Replication Data for: Closure Law Model Uncertainty Quantification

The prediction uncertainty in simulators for industrial processes is due to uncertainties in the input variables and uncertainties in specification of the models, in particular the closure laws. In this work, the uncertainty in each closure law was modeled as a random variable and the parameters of its distribution were optimized to correctly quantify the uncertainty in predictions. We have developed two methods for optimization, based on the integrated quadratic distance and the energy score. The proposed methods were applied to the commercial multiphase flow simulator LedaFlow with the liquid volume fraction and pressure gradient as output variables. Two datasets were analyzed. Both describe two-phase gas-liquid flow, but are otherwise fundamentally different. One is gas-dominated stratified/annular flow and the other is liquid-dominated slug flow.

工业过程模拟器的预测不确定性，源于输入变量的不确定性以及模型的规格设定（尤其是封闭律（closure laws））的不确定性。本研究将每条封闭律的不确定性建模为随机变量，并对其分布参数进行优化，以准确量化预测结果的不确定性。我们基于积分二次距离与能量得分，开发了两种优化方法。所提方法被应用于商用多相流模拟器LedaFlow，以液体体积分数与压力梯度作为输出变量。本次研究分析了两个数据集，二者均描述气液两相流，但本质上存在根本差异：其一为气相主导的分层/环状流，其二为液相主导的段塞流。