Banca de DEFESA: VINICIUS VIENA SANTANA

Uma banca de DEFESA de MESTRADO foi cadastrada pelo programa.
STUDENT : VINICIUS VIENA SANTANA
DATE: 29/08/2022
TIME: 09:00
LOCAL: videoconferência na plataforma RNP (sala PEI-UFBA)
TITLE:

SIMULATED MOVING BED REACTOR: DYNAMIC MODELING, ABNORMAL OPERATION TRACKING AND OPTIMIZATION.

KEY WORDS:

Deep Learning Identification. Dynamic Chemical Systems. Simulated Moving Bed Reactor. Artificial Intelligence. Abnormal operation monitoring. Big data analytics. Gram-Schmidt Orthogonalization. Process Intensification. n-Propyl Propionate.

PAGES: 117
BIG AREA: Engenharias
AREA: Engenharia de Produção
SUMMARY:

This work proposes filling gaps in three areas of PSE applied to the cyclic process, specifi-
cally, the Simulated Moving Bed Reactor: Dynamic sensitivity analysis, model acceleration
with deep neural networks surrogates, robust optimization and fault diagnosis.
To start, this work proposes two Simulated Moving Bed Reactor (SMBR) units for the synthesis
of n-Propyl propionate (ProPro): the Simulated Moving Bed Reactor with 4 columns
(SMBR-4) and the SMBR-8. Their simulated performance was compared to the True Moving
Bed Reactor (TMBR). This work provides two detailed analyses to assess the system
dynamic behavior of the three units: an orthogonalization based analysis and the process
reaction curve. The first showed that the switching time and feed concentration in uence
the system the most, though at dierent extensions for each unit. The second highlighted
that the SMBR/TMBR commonly used equivalence is not valid during transient conditions.
When it concerns neural surrogates, it is known that Recurrent Neural Networks (RNNs)
are suitable predictors for Nonlinear Output Error (NOE)-type model family. NOE is one
of the most important representations in chemical engineering when long-term predictions
(simulation) are required. Even so, the chemical engineering literature presents notorious
gaps in this topic. Therefore, this work proposes a novel and complete framework for
identifying Deep Recurrent Neural Networks (DRNNs) for NOE representation with focus
on chemical engineering systems. The framework incorporates a novel training method
and the employment of a state-of-the-art automatic hyperparameter selection algorithm.
The results show that by using the proposed framework, DRNNs can model the SMBR
process with satisfactory accuracy, relying exclusively on input signals and past predictions
(free-simulation).
When dealing with model-based optimization of Simulated Moving Bed Reactor (SMBR),
ecient solvers and significant computational power are required. Over the past years,
surrogate models have been considered for such computationally-demanding optimization
problems. In this sense, Artificial Neural Networks have found applications for modeling the
SMB unit, but not yet been reported for the reactive SMB. However, a consistent method
for optimality assessment using surrogate models is still an open issue in the literature. As
such, two main contributions can be highlighted: the SMBR optimization based on Deep
Recurrent Neural Network (DRNN) and the characterization of the Feasible Operation
Region. This is done by recycling the data points from a meta-heuristic technique {
optimality assessment. The results demonstrate that the DRNN-based optimization can
address such complex optimization, while meeting optimality.
Finally, it is well known that Big Data plays a crucial role in Industry 4.0 by oering
tools to improve the decision-making process. Among the industrial sectors, the chemical
process industry already holds mature data management structures, but poorly explored
analytical tools. In this sense, this work proposes an online analytical tool which can deal
with big data to be used for identifying abnormal operation in chemical processes. It deals
with a modified Dynamic Sensitivity Matrix (DSM) and Gram-Schmidt Orthogonalization
(GSO) to prioritize process variables under abnormal behavior and scaling the impact they
have on plant performance.

BANKING MEMBERS:
Externo à Instituição - Galo Antonio Carillo Le Roux
Externa à Instituição - IDELFONSO BESSA DOS REIS NOGUEIRA
Externo à Instituição - LUIS CLAUDIO OLIVEIRA LOPES
Presidente - 2042153 - MARCIO ANDRE FERNANDES MARTINS