Data Ecosystems for Scientific Experiments: Managing Combustion Experiments and Simulation Analyses in Chemical Engineering

Abstract

The development of scientific predictive models has been of great interest over the decades. A scientific model is capable of forecasting domain outcomes without the necessity of performing expensive experiments. In particular, in combustion kinetics, the model can help improving the combustion facilities and the fuel efficiency reducing the pollutants. At the same time, the amount of available scientific data has increased and helped speeding up the continuous cycle of model improvement and validation. This has also opened new opportunities for leveraging a large amount of data to support knowledge extraction. However, experiments are affected by several data quality problems since they are a collection of information over several decades of research, each characterized by different representation formats and reasons of uncertainty. In this context, it is necessary to develop an automatic data ecosystem capable of integrating heterogeneous information sources while maintaining a quality repository. We present an innovative approach to data quality management from the chemical engineering domain, based on an available prototype of a scientific framework, SciExpeM, which has been significantly extended. We identified a new methodology from the model development research process that systematically extracts knowledge from the experimental data and the predictive model. In the paper, we show how our general framework could support the model development process, and save precious research time also in other experimental domains with similar characteristics, i.e., managing numerical data from experiments.

Keywords: combustion kinetics; data management; data quality; data validation; experiments management; scientific experiments; scientific model development; simulation analysis.

FIGURE 1

Continuous-improvement process of the CRECK predictive kinetic model - Ranzi et al. (2012).

FIGURE 2

A sketch of the logical workflow of the development process of a scientific model that involves the continuous interaction with experiments. Filled in grey are the steps that are not a subject of the SciExpeM framework but are part of the overall process.

FIGURE 3

Sketch of the architecture adapted from the NIST Big Data Reference Architecture (cfr. Chang et al. (2019)) for the management of experiments to support the development of scientific models.

FIGURE 4

Class diagram of the database used in SciExpeM to represent experiment, simulation, models, and analyses (PK) stands for primary key. Some fields of the entities can be omitted for simplicity.

FIGURE 5

Main experiment related services of SciExpeM system with their dependencies.

FIGURE 6

Business Process Model and Notation (BPMN) for the request of an experiment simulation. We can observe the interaction of the various SciExpeM services to fulfill the requirements about the reuse of resources and a non blocking request.

FIGURE 7

Combustion kinetics model comparison for an experiment (cfr. Chaumeix et al. (2007)) using Curve Matching as similarity score.

FIGURE 8

Heatmap of the average Curve Matching scores for each combination of species-model. Higher is better.

FIGURE 9

Heatmap of the minimum Curve Matching scores for each combination of species-model. Higher is better.

FIGURE 10

Heatmap of the Curve Matching scores against three combustion kinetic models regarding a collection of Methanol experiments.