462943 Data Auditing: Analysis Techniques for the Compatibility of Experimental Data for Molecular-Level Kinetic Models

Wednesday, November 16, 2016
Grand Ballroom B (Hilton San Francisco Union Square)
Scott R. Horton1, Triveni Billa1, Pratyush Agarwal1, Juan Lucio-Vega1, Mayuresh Sahasrabudhe2, Chandra Saravanan2 and Michael T. Klein1, (1)Chemical and Biomolecular Engineering and Energy Institute University of Delaware, University of Delaware, Newark, DE, (2)Reliance Industries, Navi Mumbai, India

Data Auditing: Analysis techniques for the compatibility of experimental data for molecular-level kinetic models

Scott R. Horton*, Triveni Billa, Pratyush Agarwal, Juan Lucio, and Michael T. Klein

Department of Chemical & Biomolecular Engineering, University of Delaware


Mayuresh Sahasrabudhe and Chandra Saravanan (Saru)

Reliance Industries

In modeling industrial reactors, vast quantities of data are often available for years of reactor operation. These data allow for the evaluation and modeling of reactor performance through catalyst deactivation and feedstock variation. The sheer quantity of data also gives the motivation for automated methods of analysis of the data and their utility in detailed kinetic models. In this work, automated methods are developed and applied to five years of industrial data on hydrocarbon processing to select data sets for model evaluation and parameter optimization.

A software tool has been developed to perform an ‘audit’ of the data. First, the software searches for incompatibilities and outliers within the entire set of data. This allows for the identification of erroneous data entry or equipment failure. For models where solution time is an issue, the user might want to tune parameters on a sub-set of the entire data sets. To this end, a second feature of the app is the ability to find representative data sets within the full set of data. To select these data sets, the algorithm analyzes the measurable inputs to find data sets which best span the range of possible experimental values. This allows the user to tune kinetic parameters to the full range of possible data with a limited number of data sets.

The final set of features in this app analyze the compatibility of the experimental data with the reaction network in the molecular-level kinetic model. For instance, this app, without kinetic parameter tuning, will inform the user if the output measurements are possible given the input measurements and reaction network. In the case where the analysis returns red flags, the user can then analyze the structure of the reaction network and the validity of the experimental measurements. Because this analysis is performed before tuning, incompatibilities of data and reaction network can be discovered and remedied in moments rather than days-weeks.

Extended Abstract: File Not Uploaded