American Institute of Aeronautics and Astronautics 1 Investigation of Dataset Construction Parameters and their Impact on Reaction Model Optimization using PrIMe A. Mirzayeva 1 , N.A. Slavinskaya 2 , U. Riedel 3 Institute of Combustion Technology, German Aerospace Center (DLR), Stuttgart, 70569, Germany M. Frenklach 4 , A. Packard 5 , W. Li 6 , J. Oreluk 7 , A. Hegde 8 Department of Mechanical Engineering, University of California, Berkeley, CA 94720, USA The current paper presents a continuation of the development of a modern methodology for the construction of uncertainty-quantified chemical reaction models on the base of the Bound-to-Bound Data Collaboration (B2BDC) module of the automated data-centric infrastructure PrIMe. Some problems, postulated in the recent studies, are in the focus of the present investigation. The question of targets amount (experimental data, Quantities of Interest (QoI)) selected for the analysis has been studied. To investigate this, the PrIMe dataset is augmented. The influence of dataset extension on the dataset consistency, feasible parameter set, and model optimization is studied and an algorithm for the selection of QoI in each experimental set is postulated. The approach of combined methods of scalar consistency measure, SCM, and vector consistency measure, VCM, for consistency analysis are adapted and successfully implemented. Predictions of the LS-  optimized mechanism are compared against a wide range of experimental data of laminar premixed flames and shock tube ignition delay times. Good agreement of model predictions with the experimental measurements is obtained. Nomenclature ϕ = equivalence ratio P 5 = pressure behind reflected shock waves in shock-tube experiments QoI = quantity of interest T 5 = temperature behind reflected shock waves in shock-tube experiments T 0 = initial temperature in laminar flame experiments UB = uncertainty bounds I. Introduction NE of the most important properties of a reaction model in chemical engineering is to make predictions about the system when certain settings are changed. Appropriate handling and archiving of experimental data from different sources, and of the many uncertainties in the data embedded in the kinetic models, is a major challenge the chemical kinetics community has to tackle before becoming a predictive science. Methods for determining whether or not the model predictions are consistent with experimental data have been of great interest in combustion research over decades. Developing predictive models 1 has become the goal in many of the modelling studies on reaction systems. Numerical optimization of complex reaction networks, of the kind that guided the development of GRI- Mech, 1-3 e.g., has now been accepted as one of the underlying methods in this pursuit. 4-6 In order to develop a 1 PhD Student, Chemical Kinetics Department, Aziza.Mirzayeva@dlr.de 2 Senior research fellow, Chemical Kinetics Department, Nadja.Slavinskaya@dlr.de, AIAA Senior Member 3 Head of Chemical Kinetics Department, Uwe.Riedel@dlr.de, AIAA Senior Member 4 Prof. of University of California, Berkeley, frenklach@berkeley.edu, AIAA Member 5 Prof. of University of California, Berkeley, apackard@berkeley.edu. 6 PhD Student of University of California, Berkeley, wenyuli@berkeley.edu. 7 PhD Student of University of California, Berkeley, jim.oreluk@berkeley.edu. 8 PhD Student of University of California, Berkeley, arun.hegde@berkeley.edu. O Downloaded by UNIV CALIFORNIA BERKELEY on January 9, 2018 | http://arc.aiaa.org | DOI: 10.2514/6.2018-0143 2018 AIAA Aerospace Sciences Meeting 8–12 January 2018, Kissimmee, Florida 10.2514/6.2018-0143 Copyright © 2018 by the American Institute of Aeronautics and Astronautics, Inc. All rights reserved. AIAA SciTech Forum