Two different approaches for RDC modelling when simulating a solvent deasphalting plant

• Type: Article in International Scientific Journal
• Date: 2002

Authors

aparicio, jjc (Author) Other; jeronimo, mas (Author) Other; martins, fg (Author) FEUP; coelho, ma (Author) FEUP; martins, c (Author) Other; braga, as (Author) Other; costa, ca (Author) FEUP

Journal

Title: Computers and Chemical Engineering

Vol. 26

Pages: 1369-1377

ISSN: 0098-1354

Publisher: Elsevier

Indexing

ISI Web of Knowledge - 8 Citations

Scopus - 12 Citations

Scientific classification

FOS: Natural sciences > Computer and information sciences

Other information

Authenticus ID: P-000-MP4

DOI: 10.1016/s0098-1354(02)00114-x

Abstract (EN): Due to the complexity of hydrodynamics and mass transfer phenomena involved in rotating disk contactors, plus the lack of accurate information of crude oil composition, simulation of a propane deasphalting plant in a common process simulator can be a very difficult task. Two different approaches are presented in this work, aiming the development of a reliable simulation model for propane deasphalting rotating disc contactor in a chemical process simulator. The first approach consists in the development of a black box-type model, based on feedforward artificial neural networks (FANN) to predict the RDC performance. The second one involves a recharacterisation of vacuum residue feed and a simulation of the RDC with the process simulator algorithms. Comparing the results for both approaches with real operating data, a maximum relative error of 9% was found in the FANN predictions and a maximum relative error of 4.5% was found in the second approach, showing that both have good predictive performances. For this particular case and concerning the objectives of this research, the K-UOP methodology, based on fundamental chemistry, is more suitable than FANN approach because it can be applied in a wider range of operating conditions of the extraction unit. However, in similar extraction problems with a wider range of process variables as training data set, the FANN could be an alternative methodology due to its simplicity and easy use in simulation processes.

Language: English

Type (Professor's evaluation): Scientific

No. of pages: 9