Hydrogen is a good candidate as an alternative fuel for power generation, due to the absence of carbon. Retrofitting available combustion systems to run hydrogen or its blends with natural gas is an effective option t...
详细信息
ISBN:
(纸本)9780791887035
Hydrogen is a good candidate as an alternative fuel for power generation, due to the absence of carbon. Retrofitting available combustion systems to run hydrogen or its blends with natural gas is an effective option to decarbonize this sector. This demands a reassessment of the system response for the entire range of operation. Experimentation can be expensive and prohibitive from safety standpoints, and though computational models can provide data to fill the gaps, they come with their own sets of challenges. data-drivensurrogates use machine learning methods to learn from the sparse available data and can fill this knowledge gap by learning statistical correlations that describe the system. In this paper, a data-drivensurrogate model is developed for a Capstone C65 microturbine combustor that is modified to burn pure natural gas and blends with up to 60% hydrogen for various power demands. Gaussian process (GP) regression modeling is used to learn from this dataset to emulate the system characteristics. Active learning is also invoked to learn a good model using as few data points as possible. Additional data is, then, generated using Reynolds-averaged Navier-Stokes (RANS) simulations of the same combustor geometry across a wider range of operating conditions i.e., 0-100% hydrogen for 0-65kW power loads. This provides low-fidelity information that can be included in a multi-fidelity learning setting which shows distinct improvements over the single-fidelity model. The novel multi-input-multi-output multi-fidelity GP surrogatemodeling framework used in this work, is developed in-house using GPyTorch, which runs on a PyTorch backend, and is capable of superior scaling compared to traditional GP approaches.
This paper presents a data-driven surrogate modeling methodology for computational fluid dynamics (CFD) to deliver a more accurate and efficient modeling for simulating segmented simulating spray fluidized bed granula...
详细信息
ISBN:
(纸本)9798350334722
This paper presents a data-driven surrogate modeling methodology for computational fluid dynamics (CFD) to deliver a more accurate and efficient modeling for simulating segmented simulating spray fluidized bed granulation (SFBG) which fully considers the fluidization dynamics of being in actual production. Specifically, a data-drivensurrogate model is developed for replacing the actual CFD simulations to calculate the time-varying fluidization parameters, and then integrating with a two-compartmental population balance model (TCPBM) to formulate a surrogate-assisted CFD-TCPBM coupled simulation modeling framework. The proposed surrogate-assisted methodology is illustrated by its application to a case study, in which the simulation results indicate the feasibility and effectiveness of surrogate-assisted CFD in simulating SFBG compared to actual CFD simulations. Therefore, one can conclude that the proposed methodology has great potential to deliver more accurate and realistic results in simulating SFBG of being in actual production.
Rapid and accurate system evolution predictions are crucial in scientific and engineering research. However, the complexity of processing systems, involving multiple physical field couplings and slow convergence of it...
详细信息
Rapid and accurate system evolution predictions are crucial in scientific and engineering research. However, the complexity of processing systems, involving multiple physical field couplings and slow convergence of iterative numerical algorithms, leads to low computational efficiency. Hence, this paper introduces a systematic deep-learning-based surrogatemodeling methodology for multi-physics-coupled process systems with limited data and long-range time evolution, accurately predicting physics dynamics and considerably improving computational efficiency and generalization. The methodology comprises three main components: (1) generating datasets using a sequential sampling strategy, (2) modeling multi-physics spatio-temporal dynamics by designing a heterogeneous Convolutional Autoencoder and Recurrent Neural Network, and (3) training high-precision models with limited data and long-range time evolution via a dual-phase training strategy. A holistic evaluation using a 2D low-temperature plasma processing example demonstrates the methodology's superior computational efficiency, accuracy, and generalization capabilities. It predicts plasma dynamics approximately 105 5 times faster than traditional numerical solvers, with a consistent 2% relative error across different generalization tasks. Furthermore, the potential for transferability across various geometries is explored, and the model's transfer capability is demonstrated with two distinct geometric domain examples.
When repeated evaluations for varying parameter configurations of a high-fidelity physical model are required, surrogatemodeling techniques based on model order reduction are desirable. In absence of the governing eq...
详细信息
When repeated evaluations for varying parameter configurations of a high-fidelity physical model are required, surrogatemodeling techniques based on model order reduction are desirable. In absence of the governing equations describing the dynamics, we need to construct the parametric reduced-order surrogate model in a non-intrusive fashion. In this setting, the usual residual-based error estimate for optimal parameter sampling associated with the reduced basis method is not directly available. Our work provides a non-intrusive error-estimator-based optimality criterion to efficiently populate the parameter snapshots, thereby, enabling us to effectively construct a parametric surrogate model. We consider parameter-specific proper orthogonal decomposition subspaces and propose an active-learning-drivensurrogate model using kernel-based shallow neural networks (KSNNs), abbreviated as ActLearn-POD-KSNN surrogate model. The center location for each kernel, along with center-dependent kernel widths, can be learned for the KSNN by using an alternating dual-staged iterative training procedure. To demonstrate the efficiency of our proposed ideas, we present numerical experiments using four physical models, including incompressible Navier-Stokes equations. The ActLearn-POD-KSNN surrogate model efficiently predicts the solution at new parameter locations, even for a setting with multiple interacting shock profiles and a fluid flow scenario with Hopf bifurcation. We also provide an investigation of the surrogate's performance when the available data is noisy.
暂无评论