Development of an Uncertainty Quantification Predictive Chemical Reaction Model for Syngas Combustion

Abstract

An automated data-centric infrastructure, Process Informatics Model (PrIMe), was applied to validation and optimization of a syngas combustion model. The Bound-to-Bound Data Collaboration (B2BDC) module of PrIMe was employed to discover the limits of parameter modifications based on uncertainty quantification (UQ) and consistency analysis of the model−data system and experimental data, including shock-tube ignition delay times and laminar flame speeds. Existing syngas reaction models are reviewed, and the selected kinetic data are described in detail. Empirical rules were developed and applied to evaluate the uncertainty bounds of the literature experimental data. The initial H 2 /CO reaction model, assembled from 73 reactions and 17 species, was subjected to a B2BDC analysis. For this purpose, a dataset was constructed that included a total of 167 experimental targets and 55 active model parameters. Consistency analysis of the composed dataset revealed disagreement between models and data. Further analysis suggested that removing 45 experimental targets, 8 of which were self-inconsistent, would lead to a consistent dataset. This dataset was subjected to a correlation analysis, which highlights possible directions for parameter modification and model improvement. Additionally, several methods of parameter optimization were applied, some of them unique to the B2BDC framework. The optimized models demonstrated improved agreement with experiments compared to the initially assembled model, and their predictions for experiments not included in the initial dataset (i.e., a blind prediction) were investigated. The results demonstrate benefits of applying the B2BDC methodology for developing predictive kinetic models.

Consistency Analysis for Massively Inconsistent Datasets in Bound-to-Bound Data Collaboration

“…The resulting dataset is comprised of 159 model-data constraints in 55 uncertain parameters. The experimental uncertainty came in the form of expert-assessed bounds [44].…”

“…In Feeley et al [19], model inadequacy was primarily attributed to the parameters and certain experimental bounds were revised. In Slavinskaya et al [44], the (temporary) removal of QOIs was warranted due to perceived deficiencies in their instrument models rather than the underlying reaction model. In Iavarone et al [28], a new model form consistent with experimental data was proposed to replace a previously inconsistent model.…”

Investigation of Dataset Construction Parameters and their Impact on Reaction Model Optimization using PrIMe

“…In our recent research, 10 the initial H 2 /CO reaction model, consisting of 73 reactions and 17 species, was subjected to a B2BDC analysis. A complete dataset including a total of 167 experimental targets (122 ignition delays, and 45 laminar flame speeds) and 55 active model parameters was constructed.…”

“…At the moment, it is unfortunately not guaranteed that a huge amount of data leads to a perfect well-predictive model, partly because the uncertainty prediction cannot be measured. In this work, we have augmented the dataset through a higher number of QoIs for syngas combustion by addition of more targets for already implemented 10 and newly published experimental observation sets. The selection criterion is the minimum amount of units sufficient to reflect the experimental trend.…”

“…Experimental conditions of the selected measurements are summarized in Table 1. The uncertainty boundaries of the selected QoIs are assessed by evaluation with the empirical rule described in Slavinskaya et al 10 The final number of QoIs in the complete is 477 QoIs, including 319 ignition delays and 158 laminar flame speeds. The selected QoIs of ignition delay times and laminar flame speeds along with their evaluated uncertainties can be found in Appendixes 1-2.…”