QUANTIS: Data quality assessment tool by clustering analysis

Abstract

Automatically generated kinetic networks are ideally validated against a large set of accurate, reproducible, and easy‐to‐model experimental data. However, although this might seem simple, it proves to be quite challenging. QUANTIS, a publicly available Python package, is specifically developed to evaluate both the precision and accuracy of experimental data and to ensure a uniform, quick processing, and storage strategy that enables automated comparison of developed kinetic models. The precision is investigated with two clustering techniques, PCA and t‐SNE, whereas the accuracy is probed with checks for the conservation laws. First, the developed tool processes, evaluates, and stores experimental yield data automatically. All data belonging to a given experiment, both unprocessed and processed, are stored in the form of an HDF5 container. The demonstration of QUANTIS on three different pyrolysis cases showed that it can help in identifying and overcoming instabilities in experimental datasets, reduce mass and molar balance closure discrepancies, and, by evaluating the visualized correlation matrices, increase understanding in the underlying reaction pathways. Inclusion of all experimental data in the HDF5 file makes it possible to automate simulating the experiment with CHEMKIN. Because of the employed InChI string identifiers for molecules, it is possible to automate the comparison experiment/simulation. QUANTIS and the concepts demonstrated therein is a potentially useful tool for data quality assessment, kinetic model validation, and refinement.

The merit of pressure dependent kinetic modelling in steam cracking

“…Principal component analysis (PCA) is performed on this data using an in-house developed data analysis tool, Quantis. 31 After PCA, outliers for each of the conditions are removed and the remaining product yields are averaged for each of the conditions. The outliers identified by PCA are generally the first injection after a new temperature profile is set and thus it can be assumed that these outliers occur because steady state operation has not been fully reached.…”

Active learning-based exploration of the catalytic pyrolysis of plastic waste

“…Based on the FID response for the internal standard, the yields of all other products were calculated by using the molar response factor (MRF) approach [73]. All reported yields were normalized via the experimental postprocessing tool QUANTIS [74].…”

Active Machine Learning for Chemical Engineers: a Bright Future Lies Ahead!

“…This makes the post-processing of the experiments less trivial as the effect of the factors is not isolated. As a result, a statistical analysis is required to draw conclusions from the experimental campaign [52]. These tools are incorporated in regular DoE software but not in the active machine learning packages that are available as of today.…”