Proposed thesis on data driven modeling of conversion units in refineries
Thesis title “Data driven modeling of conversion units’ product properties for control applications”
In cooperation with Helleniq Energy Aspropyrgos Refinery.
Description
The goal of this project is to utilize advanced data driven techniques in order to predict, and ultimately control, key properties of the main product streams of conversion units that focus on gasoline production. Despite the growing shift towards renewable energy alternatives, gasoline is still the dominant fuel for transportation, accounting for more than 40% in the fuel mix [1]. Gasoline is made up of various components. Two of the most important intermediates of gasoline production are isomerate and reformate. Reformate is produced from a Continuous Catalytic Reforming (CCR) unit, which is based on a process that converts low-octane (heavy) naphtha into high-octane reformate [2]. The main goal is to rearrange hydrocarbon molecules using a catalyst. The process operates continuously and the catalyst is regenerated on-site to sustain efficiency and product yield. The CCR unit not only increases gasoline production but also produces valuable by-products like hydrogen, used in other refining processes. Isomerate is produced in an isomerization unit [2]. This unit is charged with (light) naphtha of low-octane number and transforms it into their higher-octane isomers, like isobutane or isopentane, which are valuable for blending into gasoline to improve its octane rating. The process typically involves using a catalyst in the presence of hydrogen to rearrange the molecular structure of the hydrocarbons. The proposed project aims to construct data-driven models that are able to predict the key properties of reformate and/or isomerate products, as well as their yield. The inputs to these models will be continuously measured process variables, lab data (infrequent discrete measurements), as well as data from upstream units. Physics informed Neural Networks (PINNs), pattern recognition methodologies appropriate filtering techniques and possibly uncertainty quantification will be used to develop reliable hybrid data driven models o the real-world process. The project will be conducted in collaboration with Helleniq Energy Aspropyrgos Refinery. It will make use of real-world data, which are notorious for noise, measurement errors, discontinuities, etc. Challenges arise from the fact that data is not typically at steady-state, the process incorporates various (unknown) delays, the variability of the data is limited to the ufdvsdgsnit operating window, and that the model inputs are different in nature and sampling times. Ultimately (and beyond the scope of the current project) successful hybrid models will be incorporated in the Model Predictive Control application of the unit to extend its predictive capability and therefore enhance control performance.
References
1. European Commission: Directorate-General for Mobility and Transport, Transport in the European Union – Current trends and issues, Publications Office of the European Union, 2024, https://data.europa.eu/doi/10.2832/131741
2. Jones, D. S., & Pujadó, P. P. (Eds.). (2006)
Interested students should contact Professor Kostas Theodoropoulos, University of Manchester, UK, visiting Professor at the NTUA School of Chemical Engineering; also Professor Andreas Boudouvis.