Predicting the performance uncertainty of a 1-MW pilot-scale carbon capture system after hierarchical laboratory-scale calibration and validation

Abstract

A challenging problem in designing pilot-scale carbon capture systems is to predict, with uncertainty, the adsorber performance and capture efficiency under various operating conditions where no direct experimental data exist. Motivated by this challenge, we previously proposed a hierarchical framework where relevant parameters of physical models were sequentially calibrated from different laboratory-scale carbon capture unit (C2U) experiments. Specifically, three models of increasing complexity were identified based on the fundamental physical and chemical processes of the sorbent-based carbon capture technology. Results from the corresponding laboratory experiments were used to statistically calibrate the physical model parameters while quantifying some of their inherent uncertainty.In this study, the parameter distributions obtained from laboratory-scale C2U calibration runs are used to facilitate prediction at a larger scale with no corresponding experimental results available. Initially, we describe the multiphase reactive flow model for a sorbent-based 1megawatt carbon capture system then analyze results from an ensemble of simulations with the upscaled model. The simulation results are used to quantify uncertainty regarding the

Sub-grid models for heat transfer in gas-particle flows with immersed horizontal cylinders

“…Additionally, the model is currently being used to simulate a conceptual 1 MWe pilot-scale solid-sorbent carbon-capture reactor [32]; an otherwise computationally intractable system. Preliminary results show great mixing and heat regulation within the system and fair agreement with a previously developed 1D process-level model.…”

Parameter estimation and prediction uncertainties for multi-response kinetic models with uncertain inputs

“…Past researchers have sampled from the distribution of the parameters and inputs simultaneously to propagate the corresponding uncertainties into the model predictions. [10][11][12][13][14][15][16][17][18][19][20] Table 2 provides an algorithm for quantifying uncertainties in model predictions using a cloud of 𝑏 𝑚𝑎𝑥 parameter estimates available from bootstrapping. This algorithm quantifies the uncertainties in a vector of model predictions 𝒚 ̂𝒄 at an experimental condition of interest 𝑐.…”

Fast Stencil-Code Computation on a Wafer-Scale Processor

“…Uncertainty quantification (UQ) relies on the assessment of the effect of model and parameter uncertainty, usually requiring hundreds or thousands of simulations to sweep relevant ranges of parameters. For example, to predict the performance uncertainty of a 1MW pilot scale carbon capture system providing 90% capture efficiency with 95% confidence [18], a total of 1,505 simulations were run, each needing about 600 seconds of simulation to reach steady state. This kind of simulation campaign is currently achievable only by simplifying the geometry (two-dimensional geometry), and using filter models to account for the coarse meshes.…”