56016 research outputs found
Sort by
The Development of Na Metal Electrode-based Desalination Batteries
No full textSchool of Energy and Chemical Engineering (Energy Engineering (Battery Science and Technology))clos
????????????????????? ???????????? ?????? ?????????-????????? ???????????? ?????? ????????? ?????????????????? ?????? ??????
No full textSchool of Energy and Chemical Engineering (Energy Engineering (Battery Science and Technology))clos
????????? ?????? ????????? ?????? ?????? ???????????? ????????? ????????? ?????? ?????? ????????? ?????? ?????????????????? ????????????
No full textSchool of Energy and Chemical Engineering (Energy Engineering (Battery Science and Technology))clos
??????????????? ????????????????????? ??????????????? ???????????? ??? ????????? ??????????????? ?????????
No full textSchool of Energy and Chemical Engineering (Chemical Engineering)clos
Scalable Integrated Photoelectrochemical Water Splitting Systems for Unassisted Solar H2 Production
No full textSchool of Energy and Chemical Engineering (Chemical Engineering)clos
???????????? ????????? ????????? ??????????????? ??????????????? ??????????????? ??????
No full textSchool of Energy and Chemical Engineering (Chemical Engineering)clos
Development of Reactor Multiphysics framework to analyze the effect of crossflow and dynamic gHTC for Depletion and REA transient
Get PDFDepartment of Nuclear EngineeringThe aim of this research is to create a Multiphysics coupling framework called MPCORE (Multi-Physics CORE) to analyze the behavior of nuclear reactors. This framework couples fuel performance (FP) with neutron kinetics (NK) and thermal hydraulics (TH) modules for depletion and transient analysis. Coupling the FP code allows for accurate modeling of dynamic gap heat transfer for each pin. Converging all modules together provides a more meaningful insight into the variation of reactor parameters. Depletion studies with Multiphysics parameters are essential to understand safety parameters throughout a nuclear reactor's life. The study investigates the passive response of the reactor core to reactivity insertions caused by rod ejection accidents (REA). Most coupling frameworks only couple NK with TH, but this research also includes FP and uses two-way coupling between TH and FP modules to examine the impact on critical safety parameters.
The adaptive time-step feature of MPCORE reduces execution time, and the framework performs in-memory data transfer between modules. Verification and validation work for MPCORE coupled modules (RAST-K for NK, CTH1D/CTF for TH, and FRAPI for FP) has been performed for single assembly, 3x3 mini-core, and whole-core problems. The performance of the TH module is evaluated with and without crossflow for transient calculations in whole-core problems. The effect of dynamic and static gap heat transfer coefficient models on the FP module is quantified for assembly, mini-core, and whole-core transient problems. Difference between one-way and two-way coupling between FP and TH modules is quantified for whole-core depletion problems.
The study compares safety parameters such as departure from nucleate boiling ratio, linear power, fuel enthalpy, fuel centerline temperature, cladding outer surface temperature, coolant temperature, and cladding hydrogen concentration for different models. A best-estimate coupling framework has been developed and tested for uncertainty quantification (UQ) studies for assembly and mini-core problems. Random sampling and Latin hypercube sampling options are available for UQ studies in MPCORE. Standard deviation of different parameters in case of dynamic gap conductance has increased due to the difference of gap heat transfer in different cases.clos
Score-based Generative Models with L??vy Processes
Get PDFGraduate School of Artificial IntelligenceInvestigating the optimal stochastic process beyond Gaussian for noise injection in a score-based generative model remains an open question. The commonly used Brownian motion, which is a light- tailed process with continuous paths, suffers from slow convergence rates in terms of the Number of Function Evaluation (NFE). Additionally, diffusion models based on Brownian motion often encounter mode-collapse issues when dealing with imbalanced data.
Recent studies derive exact reverse-time stochastic differential equation driven by the L??vy process and propose the corresponding fractional denoising score matching. However, the performance of the model in both unconditional and conditional generation tasks is insufficient, and there is a lack of mo- tivation to address these limitations. To alleviate the limitations and investigate motivation, I improve a previous score-based generative model, referred to as L??vy-Ito ?? Model (LIM). LIM utilizes isotropic a- stable L??vy processes, which are heavy-tailed in nature. By leveraging the heavy-tailed properties of the L??vy process, LIM enables faster and more diverse sampling while maintaining high fidelity compared to existing diffusion models.
Experimental results demonstrate the effectiveness of LIM across various image datasets, including CIFAR10, CelebA, and the imbalanced dataset CIFAR10LT. While DDPM achieves Fr??chet Inception Distance (FID) of 3.21 and a Recall of 0.6437 in unconditional generation on the CelebA dataset, LIM achieves significantly better results with a FID of 1.58 and a Recall of 0.7006 even using the same architecture. Furthermore, the best performance of LIM shows 2 times faster total wall-clock time than the baseline by using half of NFE(500).
Overall, the introduction of the LIM presents a promising approach to address the limitations of Gaussian-based noise injection in score-based generative models, demonstrating improved performance in terms of mode estimation, sampling diversity, and convergence rate for NFE.clos
A Bipolar Impedance Plethysmography (IPG) Sensor for Non-invasive and Continuous Blood Pressure Monitoring with a Distensibility Correction Technique (DCT)
No full textDepartment of Electrical Engineeringclos
Learning Spherical Deformation with Guided-Attention for Sulcal Labeling
No full textDepartment of Computer Science and Engineeringclos