors.
COMPUTERS IN BIOLOGY AND MEDICINE
(2024)
Article
Biology
Qing Xu, Wenting Duan
Summary: In this paper, a dual attention supervised module, named DualAttNet, is proposed for multi-label lesion detection in chest radiographs. By efficiently fusing global and local lesion classification information, the module is able to recognize targets with different sizes. Experimental results show that DualAttNet outperforms baselines in terms of mAP and AP50 with different detection architectures.
COMPUTERS IN BIOLOGY AND MEDICINE
(2024)
Article
Biology
Kaja Gutowska, Piotr Formanowicz
Summary: The primary aim of this research is to propose algorithms for identifying significant reactions and subprocesses within biological system models constructed using classical Petri nets. These solutions enable two analysis methods: importance analysis for identifying critical individual reactions to the model's functionality and occurrence analysis for finding essential subprocesses. The utility of these methods has been demonstrated through analyses of an example model related to the DNA damage response mechanism. It should be noted that these proposed analyses can be applied to any biological phenomenon represented using the Petri net formalism. The presented analysis methods extend classical Petri net-based analyses, enhancing our comprehension of the investigated biological phenomena and aiding in the identification of potential molecular targets for drugs.
COMPUTERS IN BIOLOGY AND MEDICINE
(2024)
Article
Biology
Hansle Gwon, Imjin Ahn, Yunha Kim, Hee Jun Kang, Hyeram Seo, Heejung Choi, Ha Na Cho, Minkyoung Kim, Jiye Han, Gaeun Kee, Seohyun Park, Kye Hwa Lee, Tae Joon Jun, Young-Hak Kim
Summary: Electronic medical records have potential in advancing healthcare technologies, but privacy issues hinder their full utilization. Deep learning-based generative models can mitigate this problem by creating synthetic data similar to real patient data. However, the risk of data leakage due to malicious attacks poses a challenge to traditional generative models. To address this, we propose a method that employs local differential privacy (LDP) to protect the model from attacks and preserve the privacy of training data, while generating medical data with reasonable performance.
COMPUTERS IN BIOLOGY AND MEDICINE
(2024)
Article
Biology
Siwei Tao, Zonghan Tian, Ling Bai, Yueshu Xu, Cuifang Kuang, Xu Liu
Summary: This study proposes a transfer learning-based method to address the phase retrieval problem in grating-based X-ray phase contrast imaging. By generating a training dataset and using deep learning techniques, this method improves image quality and can be applied to X-ray 2D and 3D imaging.
COMPUTERS IN BIOLOGY AND MEDICINE
(2024)