Location via proxy:   [ UP ]  
[Report a bug]   [Manage cookies]                

Self-Optimizing Control of A Gas Oil Hydrodesulfurization

Download as pdf or txt
Download as pdf or txt
You are on page 1of 167

POLITECNICO DI MILANO

Scuola di Ingegneria Industriale e dell'Informazione


Corso di Laurea in Ingegneria Chimica

Self-Optimizing Control of a Gas Oil Hydrodesulfurization


Plant

Relatori: Prof. Flavio MANENTI, Politecnico di Milano


Prof. Sigurd SKOGESTAD, NTNU – Trondheim

Tesi di Laurea Magistrale di:


Francesco Morandi, matricola n°783496

Anno Accademico 2013 – 2014


INDEX
Chapter 1 - Motivations, introduction and approach of the hydrodesulfurization
process ...........................................................................................................................1
1.1 INTRODUCTION ....................................................................................................1
1.2 THE APPROACH ....................................................................................................5
Chapter 2 - The hydrodesulfurization process: history, scheme and equipment ..........8
2.1 INTRODUCTION ....................................................................................................8
2.2 HISTORY ................................................................................................................8
2.3 THE PROCESS ...................................................................................................... 12
2.3.1 REACTOR SECTION ..................................................................................... 17
2.3.1a Feed Filters ................................................................................................. 17
2.3.1b Feed/effluent exchangers ............................................................................ 17
2.3.1c The make-up hydrogen system .................................................................... 18
2.3.1d The reactor..................................................................................................19
2.3.2 FRACTIONATION SECTION ........................................................................ 21
2.4 THE CHEMISTRY OF REACTION ...................................................................... 22
2.4.1 SULFUR .......................................................................................................... 24
2.4.1a Mercaptans ................................................................................................. 25
2.4.1b Sulfides....................................................................................................... 25
2.4.1c Di-Sulfides ..................................................................................................25
2.4.1d Thiphenes ...................................................................................................25
2.4.1e Benzothiophenes ......................................................................................... 26
2.4.1f Di-Benzothiophenes .................................................................................... 26
2.4.2 NITROGEN .....................................................................................................27
2.4.2a Pyrrole ........................................................................................................ 27
2.4.2b Pyridine ...................................................................................................... 28
2.4.2c Quinoline .................................................................................................... 28
2.4.3 OXYGEN ........................................................................................................ 29
2.4.3a Phenols ....................................................................................................... 29
2.4.3b Oxygenates ................................................................................................. 29
2.4.4 AROMATIC SATURATION........................................................................... 30
2.4.4a Toluene ....................................................................................................... 30
2.4.4b Phenanthrene .............................................................................................. 30
2.4.4c Naphthalene ................................................................................................ 30
2.5 THE CATALYST ..................................................................................................31
2.5.1 COBALT-MOLYBDENUM ............................................................................ 32
2.5.2 NICKEL-MOLYBDENUM ............................................................................. 33
2.5.3 NICKEL-TUNGSTEN and NICKEL-COBALT-MOLYBDENUM ................. 33
2.6 REACTORS ........................................................................................................... 35
Chapter 3 - Process Design: layout and modification ................................................... 38
3.1 INTRODUCTION ..................................................................................................38
3.2 LITERATURE BACKGROUND ............................................................................ 38
3.3 PLANT SCHEME from LITERATURE ................................................................. 38
3.3.1 NAPHTHA HDS UNIT ................................................................................... 39
3.3.1a Process Description ..................................................................................... 40
3.3.2 KEROSENE HDS UNIT .................................................................................. 42
3.3.2a Process Description ..................................................................................... 42
3.3.3 GAS OIL HDS UNIT........................................................................................ 44
3.3.3a Process Description ..................................................................................... 44
3.4 PLANT SCHEMES ................................................................................................ 46
3.4.1 FIRST MODEL: HDS from literature .............................................................. 48
3.4.1a Considerations ............................................................................................ 48
3.4.2 THE SIMPLEST MODEL: “naked” HDS ........................................................ 50
3.4.3 THE FIRST MODIFICATIONS ...................................................................... 51
3.4.4 AMINE LOOP STRIPPER .............................................................................. 52
3.4.5 SETTING PRESSURES and TEMPERATURES ............................................. 54
3.4.6 THE LAST MODEL ........................................................................................ 55
3.5 KINETICS.............................................................................................................. 56
3.5.1 FIRST KINETICS: FORMENT et al. ............................................................... 57
3.5.2 SECOND KINETICS: D. TSAMATSOULIS and N. PAPAYANNAKOS ....... 57
3.5.3 THIRD KINETICS: AYSAR T. JARULLAH, IQBAL M. MUJTABA,
ALASTAIR S. WOOD ............................................................................................. 58
3.6 THE FEEDSTOCK ................................................................................................ 60
Chapter 4 - Steady State simulation: theory, analysis and results ............................... 62
4.1 INTRODUCTION ..................................................................................................62
4.2 THE THEORY ....................................................................................................... 65
4.3 THE PROCEDURE ................................................................................................ 68
4.3.1 LUYBEN‟S PROCEDURE .............................................................................. 70
4.3.2 SKOGESTAD‟S PROCEDURE ...................................................................... 71
4.4 SKOGESTAD‟S PROCEDURE ............................................................................. 73
4.4.1 DEGREES OF FREEDOM .............................................................................. 73
4.4.1a Steady-State DOFs ...................................................................................... 73
4.4.2 TOP-DOWN .................................................................................................... 77
4.4.2a Step 1: Define Operational Objectives (Cost J and constraints) ................... 77
4.4.2b Step 2: Determine the steady-state Optimal Operation ................................ 78
4.4.2c Step 3: Select Economic (Primary) Controlled Variables, CV 1 .................... 82
4.4.2d Step 4: Select the Location of TPM (Decision 3) ......................................... 93
4.4.3 BOTTOM-UP ..................................................................................................95
4.4.3a Step 5: Select the Structure of Regulatory (Stabilizing) Control Layer ........ 95
4.4.3b Step 6: Select Structure of Supervisory Control Layer ................................. 98
4.4.3c Step 7: Structure of Optimization Layer (RTO) (related to Decision 1) ....... 99
4.5 ANALYSIS AND RESULTS ............................................................................... 100
4.5.1 Step 1:Define Operational Objectives ............................................................. 100
4.5.2 Step 2: Determine the steady-state Optimal Operation .................................... 104
4.5.2a Results of the Optimization ....................................................................... 108
4.5.3 Step 3: Select Economic (Primary) Controlled Variables, CV 1 ....................... 114
4.5.3a The Nullspace Method .............................................................................. 115
4.5.4 Step 4: Select the Location of TPM (Decision 3) ............................................ 125
4.5.5 Step 5: Select the Structure of Regulatory (Stabilizing) Control Layer ............ 126
4.5.6 Step 6: Select Structure of Supervisory Control Layer .................................... 126
4.5.7 Step 7: Structure of Optimization Layer (RTO) (related to Decision 1) ........... 129
Chapter 5 - Dynamic Simulation: tuning and scenarios ............................................. 130
5.1 INTRODUCTION ................................................................................................ 130
5.2 PID ....................................................................................................................... 130
5.3 TYPICAL REGULATOR CHOICE ..................................................................... 131
5.3.1 FLOOD .......................................................................................................... 132
5.3.2 PRESSURE.................................................................................................... 132
5.3.3 LEVEL .......................................................................................................... 132
5.3.4 TEMPERATURE ........................................................................................... 133
5.3.5 COMPOSITION ............................................................................................ 133
5.4 THE PLANT CONTROL ..................................................................................... 133
5.5 RESULTS ............................................................................................................ 136
Chapter 6 - Conclusions and future perspectives ....................................................... 142
6.1 CONCLUSIONS .................................................................................................. 142
Bibliography ................................................................................................................. 146
Volume: ..................................................................................................................... 146
Paper: ......................................................................................................................... 147
Site: ............................................................................................................................ 152
Guide: ........................................................................................................................ 154
INDEX of FIGURES
Figure 1: ideal Otto cycle, NASA, Otto Cycle,http://www.grc.nasa.gov/WWW/k-
12/airplane/otto.html, accessed on March 2014 ..................................................................1
Figure 2: ideal Diesel, HYPERPHYSIC, Diesel Cycle, http://hyperphysics.phy-
astr.gsu.edu/hbase/thermo/imgheat/diescyc.gif, accessed on March 2014 ...........................2
Figure 3: Diesel car penetration in major world markets, MICHEL CAMES, ECKARD
HELMERS, Critical evaluation of the European diesel car boom - global comparison,
environmental effects and various national strategies, Environmental Sciences Europe
2013, 25:15 ........................................................................................................................2
Figure 4: sulfur content of crude oil to refineries, EIA, http://www.eia.gov/, accessed on
March 2014 ........................................................................................................................5
Figure 5: density and sulfur content of selection crude oils, EIA, http://www.eia.gov/,
accessed on March 2014 ....................................................................................................6
Figure 6: typical process plant scheme; OZREN OCIC, Oil Refineries in the 21st Century:
Energy Efficient, Cost Effective, Environmentally Benign, Wiley 2005 ........................... 12
Figure 7: simplified process diagram; DAVID S. J. JONES, PETER R. PUJADÓ, Handbook of
Petroleum Processing Handbook of Petroleum Processing, Edited by DAVID S. J.
“STAN”, 2006 ................................................................................................................. 15
Figure 8: VGO Unionfinning Unit (Reactor Section), DAVID S. J. JONES, PETER R. PUJADÓ,
Handbook of Petroleum Processing, Springer 2006 ......................................................... 15
Figure 9: VGO Unionfinning Unit (Fractionation Section), DAVID S. J. JONES, PETER R.
PUJADÓ, Handbook of Petroleum Processing, Springer 2006 ........................................... 16
Figure 10: Francis‟ Diagram, GIULIO NATTA, ITALO PASQUON, PRICIPI DELLA CHIMICA
INDUSTRIALE, MILANO: TAMBURINI, 1966.......................................................................... 23
Figure 11: typical shape of hydroprocessing catalyst, DAVID S. J. JONES, PETER R. PUJADÓ,
Handbook of Petroleum Processing, Springer 2006 ......................................................... 32
Figure 12: Hot Soaking, DAVID S. J. JONES, PETER R. PUJADÓ, Handbook of Petroleum
Processing, Springer 2006 ............................................................................................... 34
Figure 13: typical vessel shape, DAVID S. J. JONES, PETER R. PUJADÓ, Handbook of
Petroleum Processing, Springer 2006 .............................................................................. 35
Figure 14: Surinder Parkash , Refining Processes Handbook, Gulf Professional Publishing,
2003................................................................................................................................. 39
Figure 15: Surinder Parkash , Refining Processes Handbook, Gulf Professional Publishing,
2003................................................................................................................................. 42
Figure 16: Surinder Parkash , Refining Processes Handbook, Gulf Professional Publishing,
2003................................................................................................................................. 44
Figure 17: Aspen HYSYS® model adapted from Surinder Parkash , Refining Processes
Handbook, Gulf Professional Publishing, 2003 ................................................................ 48
Figure 18: the easiest way to configure an HDS unit ........................................................ 50
Figure 19: HDS unit with a previous selection of the desired layout ................................. 51
Figure 20: HDS unit with the regeneration unit ................................................................ 52
Figure 21: HDS unit with the optimal configuration of the equipment .............................. 54
Figure 22: Aspen HYSYS® workbook............................................................................. 54
Figure 23: final HDS unit layout ...................................................................................... 55
Figure 24: how to select the reaction set in Aspen HYSYS® ............................................ 56
Figure 25: results from Aspen HYSYS® .......................................................................... 59
Figure 26: Gilbert F. Froment results ................................................................................ 60
Figure 27: Aspen HYSYS® composition ......................................................................... 61
Figure 28: Aspen HYSYS® feed conditions.....................................................................61
Figure 29: hierarchical control ......................................................................................... 63
Figure 30: PhD control .....................................................................................................63
Figure 31: Tennessee Eastman challenge problem (Downs, 1991) .................................... 64
Figure 32: different control levels .................................................................................... 68
Figure 33: a heat exchanger .............................................................................................. 75
Figure 34: DOFS of a distillation column ......................................................................... 76
Figure 35: a distillation column ........................................................................................ 79
Figure 36: objective function J versus cost energy............................................................ 82
Figure 37: objective function J versus DOFs .................................................................... 82
Figure 38: self-optimizing control of a marathon runner ................................................... 84
Figure 39: active output constraints .................................................................................. 86
Figure 40: “squeeze and shift” rule, Richalet© ................................................................. 87
Figure 41: simple scheme for transfer function ................................................................. 90
Figure 42: control scheme for the exact local method ....................................................... 91
Figure 43: TPM at inlet (feed), inventory control in direction of flow ............................... 94
Figure 44: TPM at outlet, inventory control in direction opposite the flow ....................... 94
Figure 45: general case with TPM inside the plant, Radiation Inventory Control .............. 94
Figure 46: spreadsheet with all contributes to the final cost function .............................. 101
Figure 47: Table 2 .......................................................................................................... 105
Figure 48: the plant ........................................................................................................ 105
Figure 49: fmincon function, MATHWORKS, fmincon,
http://www.mathworks.it/it/help/optim/ug/fmincon.html, accessed on March 2014 ........ 108
Figure 50: cost function optimized, no disturbances ....................................................... 109
Figure 51: changing in feed flood ................................................................................... 110
Figure 52:changing in feed flood .................................................................................... 110
Figure 53: changing in feed composition ........................................................................ 111
Figure 54: changing in composition ............................................................................... 111
Figure 55: changing in one feed‟s compound‟s composition........................................... 112
Figure 56: comparison between the different optimizations with the respect of disturbances
and their expected range ................................................................................................. 113
Figure 57: results of the Nullspace method applied to the previous optimal configuration,
first two matrices ........................................................................................................... 122
Figure 58: results of the Nullspace method applied to the previous optimal configuration,
second two matrices ....................................................................................................... 123
Figure 59: results of the Nullspace method applied to the previous optimal configuration,
with the second set of measurements .............................................................................. 124
Figure 60: location of the TPM, select at the beginning of the plant ............................... 125
Figure 61: HDS control structure ................................................................................... 128
Figure 62: the HDS control ............................................................................................ 134
Figure 63: valve signal on feed control valve ................................................................. 137
Figure 64: feed flood control and its disturb with increase in flood value from 60600 [kg/h]
to 63600[kg/h] ............................................................................................................... 137
Figure 65: inlet pressure and its control .......................................................................... 138
Figure 66: reactor temperature control (slow dynamics) and its response after the
disturbance ..................................................................................................................... 138
Figure 67: absorption column pressure control and the disturbance effect....................... 139
Figure 68: absorption column pressure control with autotuning results ........................... 139
Figure 69: desulfurized conditions ................................................................................. 140
Figure 70: level control of the desulfurized charge, with the result after the disturb in feed
flood .............................................................................................................................. 140
Figure 71: composition after disturb in total sulfur content ............................................. 141
INDEX of TABLES
Table 1: applications of Unicracking Process ...................................................................13
Table 2: different hydrotreating processes and their products ........................................... 22
Table 3: Operative Conditions Naphtha Unit, SURINDER PARKASH , Refining Processes
Handbook, Gulf Professional Publishing, 2003 ................................................................ 41
Table 4: Operative Conditions Kerosene Unit, SURINDER PARKASH , Refining Processes
Handbook, Gulf Professional Publishing, 2003 ................................................................ 43
Table 5: Operative Conditions Gas Oil Unit, SURINDER PARKASH , Refining Processes
Handbook, Gulf Professional Publishing, 2003 ................................................................ 46
Table 6: HDS kinetic parameters of D. TSAMATSOULIS AND N. PAPAYANNAKOS .............. 58
Table 7: HDS kinetic parameters of AYSAR T. JARULLAH, IQBAL M. MUJTABA, ALASTAIR S.
WOOD .............................................................................................................................. 59
Table 8: top-down/bottom up procedure ........................................................................... 71
Table 9: Potential steady-state degrees of freedom ........................................................... 76
Table 10: costs of electricity and natural gas .................................................................. 102
Table 11: price and value of raw materials ..................................................................... 103
Table 12: operational constraints set to the system ......................................................... 107
Table 13: DOFs with their unit of measurement ............................................................. 109
Table 14: results of the third configuration ..................................................................... 113
Table 15: first candidate measurements .......................................................................... 118
Table 16: second candidate measurements ..................................................................... 119
Table 17: disturbances and their value after small perturbation ....................................... 119
Table 18: positive disturbance in mass flow, first candidate measurements .................... 119
Table 19: negative disturbance in mass flow, first candidate measurements .................... 120
Table 20: positive disturbance in composition, first candidate measurements ................. 120
Table 21: negative disturbance in composition, first candidate measurements ................ 120
Table 22: positive disturbance in mass flow, second candidate measurements ................ 121
Table 23: negative disturbance in mass flow, second candidate measurements ............... 121
Table 24: positive disturbance in composition, second candidate measurements ............. 121
Table 25: negative disturbance in composition, second candidate measurements ............ 122
ABSTRACT
This work aims to give an appropriate introduction to an innovative technique of plantwide
control, based on economics, recently developed by Professor Sigurd Skogestad of NTNU
University of Trondheim (NOR); and then its application to a well known post-treatment
process: a hydrodesulfurization process plant. The procedure is note as “self-optimizing
control” and it is based on the search of some variables which kept near to their optimal
point can give acceptable results without too much loss. With this project the main steps of
the procedure will be analyzed and then applied and finally a control structure will be
developed, leaving to next work the possibility to increase and amply the choice of
different controllers or different value of parameters. The unit plant has been modeled in
AspenHYSYS®, starting from different examples found in literature; the same procedure
has been followed for the kinetics adopted. This model has been the basis on which test the
so-called “self-optimizing control”. The structure has been tested on two main
disturbances: feed flow and its composition. This choice has been done in order to make it
possible a right approaching to the technique, because even if it its simplicity is its main
characteristic, it is worth to note that its application on a whole unit plant it is still at the
beginning. This approach to plantwide control is innovative because it can give the right
orderliness and a mental guideline to follow when whoever starts thinking about the
control everything: from the single equipment till the complete plant.
Starting from the choice of the main variables to control the procedure ends with the
selection of the appropriate control structure, whose nature remains free to be chosen
among the different ones present in literature.
Reasonable results have been obtained from the point of view of sulfur removed, in line
with what European Commission decided and at steady state conditions. However even if
the dynamic simulation has been found to be correct, the dynamic behavior of the plant is
difficult to control and some numerical problems gives the wrong way to the system.
ABSTRACT
Questo lavoro di tesi prevede l‟utilizzo di una nuova procedura tecnica di controllo
dell‟impianto basata fondamentalmente sull‟economia dello stesso. Questo approccio
avveniristico sviluppato negli ultimi anni dal Professor Sigurd Skogestad dell‟Università
NTNU di Trondheim (Norvegia), il cosiddetto “self-optimizing control” prende il nome da
alcune variabili peculiari sulla cui ricerca si basa essenzialmente la procedura stessa: “self-
optimizing variables” il cui controllo se effettuato efficacemente e con logica permette di
mantenere il sistema su valori vicini all‟ottimo di normale funzionamento senza di
conseguenza avere perdite dal punto di vista economico.
La struttura del progetto ha previsto differenti fasi: dal periodo presso l‟Università di
Trondheim fino al successivo completamento preso il Politecnico di Milano.
L‟articolazione del lavoro di tesi invece segue una svolgersi più regolare, attraverso una
breve introduzione all‟argomento si passa alla storia del processo (tornato di recente in
auge viste le sempre più restrittive norme in materia di inquinamento e di emissioni nocive
da parte di autoveicoli/industrie stesse); quindi viene affrontata il processo di modellazione
a partire da differenti esempi tratti dalla letteratura, dai quali si studia la collocazione dei
differenti dispositivi in campo e i valori delle grandezze in gioco. Seguendo la
presentazione di diversi lavori sull‟argomento dell‟idrodesolforazione delle cariche di
gasolio e altre più pesanti si prendono in atto differenti articoli trattanti le cinetiche di
questo processo. Vengono selezionate tre cinetiche per l‟occasione e ne vengono vagliate
le caratteristiche e soprattutto la riproducibilità di risultati coerenti con quanto richiesto da
un siffatto impianto; verificate anche dal punto di vista della disponibilità di tali dati in
quanto è stato possibile sperimentare direttamente l‟impossibilità di fruire di alcune
cinetiche in quanto coperte da brevetto e parte di un pacchetto commerciale di simulazione
(nel caso specifico la cinetica proposta da Gilbert Froment e non più disponibile a detta
dello stesso insigne Professore). Selezionata la cinetica si è passato alla modellazione reale
dell‟impianto, andando a simularlo attraverso AspenHYSYS®, noto pacchetto
commerciale di simulazione della AspenTech®. Attraverso successive semplificazioni
dell‟impianto si è giunti ad una configurazione stabile in grado di predire i risultati ricavati
dagli articoli precedenti. Il modello è stato testato e successivamente si passati alla vera e
propria procedura per il controllo.
L‟approccio proposto dal Professor Sigurd Skogestad è stata quindi spiegata passo a passo
e illustrata attraverso l‟esemplificazione dei passaggi e il confronto con altri significativi
sistemi di controllo sulla base dei quali in parte è stato costruito quest‟ultimo. I differenti
metodi matematici adottati sono stati integrati con materiale ad hoc e il tutto è stato quindi
applicato al modello e nel dettaglio al processo: brevemente il modello è stato sottoposto a
differenti cicli di ottimizzazione atti ad ottenere i valori ottimali del processo che saranno
quindi i set-point da adottare nei controllori, sulla base dei “gradi di libertà” del sistema in
esame stesso; terminata questa prima fase si è indagato a fondo sulle “self-optimizing
varibles” da controllarsi insieme ai suddetti gradi di libertà per incrementare l‟efficacia e la
robustezza del controllo.
Terminata la scelta delle variabili supposte essere le migliori per il controllo dunque, il
tutto è stato implementato in una struttura di controllo e questa è stata verificata per
verificarne l‟efficacia; i risultati della simulazione in dinamico sono quindi stati riportati
per completezza di informazione.
È stato possibile verificare dunque la possibilità di applicazione di questa procedura su un
impianto a livello industriale, pur rimanendo vero il fatto che notevoli semplificazioni sono
state effettuate in modo tale da permetterne la realizzazione pratica. Questo risulta essere
un notevole spunto in vista di miglioramenti e quindi l‟introduzione di sistemi di controllo
più articolati quali MPC o se la potenza di calcolo dovesse consentirlo, nel caso specifico
la velocità di calcolo, l‟implementazione di una RTO. La maggior parte dell‟attenzione è
stata volta n ella ricerca e nello studio del processo.
FRANCESCO MORANDI SELF-OPTIMIZING CONTROL

Chapter 1
Motivations, introduction and approach of the
hydrodesulfurization process

“One”
U2
1.1 INTRODUCTION
The automotive industries has today the necessity to face and try to satisfy the
environmental prescriptions in terms of human health and reducing air pollution.
Even if in the last decades there have been a wide development focused to new energy
solutions (i.e. hydrogen cars, solar cells, PEM and electric cars more in general); we have
not to forget that the greatest part of the automotive industry is covered by gasoline and
diesel cars. For this reason we have not to surprise if the major Environmental Agencies in
whatever country aim to reduce the pollutants coming from the exhaust gases of these two
type of engines. In this sense we mean the engine that works following the Otto‟s cycle,
and that is fed by gasoline and the Diesel‟s cycle which uses for its right working a mixture
of higher weighted hydrocarbons (i.e. paraffin oils) compared with Otto‟s fuel.

Figure 1: ideal Otto cycle, NASA, Otto Cycle,http://www.grc.nasa.gov/WWW/k-12/airplane/otto.html, accessed on


March 2014

1
FRANCESCO MORANDI SELF-OPTIMIZING CONTROL

Figure 2: ideal Diesel, HYPERPHYSIC, Diesel Cycle, http://hyperphysics.phy-


astr.gsu.edu/hbase/thermo/imgheat/diescyc.gif, accessed on March 2014

In general for what concerns the distribution of this two main groups of cars we have to
note that in the last years diesel engines have raised their spread 1, losing its natural tiling to
the thinking of it as only truck engine; and this because of the improvements of the diesel
engines performances (e.g. the introduction of the Common Rail® 2 technology and
Multijet® system3) and reduced consumes. Anyway a trend line can be traced, highlighting
that gasoline cars are the most used for civil and private transportation, while the diesel
engines are used for power plants, energy supply and naval transportations.
Considering all these application we can note how really is spread this typology of engine.

Figure 3: Diesel car penetration in major world markets, MICHEL CAMES, ECKARD HELMERS, Critical evaluation of the
European diesel car boom - global comparison, environmental effects and various national strategies ,
Environmental Sciences Europe 2013, 25:15

1
MICHEL CAMES, ECKARD HELMERS, Critical evaluation of the European diesel car boom - global comparison,
environmental effects and various national strategies, Environmental Sciences Europe 2013, 25:15
2
FIAT, Common Rail,http://www.fiat.it/fiatpedia/glossario/common-rail, accessed on March 2014
3
Ibidem

2
FRANCESCO MORANDI SELF-OPTIMIZING CONTROL

Consequently it is obvious the reason that pull the most important environmental
institutions to take care and control the emission of diesel engines.
How could it be possible to reduce the noxious pollutant from these engines? It depends.
There are several ways to act in this sense; but first of all it is necessary to distinguish
between two groups of treatment: pre-treatment and post-treatment. The difference lies
where the process studied and applied, works. Basically all pre-treatment aims to reduce
upstream the presence of pollutant, letting the post-treatment to keep on reducing them till
the desired values; for what concerns instead the post-treatment they work properly
downstream with the use of post-combustion system or particulate filter in order to
mechanically reduce the non-desired combustion byproducts.
It is clear that the best result is achieved with both the system that act concurrently; but
while post-treatment can be mounted directly on the device uses the diesel engine, a pre-
treatment has is goal on the fuel fed and for this reason we have to operate on large scale
and particularly on a unit of a plant.
Among the major cause of pollution and damage to human health there are the so-called
combustion oxides, i.e. Co x, NOx and SOx. Where the “x” means the generic number of
oxidation where the oxide could be present.
These kind of oxides are noxious for their effects both on the environment and the human
health. Especially the sulfur oxides are responsible of the so called “acid rains 4” and some
human pathologies on lung 5 and respiratory system.
This is a strong motivation to the will of reduce this kind of emissions. Diesel fuel
specifications are being tightened throughout the world as part of efforts to improve air
quality. At the same time, the demand for diesel is increasing necessitating use of lower
quality feedstocks. All of this results in a great burden on refineries, which have to face
this important task. New hydrotreating capacity and revamp of existing facilities are
needed to meet the future diesel specifications and even if today‟s emphasis is on the
reduction of sulfur content, it is not strange that in the next future the focus will be set to
all the other toxic and noxious compounds presents in diesel exhaust gases.

4
SO2 + H2O  H2SO3
SO3 + H2O  H2SO4
5
IARC, IARC: DIESEL ENGINE EXHAUST CARCINOGENIC,
http://www.iarc.fr/en/media-centre/pr/2012/pdfs/pr213_E.pdf, accessed on March 2014

3
FRANCESCO MORANDI SELF-OPTIMIZING CONTROL

There are different normative dealing with the maximum sulfur content in the emissions of
car vehicles; different from country to country with a common intent to unify them in the
next future.
The Environmental Protection Agency (EPA) has proposed new guidelines to limit sulfur
in diesel fuels from low sulfur diesel (LSD) of 500 [ppm wt/wt] to ultra low sulfur diesel
(ULSD) of less than 15 [ppm wt/wt] in 20106; in California, transit bus fleets must use a 15
[ppm wt/wt] sulfur diesel by October, 2002 and it is expected that this limit will be
extended to more fleets shortly afterward. The same procedure has been proposed by
European Commission7, while the same path is going to be followed by other legislations,
introducing, moreover, limitations on aromatic content in diesel fuels to less than 20 % in
the near future.
In the EU, current diesel specifications (2000) limit sulfur is to a maximum of 350 [ppm
wt/wt]. By the year 2005, the sulfur content must be reduced to 50 [ppm wt/wt], and diesel
containing a maximum of 10 [ppm wt/wt] must also be available. By the first of January of
2009 all on-road diesel in EU would have a specification of maximum 10 [ppm wt/wt]
sulfur. Another example in Europe in Germany 10 [ppm wt/wt] sulfur diesel will be
mandatory from 2003.
In Japan, there are the more severe restrictions and the government is considering plans to
reduce sulfur content to below 10 [ppm wt/wt] by 2008. Another interesting point is that
the demand for diesel is growing in both Asia and Europe: we can see it clearly for what
concerning the Asia Pacific demand which has remained still from 1996-1998, while it has
raised to 4.6% in 1999. Such a request is forecast to be equivalent of a 5-8% per year.
In Europe has been quite constant and fixed at about 1.2% per annum8.
In parallel we have a decreasing demand for some fractions not more necessary for the
sustaining of the economy (i.e. home heating oil and fuel oil cutter stocks). That is index of
an evolution in term of quality and increasing in specification for everyday‟s use.

6
Kim G. Knudsen and Barry H. Cooper, Ultra Deep Desulfurization of Diesel: How an understanding of the
underlying kinetics can reduce investment costs, Practical Advances in Petroleum Processing, pp 297-316,
Springer 2006
7
http://ec.europa.eu/italia/attualita/archivio/trasporti_energia/11078bab270_it.htm
8
Ibidem 6

4
FRANCESCO MORANDI SELF-OPTIMIZING CONTROL

Figure 4: sulfur content of crude oil to refineries, EIA, http://www.eia.gov/, accessed on March 2014

We have to note that a reduction in aromatic content is useful first of all to decrease gas
emission (hydrocarbons and particulate matter), to improve the cetane index (i.e. the
equivalent of the octane number for gasoline) and decrease the density of the fuel. As
explained before it is expected that ultra deep hydrodearomatization (HDA) and
hydrodesulphurization (HDS) of gas oils may be considered necessary in order to satisfy
the market and environmental requirements.

1.2 THE APPROACH


It is clear that to make possible this drastic reduction and then accomplish the new
environmental requirements, it is necessary the removal of refractory sulfur aromatics
compounds. It is note anyway that these hindered compounds are remaining in the diesel
fuel after the sulfur reduction to 500 [ppm wt/wt] level by conventional HDS process
because of the inhibiting effects of co-existing polyaromatics and nitrogen compounds in
the feed, responsible for a H2 consumption (even if it is necessary to specify that those
compounds are more stable to bland operative conditions); moreover we have to consider
that the kinetics of deep desulfurization is governed by the extent to which desulfurization
(HDS) occurs by direct sulfur extraction, or by hydrogenation of the sulfur -containing
molecule, and that the presence of H2S is an inhibiting factor on the overall reaction of
hydrodesulfurization.
However it is more difficult for the refineries to achieve less than the 15[ppm wt/wt]
(ULSD) sulfur specification with the currently employed catalysts and use of typical
operating conditions.

5
FRANCESCO MORANDI SELF-OPTIMIZING CONTROL

To increase this result, many aspects such as the catalysts, process variables, and the
quality of feedstock could play a major role on the shift from normal desulfurization (LSD)
to ultra deep desulfurization (ULSD) of diesel feeds.

It is also clear that the beginning sulfur content is function of the coming crude oil, as we
can observe in Figure 5:

Figure 5: density and sulfur content of selection crude oils, EIA, http://www.eia.gov/, accessed on March 2014

The shift from LSD to ULSD with low aromatics content is a very convoluted technical
issue; and it is clear that any improvements could be obtain working concurrently by taking
into consideration both catalyst and process issues (HDS and HDA of diesel fuels requires
either severe process conditions such as temperature, pressure and space velocity or the use
of novel effective catalysts). It is well known that HDS reactions at high reaction
temperature will result in higher rates, but the same increasing is responsible for the coke
formation on the catalyst, which in turn tends to deactivate the catalyst rapidly leading in
this way a lower reaction rate; moreover it is note that using low Liquid Hourly Space
Velocity (LHSV, [h-1]) will result in decreasing hydrotreating efficiency, because of the
scarce turbulence that will develop (we have to remember that a typical HDS reactor
consists in a trickle bed system where there is the concurrent presence of a liquid phase, i.e.
the charge feed, a solid phase, that is the catalyst, and a gaseous one, obviously the
hydrogen). Hence, deep understanding of chemistry involved in removing the sulfur and

6
FRANCESCO MORANDI SELF-OPTIMIZING CONTROL

the influences of the important operating conditions are extremely necessary. And more
necessary is an optimization of the above-mentioned variables. This is the real aim and
goal of whatever approach we want to adopt.

In this perspective, this work aims to give its support giving a clear and schematic
procedure about how to optimize, or better “self optimize” a hydrodesulfurization unit in
order to see what are the most important variables to take into consideration to follow the
specification imposed (accomplish them) and to make possible to provide a system able to
give a product pre-treated with a sulfur content as low as it could be possible to post-treat
till the desired specification.
This project is dealing with a common unit of hydrodesulfurization for which it is
investigated the possibility to obtain a simple and robust control structure based basically
on economics; that is done facing the problem on the basis of the so-called plantwide
control, adopting a procedure that has been developed for the last 25 years by Professor
Sigurd Skogestad of NTNU – Norwegian University of Science and Technology, with
whom I‟ve dealt with the problem.
It will follow the presentation of the project, specifying the procedure adopted and the
plant scheme studied and used to simulate the behavior of the unit. A comparison with
literature data has validate the correct simulation of this modeled system. Both kinetics and
example feed have been searched and taken from literature; hence studied, they have been
implemented in a simulated model built with AspenHYSYS® suite. In this way the process
has been understood and verified at steady state conditions (necessary condition to
implementation of the control procedure) and then controlled dynamically.
Here it will follow an explanation about the procedure and its precursors and then the real
application of it to the above mentioned process, giving an example of application.

7
FRANCESCO MORANDI SELF-OPTIMIZING CONTROL

Chapter 2
The hydrodesulfurization process: history, scheme
and equipment

“Show me the way”


Peter Frampton/The Cranberries
2.1 INTRODUCTION
Hydrotreating is a very general term which can be applied in all processes where there is a
feedstock which is enhanced in some manner by passing it over a catalyst in the presence
of hydrogen. Meant in this sense the terms hydrotreating, hydroprocessing, hydrocracking,
and hydrodesulfurization are used rather loosely in the industry because, in the processes
hydrodesulfurization and hydrocracking, cracking and desulfurization operations occur
simultaneously and it is relative as to which predominates.
Generally in the process, there is essentially no reduction in molecular size of the feed. The
objective has most often been to reduce the sulfur content of the feed (due to
environmental restrictions or in order to grant better performance in some oil or
lubricants); but it‟s also practiced
to reduce nitrogen content, saturate olefins, and to reduce aromatics, HDN and HDA
respectively. In particular the process of hydrodesulfurization (HDS process) has received
in the last year a greater consideration due to sulfur impact on environment (eliminate
pollutants) especially for what concerns diesel fuel.
In this process sulfur is the easiest to remove, followed by nitrogen, and the aromatics.
Saturation of olefins overlaps sulfur and nitrogen.

2.2 HISTORY
Among the petroleum refining processes, hydrotreating and hydrocracking are the oldest
catalytic ones9. Hydrotreating has its origin in the hydrogenation work done by Sabatier
and Senderens, who were the first of all that in 1897 published their discovery regarding

9
DONALD ACKELSON, UOP UNICRACKING PROCESS FOR HYDROCRACKING, in Robert A. Meyers (ed.), Handbook
of Petroleum Refining Processes, 3rd ed. (2003)

8
FRANCESCO MORANDI SELF-OPTIMIZING CONTROL

unsaturated hydrocarbons; and in particular the possibility that they could be hydrogenated
in the vapor phase over a nickel catalyst. After few years, in 1904, Ipatieff understood that
that reaction was directly connected to hydrogen partial pressure and he noted that it could
be extended in the range of feasible hydrogenation reactions by the introduction of
elevated hydrogen pressures.
We have to consider that at the time, the progress of the automobile industry was expected
to entail a considerable increase in the consumption of gasoline; and this was the context
that led to the experimental work by Bergius10, started in 1910 in Hanover, Germany who
sought to produce gasoline by cracking heavy oils and oil residues as well as converting
coal to liquid fuels. That was the first attempt to try to convert a solid fuel in a liquid one
and it is clear the reason of that work: a liquid fuel could be easier moved, transported and
then fed and used for Benz engine. He realized that to remedy the inferior quality of the
unsaturated gasoline so produced (with this Coal To Liquid precursor process), the
hydrogen removed mostly in the form of methane during the cracking operation has to be
replaced by addition of new hydrogen. In this way, formation of coke was avoided and the
gasoline produced was of a rather saturated character. Parallel with this saturation of olefin
to paraffin Bergius also noted that the sulfur contained in the oils was eliminated for the
most part as hydrogen sulfide, H2S. Instead of nickel as catalyst, ferric oxide was used in
the Bergius process to remove the sulfur. Actually, the ferric oxide and sulfides formed in
the process acted as catalysts, though the activity was very poor (ferric oxide is always
present in part in all processes as byproduct of corrosion of the equipment). The first plant
for hydrogenation of brown coal was put on stream in Leuna Germany in 1927 where they
were employed to convert lignite to gasoline and subsequently to treat petroleum residues
to distillable fractions. And the past large scale industrial development of hydrogenation in
Europe, particularly in Germany, was due entirely to military considerations: Germany
used hydrogenation extensively during World War II to produce gasoline: 3.5 million tons
were produced in 1944; we have to remember that any availability of gasoline coming
from distillation of a crude or any petroleum feed was to consider impossible due to
military accordance. So the only way to become self-sufficient from the point of view of

10
Friedrich Karl Rudolf Bergius (born October 11, 1884, Goldschmieden, near Breslau, Germany [now
Wrocław, Poland+—died March 30, 1949, Buenos Aires, Argentina), German chemist and corecipient, with
Carl Bosch of Germany, of the 1931 Nobel Prize for Chemistry;
http://www.britannica.com/EBchecked/topic/61797/Friedrich-Bergius, accessed on February 2014

9
FRANCESCO MORANDI SELF-OPTIMIZING CONTROL

the fuel production was to exploit the natural sources of their land, that means to use coal
from the inner region of Germany (e.g. Saar).
The first commercial hydrorefining installation in the United States was at Standard Oil
Company of Louisiana in Baton Rouge in the 1930s. Following World War II, growth in
the use of hydrocracking was slow. These plants were developed by Humble Oil and
Refining Company and Shell Development Company, though there was considerably less
dependence on hydrogenation as a source of gasoline. Moreover it has to be concerned
that the availability of Middle Eastern crude oils reduced the incentive to convert coal to
liquid fuels, and new catalytic cracking processes proved more economical for converting
heavy crude fractions to gasoline. Even though hydrogenation has been of interest to the
petroleum industry for many years, little commercial use of hydrogen-consuming
processes has been made because of the lack of low-cost hydrogen. In a common plant for
any hydrotreatment we have to note that the major voice to take into account for an
economic evaluation of the plant costs is the price of hydrogen; and in that period there
weren‟t sufficient alternative sources available to feed hydrogen at low price.
That changed in the early 1950s with the advent of catalytic reforming which made
available by-product hydrogen. That brought up an extensive and increased interest in
processes that will utilize this hydrogen to upgrade petroleum stocks. Indeed in the 1950s,
hydrodesulfurization and mild hydrogenation processes experienced a tremendous growth:
this was due to large availability of great quantities of by-product hydrogen were made
from the catalytic reforming of low-octane naphthas to produce high-octane gasoline. The
first modern hydrocracking operation was placed on-stream in 1959 by Standard Oil
Company of California. The unit was small, it was producing only 1000 (BPSD)11 . As
hydrocracking units were installed to complement existing fluid catalytic cracking (FCC)
units, in the refineries the hydrocracking process was quickly recognized to have the
flexibility to produce varying ratios of gasoline and middle distillate. In this sense there has
been rapid growth in U.S. hydrocracking capacity from about 3000 BPSD in 1961 to about
120,000 BPSD in 1966.
Between 1966 and 1983, U.S. capacity grew eightfold, to about 980,000 BPSD. Outside
the United States, early applications involved production of liquefied petroleum gas (LPG)
by hydrocracking naphtha feedstocks. The excellent quality of distillate fuels produced

11
barrels per stream day: a measurement used to denote rate of oil or oil-product flow while a fluid-
processing unit is in continuous operation (abbreviated BSD). McGraw-Hill Dictionary of Scientific &
Technical Terms, 6E, Copyright © 2003 by The McGraw-Hill Companies, Inc.

10
FRANCESCO MORANDI SELF-OPTIMIZING CONTROL

when hydrocracking gas oils and other heavy feedstocks led to the choice of the
hydrocracking process as a major conversion step in locations where diesel and jet fuels
were in demand. Interest in high-quality distillate fuels produced by hydrocracking has
increased dramatically worldwide. As a result of the enormous growth of hydrotreating, as
of the beginning of 2001, there were more than 1,600 hydrotreaters operating in the world
with a total capacity in excess of 39,000,000 [B/day] (4,800,000 [Mtonn/day]). And as of
2002, more than 4 million BPSD of hydrocracking capacity is either operating or is in
design and construction worldwide.

In general, hydrotreating refers to a relatively mild operation whose primary purpose is to


saturate olefins and it will refer also to reduce the sulfur and/or nitrogen content (and not to
change the boiling range) of the feed. Hydrocracking instead refers to processes whose
primary purpose is to reduce the boiling range and in which most of the feed is converted
to products with boiling ranges lower than that of the feed. It means that the operation is
conducted in more severe operative conditions. Hydrotreating and hydrocracking set the
two ends of the spectrum and those processes with a substantial amount of sulfur and/or
nitrogen removal and a significant change in boiling range of the products versus the feed
are called hydroprocessing in the wider sense of the term.
A mentioned before all these processes are processes whose aim is to catalytically stabilize
petroleum products and/or remove objectionable elements from products or feedstocks by
reacting them with hydrogen. When we speak about stabilization, we usually mean all
reactions involved to convert unsaturated hydrocarbons such as olefins and gum-forming
unstable diolefins to paraffins. The principal elements removed with this process
(hydrotreating in general) include sulfur, nitrogen, oxygen, halides, and trace metals; and
in this sense is common use to indicate the process as HDS (i.e. HydroDeSulfurization),
HDN (that is HydroDeNitrification) and so on.

For what concerns hydrotreating, it is a versatile treatment that can be applied to a wide
range of feedstocks, from naphtha to reduced crude. What change from feedstock to
feedstock is the amount of required hydrogen, the composition of the catalyst involved in
the reaction and the severity of the overall operating conditions. Moreover in the last
decades, to meet environmental regulations12, that are becoming more stringent every year,

12
http://ec.europa.eu/environment/air/quality/legislation/existing_leg.htm, seen on March 2013

11
FRANCESCO MORANDI SELF-OPTIMIZING CONTROL

it also may be necessary to hydrogenate aromatic rings to reduce aromatic content by


converting aromatics to paraffins; in particular the major attention is focused on some
delicate compounds note for their proved damage on human health: the so-called PAH, i.e.
Polycyclic Aromatic Hydrocarbon13, 14.

Although there are about 30 hydrotreating processes available for licensing 15, most of them
have essentially the same process flow for a given application. There is a common way to
operate and treat the feed fed to the plant so in general a common explanation can be done
about it. We just remember that cause the extreme variation range of feedstock, it will be
possible to find some differences between the operative conditions and the catalyst adopted
during the treatment.

2.3 THE PROCESS

Figure 6: typical process plant scheme; OZREN OCIC, Oil Refineries in the 21st Century: Energy Efficient, Cost Effective,
Environmentally Benign, Wiley 2005

13
UNITED STATES ENVIRONMENTAL PROTECTION AGENCY, OFFICE OF ENVIRONMENTAL INFORMATION,
Emergency Planning and Community Right-to-Know Act – Section 313: Guidance for Reporting
Toxic Chemicals: Polycyclic Aromatic Compounds Category, EPA 260-B-01-03, Washington, DC,
August 2001
14
http://nepis.epa.gov/Exe/ZyNET.exe/900B0H00.TXT?ZyActionD=ZyDocument&Client=EPA&Index=2000+T
hru+2005&Docs=&Query=&Time=&EndTime=&SearchMethod=1&TocRestrict=n&Toc=&TocEntry=&QField=
&QFieldYear=&QFieldMonth=&QFieldDay=&IntQFieldOp=0&ExtQFieldOp=0&XmlQuery=&File=D%3A\zyfile
s\Index%20Data\00thru05\Txt\00000011\900B0H00.txt&User=ANONYMOUS&Password=anonymous&Sort
Method=h|-
&MaximumDocuments=1&FuzzyDegree=0&ImageQuality=r75g8/r75g8/x150y150g16/i425&Display=p|f&D
efSeekPage=x&SearchBack=ZyActionL&Back=ZyActionS&BackDesc=Results%20page&MaximumPages=1&Zy
Entry=1&SeekPage=x&ZyPURL, seen on March 2013
15
JAMES H. GARY, GLENN E. HANDWERK, Petroleum Refining Institut Français du Pétrole, 4th ed., CRC
Press, 2001

12
FRANCESCO MORANDI SELF-OPTIMIZING CONTROL

As noted previously hydrocracking is one of the most versatile of all petroleum refining
processes and as we refer in general to hydrocracking we have to understand that in this
sense any fraction from naphtha to non distillables can be processed to produce almost any
desired product with a molecular weight lower than that of the chargestock. At the same
time that hydrocracking takes place, sulfur, nitrogen, and oxygen are almost completely
removed, and olefins are saturated so that products are a mixture of essentially pure
paraffins, naphthenes, and aromatics. In this way the severity of operative conditions in
this process make possible the parallel reaction of cracking of the charge, lumping the feed
into different product with a obvious different market price; then it is possible to reduce
impurities present in the chargestock in form of sulfur, nitrogen, oxygen and metal
compounds. It is specified that they are impurities cause the toxicity of the above
mentioned compound for human health and their impact on the environment. With
reference to Table 1, it is possible to illustrate the wide range of applications of
hydrocracking by listing typical chargestocks and the usual desired products.
As possible to see the first eight chargestocks are virgin fractions of petroleum crude and
gas condensates; while the last four are fractions produced from catalytic cracking and
thermal cracking.

Table 1: applications of Unicracking Process

Chargestock Products
Naphtha Propane and butane (LPG)
Kerosene Naphtha
Straight-run diesel Naphtha and/or jet fuel
Atmospheric gas oil Naphtha, jet fuel and/or distillates
Natural gas condensates Naphtha
Vacuum gas oil Naphtha, jet fuel, distillates, lubricating oils
Deasphalted oils and demetallized oils Naphtha, jet fuel, distillates, lubricating oils
Atmospheric crude column bottoms Naphtha, distillates, vacuum gas oil and low-sulfur
residual fuel
Catalytically cracked light cycle oil Naphtha
Catalytically cracked heavy cycle oil Naphtha and/or distillates
Coker distillate Naphtha
Coker heavy gas oil Naphtha and/or distillates
From: DONALD ACKELSON, UOP UNICRACKING PROCESS FOR HYDROCRACKING, in Robert A. Meyers (ed.),
Handbook of Petroleum Refining Processes, 3rd ed. (2003)

13
FRANCESCO MORANDI SELF-OPTIMIZING CONTROL

All these streams are being hydrocracked commercially to produce one or more of the
products listed; and the reason that drives this decisions is purely economic, due to the
major added value that some products have respect with other ones especially for the
feedstock. This flexibility gives the hydrocracking process a particularly important role as
refineries attempt to meet the challenges of today‟s economic climate. The combined
influences of low-quality feed sources, capital spending limitations, hydrogen limitations,
environmental regulatory pressures, and intense competition have created a complex
optimization problem for refiners.

Hydrotreaters, with this term we mean the reactor able to conduct all the above mentioned
processes, are designed for and run at a variety of conditions depending on many factors
such as type of feed, desired cycle length, expected quality of the products (we have in fact
seen a brief explanation of the different charge fed to this process), but in general they will
operate at the following range of conditions:
- LHSV: from 0.2 to 8.0
- H2 circulation from 300 to 4,000 SCFB (i.e. 50 – 675 [Nm3/m3*])
- H2 Partial Pressure: ranging from 200 to 2,000 [psia] or 14 – 138 [bar]
- SOR temperatures ranging between 550 and 700[°F], that is 290 – 370[°C], with
the lower limits representing minimum operating conditions for naphtha
hydrotreating (that has the minimum value of temperature) and the higher values
showing operating conditions used for hydrotreating atmospheric residues, those
are the feedstock which require the most severe condition of treatment.

Until about 1980, hydrotreating was a licensed technology being offered by a fairly large
number of companies. In the past 25 years, hydrotreating catalysts have become
commodities and the process has been offered without licensing fees.

Except for the cycle dealing with the purification of off-gases and the subsequently
purification for the recycle of hydrogen to the reactor, we can simplify the process diagram
as follows:

14
FRANCESCO MORANDI SELF-OPTIMIZING CONTROL

Figure 7: simplified process diagram; DAVID S. J. JONES, PETER R. PUJADÓ, Handbook of Petroleum Processing Handbook of
Petroleum Processing, Edited by DAVID S. J. “STAN”, 2006

Where with letter “R” we mean the reactor section of the plant and “F” stands for the
fractionation part of the same (i.e. product purification, or better all processes involved in
the recovery of the non-reacted H2 through first a flash tanks separator, then through an
absorption column with its subsequently recovery loop).
We can better observe what mentioned before, taking into consideration these two
zooming in of the previous plant scheme (Figure 1).

Figure 8: VGO Unionfinning Unit (Reactor Section), DAVID S. J. JONES, PETER R. PUJADÓ, Handbook of Petroleum
Processing, Springer 2006

15
FRANCESCO MORANDI SELF-OPTIMIZING CONTROL

Which shows the “R”eactor part of the simplified diagram; and:

Figure 9: VGO Unionfinning Unit (Fractionation Section), DAVID S. J. JONES, PETER R. PUJADÓ, Handbook of Petroleum
Processing, Springer 2006

which represents the section dealing with the post proceeding of the product to respect the
specification and in order to be able to recover as much as possible the most expensive raw
material of all the process: the hydrogen.
Starting from the beginning it will be given a brief explanation of all the process without
focalized in particular on a specific kind of process, while considering the most general
class of the “hydrotreating process”. It obvious it has several different applications (e.g.
desulfurization, i.e. our specific case, olefin saturation, denitrogenation, and so on) and is
used for a variety of petroleum fractions from naphtha to atmospheric residue. But as
hinted previously, practically all units have the same flow scheme: it consists of a higher
pressure reactor section and a lower pressure fractionation section as we can observe
schematically in Figure 3 – 4.
We start now analyzing the reactor section of the plant proceeding then to the fractionation
part of the same.

16
FRANCESCO MORANDI SELF-OPTIMIZING CONTROL

2.3.1 REACTOR SECTION


This section is what concerns the major pieces of equipment; we have to consider in this
part feed/effluent exchangers, reactor charge heater, reactor(s), reactor effluent condenser,
products separator, recycle gas compressor and make-up gas compressors.
In some cases it is possible to find several other pieces of equipment: fresh feed filters,
reactor effluent hot separator and recycle gas scrubber. Each equipment is typical of a
peculiar configuration and is the best solution for that process. If we want to find the
above-mentioned equipment it is sufficient to look at Figure 3, i.e. the flow diagram of a
reactor section including all the equipment.
Now, to better understand each part of the equipment will be analyzed and briefly
explained.

2.3.1a Feed Filters


When we consider a unit that has to deal with a feed, the common way, that is preferred, is
feeding the feed directly from an upstream unit to the other one without going through
intermediate storage. Why? Because if we use storage facilities, feed filters are, or better
should be used in order to make the filter to retain the particulate matter (mostly corrosion
products) picked by the feed while in storage: that is a common problem linked directly
with the nature itself of the tank/pipeline material.
The feed filters are either automatic backwash filters operating on a pressure drop setting
or manual cartridge (disposable) filters. The difference is mainly due to durability of the
former respecting with the latter. Moreover the choice of one instead of the other is
connected with the process and with the degree of purity that has to be reached before the
processing.

2.3.1b Feed/effluent exchangers


Because in a generic plant economics is really important, the major voice in the costs list is
due to energy required to heat/cool; the most commonly used is heat recovery scheme,
where the reactor effluent in a series of feed/effluent exchangers preheats the reactor
charge before entering the reactor charge heater. In this way it is possible to recover as
much heat as possible from the heat of reaction, and then avoid to spend money to heat or
preheat the charge.

17
FRANCESCO MORANDI SELF-OPTIMIZING CONTROL

Depending on the configuration of the process scheme liquid feed may be preheated
separately with reactor effluent exchange before combining with the recycle gas. Reactor
charge heater is responsible to heat together gas and liquid feed to desired reactor inlet
temperature. In case of very heavy feed, i.e. the atmospheric residue units, the liquid feed
is preheated separately with reactor effluent exchange and only the recycle gas is heated in
the heater upstream of the reactor. In this way is possible to decrease viscosity and increase
the fluidity of the feed, making possible the subsequent contact between the liquid charge
and the gaseous reactant.

2.3.1c The make-up hydrogen system


As specified before hydrogen is one of the most important cost‟s voice in the economical
balance for a hydroprocessing plant. It is then obvious that the make-up hydrogen is really
important too. This line is usually obtained from H2 produced by units of naphtha
reforming or manufacturing plants, exempli gratia steam reforming 16.
Considering that hydrogen is usually kept in storage at pressure of 1 to 6 [bar], it is
necessary to compress this reactant till the desired operative pressure. In order to achieve
this result reciprocating compressors are used; they are known also as piston compressors,
i.e. they are a positive-displacement compressor that uses pistons driven by a crankshaft to
deliver gases at high pressure. After this procedure make-up gas could be introduced into
recycle gas system. This is what regards the make-up gas, but we have to understand that
not all the hydrogen fed to the reactor reacts as wanted: is then necessary to recycle some
of this.
After the separation of gas and liquid phases in the separator, the gas phase goes to the
recycle gas compressor. It is usually sent to an amine scrubber to remove the H2S produce
after the hydrogenation process in the reactor. The recycle gas compressor is often a

16
Typical naphtha reforming reactions dealing with the production of H2 are:
- The dehydrogenation of naphthenes to convert them into aromatics:

- The dehydrogenation and aromatization of paraffins to aromatics (commonly called


dehydrocyclization):

While speaking of steam reforming we usually refer to this kind of reaction, characterized by this
stoichiometry CnHm + n H2O ↔ (n + m/2) H2 + n CO

18
FRANCESCO MORANDI SELF-OPTIMIZING CONTROL

separate centrifugal machine; but it is not strange that it could also be a part of the make-up
gas compressors, as additional cylinders. Considering the large volume of gas this kind of
compressor is designed in order to pump this volume at a relatively low compression ratio.
Any other solution would be unfeasible unless excessive costs. After the reactor, the
product is composed by two phases: a liquid phase which contains the wanted product and
partially dissolved some gases; and a gaseous one which is rich of H 2S because of the
sulfur remove. If hydrogen sulfide remained in the recycle gas stream, it would reduce the
hydrogen partial pressure leading then to catalyst‟s activity suppression. It clear and quite
obvious that higher the sulfur content higher the effect on the catalyst, with a more
pronounced effect and for the same feedstock, the heavier the cut, the higher the sulfur
content is.
In a parallel sense, increasing the recycle gas hydrogen purity will decrease the catalyst
deactivation rate. This essentially happens because the sulfur compounds responsible for
the catalyst poisoning react with a greater probability and then they do not stay or deposit
on the catalyst‟s surface. Hydrogen purity is then function of the feedstock and type of unit.
Some measures those are able to increase the hydrogen content may include hydrogen
enrichment and/or membrane separation, which contributes to greater level of purity.

2.3.1d The reactor


As possible to see in the previous diagrams, the feed and the recycle gas (with its make-up
if needed) after being preheated till the desired temperature, they can be introduced at the
top of the reactor. The system can be assumed as a trickle bed reactor 17 in which the
downward movement of a liquid and gas happen over a packed bed of catalyst particles. As
the reactants flow downward through the catalyst bed, various exothermic reactions occur
and the temperature undergoes to an increase.
Different typologies of reactor design are available; the choice is usually on multiple
catalyst beds, with the possibility to add reactor in order to meet the goal of the process
considering heat of reaction, unit capacity and/or type of hydrotreating unit. In summa
specific reactor designs will depend upon several variables. Nonetheless it is possible to
specify some “thumb rules” those made possible the sizing of reactors, or better some
guidelines. For example reactor diameter is typically set by the cross-sectional liquid flux:
as the unit capacity increases, the reactor diameter increases to the point where two parallel
17
SATTERFIELD, C. N., Trickle-bed reactors, AIChE journal [0001-1541], year:1975 volume: 21, issue:2, pages:
209 -228

19
FRANCESCO MORANDI SELF-OPTIMIZING CONTROL

trains would be considered. At the same time reactor height is a function of the amount of
catalyst and number of beds required to achieve the objective of the process.
Then for what concerns the expected heat of reaction, cold recycle gas is brought into the
reactor at the interbed quench points in order to cool the reactants and in this way to make
possible to control the reaction rate.
A good distribution of reactants at the reactor inlet and at the top of each subsequent
catalyst bed is essential for optimum catalyst performance and to avoid the presence of hot
spots guilty of a non uniform heat distribution with the associated possibility to damage the
catalyst.
For every design there is a company that patent the internals. These internals are reactor
inlet diffuser, top liquid distribution tray, quench section which includes quench inlet
assembly, quench and reactants mixing device and redistribution tray, the reactor outlet
device. But it is not a rule that all reactors have the internals described previously. They are
a kind of precautions that make possible an improvement of the reactor performance
according to the charge and the operative conditions.
After the treatment in the reactor the charge has to be cooled and most of the cooling is
achieved in the feed/effluent exchangers; while final cooling of the reactor effluent is
obtained in air fin coolers and/or water trim coolers. Water is injected into the stream
before it enters the coolers in order to prevent the deposition of salts that can corrode and
foul the coolers: it is a trick that improve and increase the reliability of the overall process.
As briefly hinted before the main impurities are represented by the sulfur and nitrogen
contained in the feed and after the treatment they are converted to hydrogen sulfide and
ammonia. Because of the presence of these two compound with opposite behavior if we
consider their acidic or basic affinity (i.e. hydrogen sulfide is an acid, while ammonia is
well known for its basic effects), they combine to form ammonium salts18 that can solidify
and precipitate as the reactor effluent is cooled. In parallel, ammonium chloride may be
formed if there is any chloride in the system, due essentially to contamination of the feed
or to some corrosion byproduct. Water introduced has the double purpose to prevent their
precipitation keeping them in solution and transport them out of the reaction loop. There
are different patented systems about the quality of the water injection, but in general use of
boiler feed water is preferred.

18

20
FRANCESCO MORANDI SELF-OPTIMIZING CONTROL

The exact method of separating vapor and liquid will vary depending on the optimum heat
integration scheme. There can be used up to four separate vessels to disengage and
individually remove vapor, water and liquid hydrocarbon. In order to recover as much as
possible the heat in the plant, a hot separator is sometimes installed after the feed/effluent
exchangers to collect the heavier hydrocarbon material from the reactor effluent and send it
to fractionation via the hot flash drum.
The overhead vapor of this separator continues through an air cooler into a cold separator.
This kind of two separator system is at the basis of what is called: heat integration scheme.

2.3.2 FRACTIONATION SECTION


This section provide the plant to be able to separate the reactor effluent into the desired
products. This can be achieved with either a one or a two-column fractionation scheme,
depending on the type of hydrotreating unit and what is the feed charge to the reactor with
the sequent products.
Taking into consideration the two columns scheme, the flash drum liquids combine and go
to a stripper column. Here, steam and/or a fired heater is used to strip naphtha (if desired)
and lighter material overhead. This is an optimal way to recover lighter products. The
stripper bottoms go to a fractionator where it is further separated into naphtha (if desired)
and heavier products. The charge to the fractionator is typically preheated with a recovery
of the heat of the bottoms and with a fired heater before entering the column. As device to
strip lighter material up the column, steam is used; this is choice is usually made because
of the availability of this source in whatever plant, for economic reason and then because it
is easier to separate after having stripped the stream. Various product strippers could be
used to pull side-cut products to the desired specifications.

21
FRANCESCO MORANDI SELF-OPTIMIZING CONTROL

2.4 THE CHEMISTRY OF REACTION


When we speak generally of hydrotreatment we mean all processes those use hydrogen as
reactant to make possible some reactions; and depending on the impurities present in the
feed to treat it could be possible that some reactions either occur in parallel or they could
be happen because there are not the right conditions.
It will be shown now the main and more important reactions in a typical operation of
hydrotreatment, dividing them into different categories based on the impurities have to be
removed, remembering that the principal ones remain sulfur, nitrogen, oxygen and some
metals19. The previous three elements are quite always present in any charge feed to treat,
while there are some reactions which impact in the overall hydrogen consumption is
function of origin and cut of the feedstock.
Table 2: different hydrotreating processes and their products

Compound Process Product


Sulfur HDS – Hydrodesulfurization H2S
Nitrogen HDN – Hydro-denitrogenation NH3
Oxygen Oxygen removal H2O
Organo-metallic Hydro-demetallation [M]2+S2- *

Olefins Olefin saturation //  / **


Aromatics HAD – Hydro-dearomatization aromatics  naphtenes
Halides Halides removal halides  hydrogen hylides ***
* where [M]2+ is whatever metal which present a number of oxidation equal to two
** organic compound with double bonds are converted to their paraffinic homologue
*** a halide is a binary compound, of which one part is a halogen atom and the other part is an element or
radical that is less electronegative (or more electropositive) than the halogen, from WIKIPEDIA, THE FREE
ENCYCLOPEDIA, Halide, seen on February 2014.

It is quite clear, for example, that a feedstock as naphtha which is extremely depending on
the source of feed stock. For example, naphtha will contain extremely poor of organo-
metallic compounds will present a lower contribute by this reaction while an atmospheric
residue which is very rich in sulfur compounds will have a great consume of hydrogen due
to hydrodesulfurization. E.g., most middle eastern crudes contain much more sulfur than
some crudes from Indonesia or North Africa20.

19
DAVID S. J. JONES, PETER R. PUJADÓ, Handbook of Petroleum Processing, Springer 2006
20
Ibidem

22
FRANCESCO MORANDI SELF-OPTIMIZING CONTROL

Take into consideration that it is not a must that impurities have always to be present. Or
better for what concerns halides, aromatics and olefins they are function of the maturation
conditions of the crude; and in this case is important to consider the Francis‟ diagram:

Figure 10: Francis’ Diagram, GIULIO NATTA, ITALO PASQUON, PRICIPI DELLA CHIMICA INDUSTRIALE, MILANO: TAMBURINI, 1966

Where we can find and see the real amount of a certain compound in a determinate feed,
on the basis of its past conditions.
Considering the previous reactions, it is possible to ordinate them in increasing order of
difficulty: first of all we find metals removal, then the saturation of olefin, followed by
hydrodesulfurization, hydro-denitrogenation, oxygen removal and the last one, i.e. the
halides removal.
Because we generally speak of hydrotreatment we have to remember that all reactions
occur only in presence of hydrogen and all of them consume it. The contaminant removal
in residue hydrotreating involves controlled breaking of the hydrocarbon molecule at the
point where the sulfur, nitrogen or whatever impurity present is joined to carbon atoms. To
have an idea of how much it influence the charge, consider that cracking occurs in residue
normally less than 20% [vol/vol] of the fresh feed charge.

23
FRANCESCO MORANDI SELF-OPTIMIZING CONTROL

For what concerns the consume of hydrogen we can have this order of magnitude, that is
representative of the real impact of the impurity on the feed:
 the desulfurization reaction consumes 100–150 [SCFB/wt% change] or 17–25
[Nm3/m3/wt% change]
 and the denitrogenation reaction consumes 200–350 [SCFB/wt% change] or 34–59
[Nm3/m3/wt% change].
Taking into consideration the heat of reactions, that it is released because hydrogenation
reactions are always exothermic is partially compensate by the mass treated and it is about
0.02 [°F/SCFB H2], 0.002 [°C/Nm3/m3 H2].
A brief summarizing of what it has been showed till this point is therefore presented: first
of all considering the rates of reactions too, hydrodesulfurization and olefin saturation are
the most rapid reactions; then while the former is very quite heat-releasing, the latter
liberates the most heat per unit of hydrogen consumed. Denitrogenation and aromatic
saturation are the most difficult reactions and obviously hydrogen consumption is related
to heat of reaction are related.
Now it will be shown the main reactions take place in a hydrotreating reactor on the basis
of the different impurity element present in order to have a clear visual of what is produced
after the treating and as byproduct.

2.4.1 SULFUR
As mentioned before, the sulfur removal has as final product H2S of the organic sulfur
compounds present in the feedstock. This element is present in the petroleum fraction in
the form of many hundreds of different organic sulfur compounds which, can be classified
in six different categories 21 (considering the generic cut from naphtha to atmospheric
residue):
1- mercaptans
2- sulfides
3- di-sulfides
4- thiophenes
5- benzo-thiophenes
6- di-benzo-thiophenes
Now it will be shown an example for each class of compound pointed before.

21
DAVID S. J. JONES, PETER R. PUJADÓ, Handbook of Petroleum Processing, Springer 2006

24
FRANCESCO MORANDI SELF-OPTIMIZING CONTROL

2.4.1a Mercaptans
Known also as thiols22, they are a class of organic chemical compounds similar to alcohol
and phenol, but they replace oxygen atom with a sulfur one. They are also remarkable for
their characteristic scent, that make possible the traceability of natural gas too.
R – SH + H2  R – H + H2S

2.4.1b Sulfides
The term usually means all the three different classes containing the element sulfur 23 .
Those three classes are inorganic, organic and phosphine sulfides. The first one is
characterized by ionic bound where there is the negative charged sulfide ion S 2-; the second
one are compounds with sulfur responsible of covalent bond; while the last one are
chemical compounds derived by the combination with phosphine and sulfur: they have
both ionic and covalent properties.
R1 – S – R2 + 2H2  R1 – H + R2 – H + H2S

2.4.1c Di-Sulfides
As seen before in the previous paragraph sulfides are whatever organic compound
containing sulfur; when there are more than one sulfur atom contained in the compound,
then we refer to them as di-sulfides. For what concerns the reaction in which they are
typically involved, it is not so far from the sulfide‟s reaction:
R1 – S – S – R2 + 3H2  R1 – H + R2 – H + 2H2S

2.4.1d Thiphenes
They are the simplest sulfur-containing aromatic compound 24 . They are very similar to
benzene in its chemical and physical properties too. Thiophene is also present with
benzene in coal tar.

22
ENCYCLOPÆDIA BRITANNICA ONLINE, Thiol, www.britannica.com/EBchecked/topic/592252/thiol, accessed on
February 2014
23
ENCYCLOPÆDIA BRITANNICA ONLINE, Sulfide, www.britannica.com/EBchecked/topic/572504/sulfide, accessed
on February 2014
24
ENCYCLOPÆDIA BRITANNICA ONLINE, Thiophene, www.britannica.com/EBchecked/topic/592288/thiophene,
accessed on February 2014

25
FRANCESCO MORANDI SELF-OPTIMIZING CONTROL

2.4.1e Benzothiophenes
This sulfur compound present an aromatic ring with another ring in which there has been a
substitution with a sulfur atom25. For what concerns the properties of this substance, it is
really affine to naphthalene, from which it presents the similar odor. It is volatile with
steam and it is soluble in common organic solvents; lignite tar presents this compound as
byproduct of distillation.

2.4.1f Di-Benzothiophenes
They are the benzothiophenes corresponding as disulfide were for sulfides. The properties
of these compound resemble those of the above mentioned benzothiophene, even if it has a
higher specific weight and more chemical stability.

The mechanism of hydrodesulfurization of aromatic compounds has been widely


investigated and finally the supposed path of reaction has been postulated and even if it is
more complicated because of the different reactions that occur contemporarily; it could be
possible to highlight a “norm” which proceed through these two steps:
1- the removal of sulfur atom from the ring

2- the saturation of the corresponding olefin

25
THE MERCK INDEX ONLINE, benzothiophene, www.rsc.org/Merck-
Index/monograph/mono1500009455/thianaphthene?q=authorize, accessed on February 2014

26
FRANCESCO MORANDI SELF-OPTIMIZING CONTROL

As possible to understand the above mentioned classes of sulfur compound are not all easy
to remove: the ease of removal is exactly the way they are presented, starting from
mercaptans (the easiest one) to dibenzothiophenes (the most difficult to process). Moreover
we must remember that the ease of removal is function of the kind of fraction we mean to
treat: a heavier fraction is usually more difficult to process than a naphtha one. This is due
to the content of aromatics in the two different cuts.

2.4.2 NITROGEN
If sulfur was found in heavier petroleum fraction, nitrogen is present in the heaviest end of
these ones, in aromatic compounds (five or six member rings). Higher the boiling point,
higher the nitrogen content. If we consider the path of reaction the process occurs in the
same previous passages of hydrodesulfurization, but starting from saturation and then
proceeding through the nitrogen removal. The byproduct of this reaction is ammonia.
As example we can take into consideration the following compounds that are often found
in heavy fraction:

2.4.2a Pyrrole26
With this term it usual to indicate an organic compound of heterocyclic series where the
ring structure is composed by four carbon atoms and one of nitrogen. The simplest one is
pyrrole itself. They are very common substances and we can find them in amino acids,
alkaloids and other compounds produced by plants.

26
ENCYCLOPÆDIA BRITANNICA ONLINE, Pyrrole, www.britannica.com/EBchecked/topic/485130/pyrrole, accessed
on February 2014

27
FRANCESCO MORANDI SELF-OPTIMIZING CONTROL

2.4.2b Pyridine27
In this case this term is applied to any class of aromatic heterocyclic characterized by six
member ring structure, whose five of them are carbon while the last is nitrogen. It is
possible to find it in coal tar.

2.4.2c Quinoline28
In this case the term stands for any of a class of organic compounds f the aromatic
heterocyclic series characterized by double-ring structure, where one is the benzene, while
the other is a pyridine ring. It is important for the manufacturing of nicotine and it is
produced by coal tar.

As anticipated briefly before, the mechanism of hydrogenation of chemical compound


containing nitrogen is very difficult, it needs more hydrogen and it occurs with opposite
passages with referring to hydrodesulfurization process: first of all we have aromatic
hydrogenation, then hydrogenolysis and the last passage is the denitrogenation itself;
exemplified:

27
ENCYCLOPÆDIA BRITANNICA ONLINE, Pyridine, www.britannica.com/EBchecked/topic/484880/pyridine,
accessed on February 2014
28
ENCYCLOPÆDIA BRITANNICA ONLINE, Quinoline, www.britannica.com/EBchecked/topic/487446/quinoline,
accessed on February 2014

28
FRANCESCO MORANDI SELF-OPTIMIZING CONTROL

1- Hydrogenation

2- Hydrogenolysis

3- Denitrogenation

2.4.3 OXYGEN
Even if in minor measure, this class of impurities can affect some petroleum fractions. The
byproduct in this case is pure water. Some examples are shown :

2.4.3a Phenols

2.4.3b Oxygenates

29
FRANCESCO MORANDI SELF-OPTIMIZING CONTROL

2.4.4 AROMATIC SATURATION


In order to complete the description of the different reactions involved during the
hydrotreating it could be interesting to see also the hydrogenation of aromatic, because it
occurs often during this refining process and because it is cause of improving qualities in
petroleum fractions.

2.4.4a Toluene

2.4.4b Phenanthrene

2.4.4c Naphthalene

The other involved reactions undergo the same steps and they have as byproduct the
corresponding hydrogenated compound with the impurity as second element. As clear
example of the above mentioned reaction of halide, after the treatment we obtain the
wanted product without halogen element and as byproduct hydrogen bounded covalently
with the halogen element. Because they are not present in great amount in the reactions of
hydrotreatment of petroleum feedstock, no more examples will be given.

In order to make it possible these reactions occur, or better, to make it possible they happen
in acceptable time and with operative conditions able to guarantee the feasibility of the
process; these reaction proceed with the help of a catalyst. It is now possible analyze the
characteristics that have to be present in a good formulation of it and highlight how the
performances are influenced by catalyst.

30
FRANCESCO MORANDI SELF-OPTIMIZING CONTROL

2.5 THE CATALYST


Catalyst, this is the chemical definition 29 , is any substance that increases the rate of a
reaction without itself being consumed. In nature this happens every day in all biological
systems: enzymes indeed are naturally occurring catalysts responsible for many essential
biochemical reactions.
As any catalyst found in chemical engineering those for hydrotreating processes are
provided by a high surface area that is made by the active component (responsible for the
correct path of reaction, with lower values of activation energy) and a promoter (specific
for the reaction it has to aid), both dispersed on a support, that provides the mechanical and
structural characteristic to the catalyst itself.
The most common catalyst support is usually the so-called γ -Al2O3, with some small parts
of silica (SiO2) or phosphorous added; the choice is on this material because of its high
surface and an optimal pore structure.
As noted before hydrotreating catalysts contain metals dispersed on a support. That support
is γ -alumina which is arrived at by synthesis. The raw materials can be used to produce
the γ –alumina are essentially the sequent:
 Gibbsite (α-alumina trihydrate)
 Bayerite (β-alumina trihydrate)
 Boehmite (α-alumina monohydrate)

For what concerns the active component it depends by the type of the feedstock and
normally it is molybdenum sulfide, MoS2; but it is not strange to find some catalyst
compositions with tungsten, generally for special applications such as lube oil processing.
Taking into consideration the molybdenum catalysts, the typical used promoters are both
cobalt (CoMo) and nickel (NiMo). The main role of these compound added in different
quantity and with a particular dispersion is to “promote” the effectiveness of the catalyst
activity; while the acidity of the support can be increased to boost the catalyst activity for
(hydro)cracking and isomerization reactions.

Typical composition is about 25% [wt/wt] promoter and 25% [wt/wt] active component.

29
ENCYCLOPÆDIA BRITANNICA ONLINE, Catalyst, www.britannica.com/EBchecked/topic/99128/catalyst, accessed
on February 2014

31
FRANCESCO MORANDI SELF-OPTIMIZING CONTROL

Figure 11: typical shape of hydroprocessing catalyst, DAVID S. J. JONES, PETER R. PUJADÓ, Handbook of Petroleum
Processing, Springer 2006

Then depending on manufacturer there are different sizes and shape as you can note in
Figure 6.
Size and shape are not casual: the choice is made as in order to minimize pore diffusion
effects in the catalyst particles (and for this reason the requirement is small size) and
pressure drop across the reactor (requiring large particle sizes).
Although the physical characteristics vary from manufacturer to manufacturer and the
intended use of the catalyst (it depends on charge quality, operative conditions and so on),
a general guideline is this:
 High surface area = 150 [m2/g] or higher
 Pore volume = 0.6–1.0 [ml/g]
 Average pore radius = 30–100 [Å]
 Compacted bulk density = 35–55 [lb/ft3]
 Crushing strength = 4–20 [lbs/in2]
 Average length (not for spheres) = 1/8–3/8 [in]
We will now analyze and highlight some peculiar characteristics of some kind of catalysts.

2.5.1 COBALT-MOLYBDENUM
CoMo catalysts have been designed primarily for desulfurization, and for this process they
can achieve the best results, then denitrogenation and demetallation is also achieved. The
particularity of this catalyst is the ability to treat feedstocks of widely varying properties.
CoMo catalysts have the lowest hydrogenation activity, therefore they have the lowest
hydrogen consumption for a given sulfur removal. And if you consider that they also have
the lowest sensitivity of H2 consumption to changes in operating pressure, the CoMo

32
FRANCESCO MORANDI SELF-OPTIMIZING CONTROL

catalysts have the highest desulfurization performance at the lower operating pressures
(pressures are at about 40 [barg]).
Considering on the other side the denitrogenation performance, this is not achieve with
great results because they have a low hydrogen activity.
The best operative conditions are set at lower pressures and short hydrogen supply.

2.5.2 NICKEL-MOLYBDENUM
NiMo catalysts are the best choice for hydrogenation and denitrogenation even if they have
been designed for desulfurization too. Due to its nature this catalyst can achieve also metal
removal. In this sense NiMo catalysts have higher denitrogenation activities than CoMo
ones and are therefore used heavy fractions where the denitrogenation is as important as
hydrodesulfurization; this means for very heavy cuts and for the treatment of lube oil.
The performance of NiMo catalysts is very good at high pressures and differing from the
CoMo catalyst, NiMo show a greater response in denitrogenation and desulfurization
performance to changes in H2 partial pressure than the previous one.
High-pressure operations (e.g. FCC and hydrocracking feed pretreatment), could use and
prefer NiMo catalysts. Their use is also useful for reforming units pre-treating.

2.5.3 NICKEL-TUNGSTEN and NICKEL-COBALT-MOLYBDENUM


There are other different compositions and formulations for these catalyst; other two
catalysts used in hydrotreating are NiW and NiCoMo. The former has applications in
treating feeds where higher hydrogenation activity is required than is available from either
NiMo or CoMo; while the latter attempt to combine the benefits of CoMo and NiMo, even
if they are rarely used.
As general rule we have to remember that their desulfurization activity is poor at the
pressure levels used in hydrotreating; while NiW in sulfided form exhibits hydrocracking
activity surpassing that of both CoMo and NiMo.
Another improvements that can increase the hydrocracking activity can be done with
promoters or zeolite.
NiW catalysts can be made selective for saturating one of the double bonds in diolefins in
light feeds.

All the catalyst mentioned before based their manufacturing on one of these methods:

33
FRANCESCO MORANDI SELF-OPTIMIZING CONTROL

 Impregnation = where the support is first made and then treated by wet
impregnation (a solution with a dispersed solution of the active metals); the shape
can be either spherical or extruded
 Co-mulling = where we have the combination of both the metals and the dry
alumina in a mix-muller, followed by the extrusion of the wanted catalyst
 Hot soaking = as you can observe in Figure , the preparation it is made mixing the
“ingredients” at high temperature, in order to eliminate directly in the mixing
passage the solvent used to solubilize the active metals.

Figure 12: Hot Soaking, DAVID S. J. JONES, PETER R. PUJADÓ, Handbook of Petroleum Processing, Springer 2006

It is important to keep in mind the different variables those influence the final product;
among those we can find:
 Mixing intensity  influences the pore size
 Peptization  responsible of mesh of active metal
 Calcination  time, temperature and concentration
 Additives to mixing
 Metals application
 Solution preparation  contacting, time, order and drying
 Handling and screening

34
FRANCESCO MORANDI SELF-OPTIMIZING CONTROL

2.6 REACTORS
After having mentioned and deepened the catalyst preparation is time to see and analyze
the structure of what contains catalyst, the core of the hydroprocessing unit: the reactor.
Basically hydrotreating reactors are downflow (they exploit gravity force to make possible
the contacting of the fluid proceeding from the top of the
reactor till the bottom), fixed-bed catalytic reactors; and
for the nature of the reactant fed they generally operate in
trickle flow regime (we have the solid phase of the
catalyst, which have to treat a liquid charge with the
hydrogen which cause its supercritical behavior remains
in these conditions too, in gas phase).
Even if some catalysts can operate at middle pressures,
because hydrotreating instead occurs at moderately high
pressure and relatively high temperature, the choice for
the thickness of the vessels‟ wall is o high depth.
The shape of these equipment is the cylinder, typical
structure for almost all the reactors‟ vessels and as usual
from the beginning of the “refining era”.
The material instead has undergone a change from the
older reactors (e.g. for hydrotreating of naphtha) made by
lower alloys, to the last 10-years equipments whose raw
materials are typically 11/4 Cr –1/2 Mo or 21/4 Cr – 1 Mo
base metal with a lining of stabilized austenitic stainless

Figure 13: typical vessel shape, DAVID S. steel for added corrosion protection (we have to
J. JONES, PETER R. PUJADÓ, Handbook of
Petroleum Processing, Springer 2006 remember that the byproduct of these treatment is almost
always acid compound which could easily damage and
disrupt the same vessel or corrode the reactor itself). With this choice, there is the
possibility to give strength to the base and at the same time the excellent corrosion
resistance.
The choice is made considering different items; and even if concerning the use of
austenitic stainless steels in hydrotreating units, the possibility of corrosion is lowered, it
exists for corrosion cracking instead if the proper procedures are not followed.

35
FRANCESCO MORANDI SELF-OPTIMIZING CONTROL

This kind of corrosion can occur through chloride attack or polythionic acid attack. The
former can be prevented by minimizing the amount of chloride in the process material,
eliminating it immediately from the process lines. Then to improve the results, during
startup and shutdown operations precautions should be taken to limit the chloride content
in any flushing, purging, or neutralizing agents used in the system.
The latter occurs as the result of the action of water and oxygen on the iron sulfide scale
that forms on all items made of austenitic stainless steel.
If one of these previous two processes happened, it causes intergranular corrosion and
cracking.
The polythionic acid attack is possible to prevent maintaining the temperature above the
dew point of water in those areas containing stainless steel: this avoid water to condensate
on the surface of the reactor. For safety reasons and because there would not be the
necessity the system is ideally without oxygen. However, when the system is depressurized
and the equipment is opened to air, in order to permit the normal maintenance operations it
becomes necessary to maintain a nitrogen purge to prevent air from entering (or whatever
neutral purge in order to avoid contacting oxygen with the surface of vessels and also
because depressurizing and lowering the temperature it could be possible to make water to
condensate on the inner surface with the possibility of polythionic corrosion).
If this is not possible, a protective neutralizing environment should be established: a 5%
soda solution is used to neutralize the austenitic stainless steel.

The bed of catalyst is function of the feed treated: most naphtha hydrotreaters only have
one catalyst bed; but it is not unusual that many reactors processing cracked feedstocks
will have several beds to facilitate temperature control by cooling with hydrogen quench
between the catalyst beds.
The reactor vessel is designed to allow maximum utilization of catalyst; and this done by
creating equal flow distribution (no preferential channel), providing maximum liquid/vapor
mixing, and providing multiple beds with quench zones for efficient catalyst use.

The internals of the reactor usually have some devices useful to improve the reactor
characteristics, even if it is not necessary they have to have. Some of them are:
 Inlet diffuser
 Top vapor/liquid distribution tray

36
FRANCESCO MORANDI SELF-OPTIMIZING CONTROL

 Quench section  only when there are multiple catalyst beds


 Catalyst support grid  only when there are multiple catalyst beds
 Outlet collector

Considering the size of the hydroprocessing reactors they vary widely on the basis of
design conditions, the LSHV and the acceptable pressure drops (function of diameter). An
important key factor to achieve the best performances is to control the temperature of the
reactor; the maximum allowable T is about 42[°C]. If that temperature is not expected to
be exceeded, a reasonable solution could be reactor mono bed and temperature control will
be exercised by changing the reactor inlet temperature.
On the other hand if the maximum reactor difference of temperature is expected to exceed
the previous value, a multiple bed reactor should be installed, where the coolant is the
hydrogen which quenches the charge inserting in the quench section of reactor (inter bed).
Considering the operative conditions of a classical hydroprocessing reactor, the
hydrotreating catalyst gradually loses some of its activity. It is a direct consequences of
fouling, sintering and coking. But in order to maintain the desired quality of the products at
the design feed rate, the solution consists in increasing the bed temperature, about 1
[°C/month].
Once the bed temperature reaches a value close to the design maximum, there no other
solution than replacing the catalyst (or regenerate it). The catalyst average lifetime is about
several years. More often the replacement or regeneration of the catalyst is necessary
because of the pressure drop due to fouling.
At the base of all these discussion there is kinetics; kinetics is the study of the rates of
reaction; these rates of reaction determine the key properties of a catalyst. In fact as
mentioned before the role of the catalyst is to reduce the activation energy, providing a
reaction path different from the thermal activation one. In hydrotreating, the key properties
are initial activity, stability, and product quality.
The temperature required to obtain the desired product at the start of the run measures the
initial activity (it is different from the use of a catalyst and without it). Catalyst stability is
a measure of change of reaction rate over time and the product quality is a measure of the
ability of the process to yield products with the desired use specification.

37
FRANCESCO MORANDI SELF-OPTIMIZING CONTROL

Chapter 3
Process Design: layout and modification

“Patterns”
Simon and Garfunkel
3.1 INTRODUCTION
Having the theoretical basis to understand how the process works and his historical
background, we can now introduce the process design path that has been adopted to
simulate it.
In this chapter it will be shown criteria and some assumptions that have made possible the
best design accordantly with the simulation program used and with the data available from
literature.

3.2 LITERATURE BACKGROUND


As mentioned before for a good design we necessary have to had a solid basis on which
have the possibility to build a realistic project.
The starting point is the literature research: from this it has been possible obtaining
different designs which have been evaluated and tried in order to select the most
appropriate model. In the same way a literature search was moved in order to find the best
kinetics able to describe the HDS processes.
In both cases the more realistic compromise has been chosen among the possible ones: the
choice has been made taking account for REALISM and ACCURACY of RESULTS,
CONVERGING TIME and REPEATABILITY between the different optimization runs.
With these evaluation criteria the project has been modeled.

3.3 PLANT SCHEME from LITERATURE


Hydrotreating processes has a specific goal: they aim at removing impurities, usually are
represented by sulfur (not by a change a distinction among oils is made for their sulfur
content) and nitrogen, from distillate fuels (naphtha, kerosene and gas oils). In order to do

38
FRANCESCO MORANDI SELF-OPTIMIZING CONTROL

this they treat the feed with hydrogen in a very high condition of severity: elevated
temperature and pressure in the presence of a catalyst.
Because in recent years hydrotreating has been extended to atmospheric residues in order
to increase their commercial value (decreasing sulfur and metal content essentially), that is
the upgrade of the feed; we obviously have that operating conditions of this kind of
treatment are a function of FEED TYPE and of desired treated product
DESULFURIZATION‟s LEVELS.
Taking account for these points we will have a wide range of different configurations of an
HDS plant.
From literature the usual feed types considered are NAPHTHA, KEROSENE, GAS OILS
and ATMOSPHERIC RESIDUES or REDUCED CRUDES in crescent grade of severity
of the treatment.
It will be now shortly present how scheme plants differ one from another based on the feed
type fed to the system. Then the focus will be point on GAS OIL plant scheme introducing
and explaining all changes made in order to set the best configuration possible and most
realistic one.

3.3.1 NAPHTHA HDS UNIT

Figure 14: Surinder Parkash , Refining Processes Handbook, Gulf Professional Publishing, 2003

39
FRANCESCO MORANDI SELF-OPTIMIZING CONTROL

3.3.1a Process Description


Naphtha (or full range naphtha) is defined as the fraction of hydrocarbons in petroleum
boiling between 303 [K] and 473 [K]30. Its composition consists of a hydrocarbon mixture
generally having between 5 and 12 carbon atom. It typically constitutes 15–30% of crude
oil, by weight.
Hydrodesulphurization of naphtha is done to make the feed suitable for subsequent
treatment, among those: catalytic reforming to improve octane or steam reforming in order
to produce hydrogen. In the first case, this takes the name pretreatment.
The process consists of treating the feed with hydrogen in the presence of a catalyst under
suitable operating conditions. Because its physical properties and being one of the lowest
petroleum cuts it is possible operate in less drastic conditions.

In this sense, following the previous chart, the naphtha feed enters the unit through charge
pump P-101. The petroleum feed is then mixed with hydrogen gas coming from a
catalyzed reforming unit or hydrogen plant. The mix is heated in succession through heat
exchange with reactor effluent (E-101) and fired heater (H-101). Now it‟s time for the mix
to enter the HDS reactor V-101. The desulfurization reactions take place over a Co-Mo on
-Al2O3 catalyst contained in the reactor. At the out of the reactor we have: the
desulfurized naphtha, excess hydrogen, H2S, and light end elements formed as a result of
reaction of sulfur in the feed with hydrogen in the presence of the catalyst.

The reactor effluent is then cooled and partially condensed through the previous heat
exchanger (E-101), an air cooler E-102, and a trim cooler E-103, before flowing into high-
pressure (HP) separator V-102. Here we obtain the separation between the vapor and liquid
phases. The vapor from this drum (essentially made by H2S, light hydrocarbons and excess
hydrogen) is purged and sent out of unit's battery limit. The liquid from V-102 is preheated
through a heat exchanger with the stripper bottom in heat exchanger E-105 before charging
to the stripper column V-103, where we have the separation between the dissolved H2 and
H2S as overhead product. The gross overhead product from V-103 is then partially
condensed (E-104). The uncondensed vapor, that is essentially made by H2S, is sent to an
amine unit. The liquid is responsible for the reflux of the column (P-103). The column is
reboiled by a fired heater H-102. P-102 is the pump used for bottom recirculation. The
30
PRESTVIC, RUNE; KJELL MOLJORD, KNUT GRANDE, ANDERS HOLMEN ,"Compositional analysis of naphtha and
reformate", 2004.

40
FRANCESCO MORANDI SELF-OPTIMIZING CONTROL

bottom product coming from the stripper is cooled by heat exchange with incoming feed in
E-105 and next in E-106 to 811 [K] before exiting the unit's battery limits.

Table 3: Operative Conditions Naphtha Unit, SURINDER PARKASH , Refining Processes Handbook, Gulf Professional
Publishing, 2003

OPERATING PARAMETERS UNITS


Reactor Inlet Temperature
SOR 593 [K]
EOR 643 [K]
Total pressure at separator drum 2.1 [MPa]
H2 partial pressure (reactor outlet) 1.1 [MPa]
Liquid Hourly Space Velocity 4.00 [h-1]
H2 consumption 45 [scf/bbl]

Note that a naptha hydrotreater (NHDT) or “light naphtha isomerization plant” is projected
to operate at specified duty under a set and range of conditions. With the progress of time
the catalyst gets deactivated and temperature is increased to compensate the deactivation.
The period between SOR and EOR is called cycle length. The lowest temperature at which
plant is commissioned to yield desired product quality is called SOR. When the catalyst
has reached the limit before having to be regenerated, we say that it has achieved EOR.

41
FRANCESCO MORANDI SELF-OPTIMIZING CONTROL

3.3.2 KEROSENE HDS UNIT

Figure 15: Surinder Parkash , Refining Processes Handbook, Gulf Professional Publishing, 2003

3.3.2a Process Description


As every HDS plant, the objective of kerosene hydrotreating is to upgrade raw kerosene to
produce products by specification suitable for market as kerosene itself and jet fuel.
The will of removing sulfur and mercaptans from the raw kerosene cuts (crude distillation
unit) lays in possible corrosion‟s problems in aircraft engines and fuel handling and storage
facilities. Beside sulfur‟s problem we have to handle the content of nitrogen in the raw
kerosene feed that can cause color stability problems.
For aviation turbine fuels (ATF), we have very specific rules; e.g.: the ASTM distillation,
flash point, freeze point and smoke point (which is a function of the aromatic type
hydrocarbons in the cut: higher aromatic content yields lower smoke point kerosene cuts;
that is we can control aromatics content by severity of the treatment).
This purpose is reached by columns‟ distillation in order to remove gases, light ends, and
heavy kerosene fractions. The upgrading is achieved by treating the feed and the hydrogen
in the presence of a catalyst, where sulfur and nitrogen compounds are converted into
hydrogen sulfide and ammonia. For the same reasons explained before, the ATF product
can have only straight run kerosene or hydrotreated blend components.
In order to understand how it‟s possible to do this we can observe the previous diagram
and analyze the process itself. Kerosene feed is pumped via charge pump (P-101) and
preheated in a heat exchanger E-103 and then followed by final heating in fired heater H-
101. This effluent next joins the recycle hydrogen coming from compressor C-101 and is
heated again in the heat exchanger E-102, subsequently by a fired heater H-102. The
heated kerosene feed and hydrogen mix flow through reactor V-101 (loaded with a Co-Mo

42
FRANCESCO MORANDI SELF-OPTIMIZING CONTROL

or Mo-Ni catalyst), where the reactions of hydrodesulfurization and hydrodenitrification


take place.
Because these reactions are exothermic, the effluents can be used in heat exchanger (E-102,
E-103 and E-104). In E-105 (air cooler) we have a further cooling before the flash in HP
separator (V-102), where the temperature is about 1000 [K]. Vapor phase (rich in
hydrogen) is compressed and sent back to the reactor by the compressor (C-101). The
hydrocarbon liquid from the same separator is flashed in drum V-104.
The flash gas is sent to the amine unit for H2S in order to remove it, while the liquid from
the flash drum is sent to a stabilizer column V-105. The overhead vapor of the column is
partially condensed in air cooler (E-106) and flows into accumulator V-106. Here the
liquid is partially returned as reflux to the column, while the remaining part is kept as it is
(wild naphtha). The bottom of the column is sent to fractionator column V-107, where we
cut a high flash naphtha as overhead product. The other cuts are withdrawn as sidestream
(e.g. light kerosene base stock). We have a further passage in side stripper V-108 in order
to adjust the flash point of the kerosene. The stabilizer column has its reboiler (a fired
heater H-103). The aviation turbine kerosene is withdrawn as an overhead product from the
splitter column V-110. It‟s heated by H-106 (fired heater) and the bottom product is cooled
in an air cooler E-114 and then in a water trim cooler (E-115). The product can be used as
it is or mixed in order to obtain different blends or cutter for various oil grades.
Table 4: Operative Conditions Kerosene Unit, SURINDER PARKASH , Refining Processes Handbook, Gulf Professional
Publishing, 2003

OPERATING PARAMETERS UNITS


Reactor Inlet Temperature (W.A.B.T.)
SOR 593 [K]
EOR 643 [K]
Total reactor pressure 9.3 [MPa]
H2 partial pressure 7.6 [MPa]
Reactor T 20 [K]
Reactor P 0.2 [MPa]
H2 consumption 555 [scf/bbl]
Recycle Ratio 3072 [scf/bbl]
% Desulfurization 99.6
% Denitrification 98.0

43
FRANCESCO MORANDI SELF-OPTIMIZING CONTROL

3.3.3 GAS OIL HDS UNIT

Figure 16: Surinder Parkash , Refining Processes Handbook, Gulf Professional Publishing, 2003

3.3.3a Process Description


Gas oil hydrodesulfurization is designed to reduce the sulfur and other impurities (e.g.
nitrogen, heavy metal and other different compounds) present in the raw gas oil cuts.
The feed to the unit could be very different: from a straight run diesel cut coming from the
crude distillation unit to a secondary units such as FCCU (Fluid Catalytic Cracking Unit)
or delayed coker. These last kind of feed may contain significant amount of olefinic
hydrocarbons, which has to be converted to saturates in the diesel hydrotreating unit
(stability problems during the storage).
As for the previous charts the primary improvement in product quality is with respect to
sulfur and conradson carbon (raw diesel cut from Middle Eastern crudes may contain as
much as 1-2% sulfur). Until a few years ago 1.0% sulfur was acceptable, but in the last
years for the extraordinary rapid increase of diesel vehicles, the pollution level has risen
exponentially; so sulfur specifications in most developed countries have fallen to 0.005
wt% or lower.
In the diagram we will analyze how to proceed a gas oil feed and to treat it in order to join
the specifications.
The diesel feed is pumped by pump P-101 to heat exchangers E-103 and E-101, then it
joins the hot recycle hydrogen stream before entering reactor V-101 (it‟s loaded with the
catalyst, Co-Mo or Ni-Mo on alumina type). Note that the recycle gas, from compressor C-
101 is heated in heat exchanger E-102 and then in a fired heater H-101; subsequently it‟s

44
FRANCESCO MORANDI SELF-OPTIMIZING CONTROL

mixed with the hydrocarbon feed before going to HDS reactor V-101. In the reactor we
have the common hydrodesulfurization and hydrodenitrification reactions.

Cause the exothermicty of the reactions of hydrodesulfurization, a cool hydrogen quench is


added to the inter bed areas. When the reaction mixture exits the reactor, it is cooled in a
series of heat exchangers from E-101 to E-104 and next in air cooler E-105. Once the
temperature is significantly lowered it is flashed into high-pressure separator (V-102).
Condensate is injected into the reactor effluent just before air cooler E-105 to dissolve and
remove ammonium salts, that result from the reaction between ammonia and H 2S
(problems in pipes for possible blockage of them). The separation of ammonium salt
solution happens in high pressure separator drum V-102 and it‟s recovered in the refinery
wastewater system.
The hydrogen-rich gas from the HP separator (almost pure hydrogen), enters H2S absorber
V-107, where with an amine washing (diethanolamine, DEA, solution) the H 2S is removed.
The H2S-free gas is mixed with makeup hydrogen coming from the hydrogen plant. We
have and intermediate compression made by centrifugal compressor C-101. As stated
before part of the recycled gas is used as an inter bed quench in the reactor, while the
remaining hydrogen (heated in heat exchanger E-102 and fired heater H-101), joins the
hydrocarbon feed before entering the reactor V-101.
The liquid hydrocarbon stream from separator V-102 is flashed in V-103. The flash gas is
sent for H2S removal before going to refinery fuel system, while the bottoms liquid stream
from V-103 is preheated in the reactor heat exchanger E-104 before flowing into the
stabilizer column V-104. This column separates the gases, light ends, and naphtha formed
as a result of HDS reactions in the reactor. The stabilizer overhead vapors are condensed in
air-cooled exchanger E-109 and then sent to accumulator drum V-105; while this liquid
(the so called unstabilized naphtha) is used as a reflux to the column itself, and the excess
is recovered as naphtha product. Heat is supplied through medium-pressure steam. The
stabilizer bottoms is the desulfurized diesel product. The product is cooled in heat
exchangers E-110 and E-111 before being stored.

45
FRANCESCO MORANDI SELF-OPTIMIZING CONTROL

Table 5: Operative Conditions Gas Oil Unit, SURINDER PARKASH , Refining Processes Handbook, Gulf Professional
Publishing, 2003

OPERATING PARAMETERS UNITS


Reactor Inlet Temperature 614 [K]
Reactor T 30 [K]
Reactor inlet pressure 15.7 [MPa]
H2 partial pressure 11.9 [MPa]
Reactor P 0.2 [MPa]
H2 consumption 422.3 [scf/bbl]
Recycle Ratio 5065 [scf/bbl]
% Desulfurization 98.9
% Denitrification 73.9

3.4 PLANT SCHEMES


We start now the study of the project plant and the modifications put in act in order to
simulate the most realistic possible a HDS unit. The project has been built in Aspen
Hysys®.
For what regards our study the goal has been focused on the gas oil plant. For the most
realistic project as possible a model from literature has been selected and then
subsequently modified in order to make it possible a good design with the possibility to
have a feasibility in the modeling.
The essential modifications applied to the model from literature are the sequent:
- Heat-exchangers
- Stabilization column
- Kinetics
For what concerns the first two modifications the main reason has been to make as easy as
possible the optimization of the plant: all the equipments have been evaluated based on
their real effectiveness on the plant specification and on their impact on optimization time.
Each train of heat-exchangers make the calculation slow in terms of numerical solution,
they burden cause every loop forces the simulation program to calculate (in the built-in
algorithms) energy balances, pressure drops and velocities. The stabilization column
instead is guilt to an inadequate solution, when it has been found; because of a instability

46
FRANCESCO MORANDI SELF-OPTIMIZING CONTROL

in evaluation counts, while the optimization program is running and it‟s changing the
initial evaluation point. The first has been substituted either with simple heater or cooler.
In other words we have to considered what is extremely necessary to make a process plant
feasible without losing reliability or without losing ourselves in too much expensive
calculations.
On the same idea are based the considerations on the kinetics and the problem here is
purely numerical; as we will analyze next, a set of different kinetics equations 3132 has been
tested and compared with data on literature in order to see what is the best solution with
regard on time and accuracy.
Now in order to understand what is the real reason for what we have so much interest in a
“speedy” evaluation we have to analyze our purpose and in particular we have to
remember that the so-called “Self Optimizing Control” is a technique developed by Prof.
Sigurd Skogestad dealing with the possibility to operate a solid and robust control on a
selected plant having chosen some “magical” variables that even if we have some
modifications on the feed or other external parameters (so called disturbances) they don‟t
get far from the optimal operating point. In order to find this theoretical optimal point (and
in this case we have to point out that we speak about an optimal point that is the best
considering the whole parameters affecting the simulation in that moment. It‟s the optimal
solution with that set of variables and disturbances but it‟s always the optimum. A wide
speech could be faced if we consider the problem of a “moving optimum” or something
that would have to update every iteration with every different set of parameters. It could be
a philosophical digression on what‟s the real meaning of optimal solution); a Matlab®
code has been written in order to find the best solution at a given set of parameters,
working on what has been selected as the best approximation of an HDS scheme plant.
The basis on which the scheme has been selected is a Gas Oil scheme plant whose design
has been found in literature. The project built in Aspen Hysys® and has been modified in
different versions in order to make the finding of a feasible solution possible with the
Matlab® code.

31
GILBERT F. FROMENT, GUY A. DEPAUW, AND VALÉRIE VANRYSSELBERGHE, Kinetic Modeling and Reactor Simulation in
Hydrodesulfurization of Oil Fractions, Amercan Chemical Society 1994.
32
AYSAR T. JARULLAH, IQBAL M. MUJTABA, ALASTAIR S. WOOD, Kinetic parameter estimation and simulation of
trickle-bed reactor for hydrodesulfurization of crude oil, Chemical Engineering Science 2011, vol. 66, pp.
859-871, 2011

47
FRANCESCO MORANDI SELF-OPTIMIZING CONTROL

3.4.1 FIRST MODEL: HDS from literature


As seen previously literature provides us some valid models and plant projects that are
representing the common way to process a certain kind of feed and treating it to remove
undesired compounds.
In this case, to make it possible reliable configuration of a scheme plant, it has been chosen
the one dealing with the treatment of a gas oil feed.
This layout of plant scheme is complete with all equipments and is presented in its first
realization on Aspen Hysys® program:

Figure 17: Aspen HYSYS® model adapted from Surinder Parkash , Refining Processes Handbook, Gulf Professional
Publishing, 2003

As explained before we can subdivided the HDS process in two main loops: the part
dealing with the hydrogenation (and then we consider the part with the reactor and its train
of heat exchangers in order to make it possible the reaching of operative conditions for the
reaction); and the post treatment of H2S as by-product of the reaction and its treating in the
amine washing loop and the regeneration of this last one.
As you can see in the picture before we have both the part regarding the real hydrogenation
process (the heart of the HDS treatment) and the loop of amine treatment-regeneration.
In the final part it has been also introduced the stabilizer column in order to obtain the
different specification products.

3.4.1a Considerations
Even if this is a project working and reliable in its features we have to consider two main
things, which will be our guideline during all the optimization analysis of the plant:

1. CALCULATION TIME: in the perspective of an optimization evaluation on all the


plant, more variables are considered (that is more equipments in the scheme) more

48
FRANCESCO MORANDI SELF-OPTIMIZING CONTROL

time is spent on calculations on every loop made by simulation program to


converge to a numerically correct and feasible solution while the optimization
program is running to find the optimum that is the optimal solution. We could
assert that we two calculation loops, one subjected to the other, both subjected to
our specification made by feasibility of the solution. That is why we want to obtain
the fastest solution possible from the simulation program because it is moved by
the optimization program on solutions surface in order to find the optimum (in this
sense we could see it as a master-slave controller: some control systems arrange
PID controllers in cascades or networks, that is a "master" control produces signals
used by "slave" controllers).
2. ACCURACY: and in this sense we want to underline the possibility to obtain
solution with a certain grade of compatibility with specification we set at the
beginning of each simulation.

As we can observe directly on the plant scheme we note first of all the complexity present
among the different parts of the plant itself; we could individuate the two main loops and
we can analyze them differently: the main loop where we have the reactor and the reaction
takes place is characterized by a series of train of heat-exchangers whose duty is to raise
the feed temperature till almost the reactor level and operate a thermal recover from the
product line in order to decrease the costs of the calories in all the process. Then we see the
loop of amine treating and the regeneration of the same itself: it‟s particularly highlighted
by the presence of the scrubber and by the regeneration column which is responsible both
for water make-up and indirectly for the H2 make-up, if we consider that higher is the
efficiency of an amine washing higher is the amount of H 2 that could be recycled to the
reactor, that means less cost of raw materials in that case H 2 itself.
Although this scheme is incredibly faithful to reality, we face with two problems: the first
is that mentioned before of a too expensive calculation time. By the clock if we change
some input data (in this case fixed the composition of the feed and having set all the
parameters regarding temperatures and pressures allowing the simulation program to
evaluate the remaining ones) we can see that every try takes about one minute (45/50 [s])
in order to reach the solution, or better in order to converge to a solution. Considering this
particular we can note that if we have the necessity of changing in a range of 1000 of

49
FRANCESCO MORANDI SELF-OPTIMIZING CONTROL

values per variable set to be optimized, calculation time reaches too high values without
having the certainty of a correct solution (in terms of optimized solution).
The other problem is represented by the stabilization column. While calculation time,
although high and unfeasible if we want to make an optimization usable directly in a RTO,
always bring to a solution; the equipment represented by the column is often guilt of a non-
correct solution or in some cases it doesn‟t make it possible the finding of a solution.
Aspen HYSYS® column environment is really susceptible of initial value and initial
estimation of the solution of the column itself and great changing in them make impossible
finding A solution. Moreover we have to take in account that for the same reason it has a
environment apart, counts to reach the solution are made in a loop apart. In this sense
besides having an unfeasible or impossible solution it could take too much time without
finding it.
The approach followed to build a realistic model and at the same time feasible in terms of
time taken for evaluation and solution, has been based on the assertion “keep it easy”.
Starting from the simplest plant design, it has been added different equipment to make it
realistic and at the same time without prejudging solution time.
The different version of the scheme will be shown and briefly commented, regarding with
the modification added and parts substituted.

3.4.2 THE SIMPLEST MODEL: “naked” HDS

Figure 18: the easiest way to configure an HDS unit

Starting from the simplest one, without considering the two mentioned before loops,
excluding all equipments more complex than a vessel, this is the result. It‟s a poor scheme

50
FRANCESCO MORANDI SELF-OPTIMIZING CONTROL

plant whose core is the reactor and where we had the stripper and the scrubber their place
has been substituted by two splitter.
Even if this work could seem useless and without any sense, we can observe instead the
real effectiveness of every single part of a chemical unit and its weight in a more complex
system.
In this case amine washing has been changed with a splitter and instead the regeneration
we split the fractions of H2S and H2 in order to send the last one to the reactor.
This preliminary analysis has underlined how important is the loop regarding scrubber and
its regeneration. The calculation time has dropped down reaching 1.2 [s] at the maximum,
while the solution was always acceptable.

Stripper itself has been changed with a splitter and we have a kind of flash in this section.
Also heat-exchangers train has been taken away, keeping only the product/feed one.
The compressor absolve the role to compensate pressure drops and no recycle as been
introduced.
The system is working and data coming from the exit of the reactor are acceptable.
This isn‟t a real unit yet. It has been useful to evaluate and understand at the basis the
process and how the different level of detail could affect the behavior of a chemical unit. In
this simple case we can analyze the role of the reactor and how it responds to different
changing in feed flow or inlet reactor temperature or different pressure/pressure drops. It‟s
a preliminary analysis on one of the two loops involved in the real process. Pointing out
how important is the choice of a right kinetics, because every added equipment has been
set to the simplest one possible.

3.4.3 THE FIRST MODIFICATIONS

Figure 19: HDS unit with a previous selection of the desired layout

51
FRANCESCO MORANDI SELF-OPTIMIZING CONTROL

Having clear in mind how the process and the reaction takes place, some little
modifications have been inserted in order to control and set some variables; in particular it
has been put the recycle of H2 to reactor and the set for the pressure assuming that the
major contribute of pressure drop is only in the reactor (as documented in literature 33) and
obviously in HP separator which acts like a flash at high temperature. Ideal behavior of
heater and cooler has been hypothesized that is in this case no pressure drops.
For what concerns scrubber and stripper the two units has been simulated treating them as
splitter as the case before.
In this simulation the importance was put on the recycle that has to make it converge in
order to evaluate correctly the necessary supply of H2 in reaction loop and directly the
make-up.
Conditions, taken from literature too, are what resumed in Table 3.
The composition is a typical gas-oil blend with a sulfur content variable from 1 to 2 wt %34.

3.4.4 AMINE LOOP STRIPPER

Figure 20: HDS unit with the regeneration unit

After all considerations regarding the reaction loop and its recycle, it has been possible
introduce some modification useful to make more realistic the model. In this sense the two
splitters set before instead of scrubber column and stripping one, it has been inserted the
“amine loop” and a simple flash separator with a steam stream to supply the necessary duty
to make possible a best separation between the gas-oil stream and gas phase of non-
condensable.

33
SURINDER PARKASH , Refining Processes Handbook, , Gulf Professional Publishing, 2003.
34
1.82 wt %; GILBERT F. FROMENT, GUY A. DEPAUW, AND VALÉRIE VANRYSSELBERGHE, Kinetic Modeling and Reactor
Simulation in Hydrodesulfurization of Oil Fractions, Amercan Chemical Society 1994.

52
FRANCESCO MORANDI SELF-OPTIMIZING CONTROL

When we speak using the expression “amine loop”, we want to underline the importance of
this sub-unit of an HDS plant for what concerns the specifications we want to obtain; such
an importance that we can describe a self standing loop. We must note in this sense that
amine loop is all the equipment tied to scrubber and amine washing: starting from scrubber
column, in cascade we have a flash separator then the regeneration column with the make-
up of water lost as vapor in gas phase in head stream.
It‟s important to make clear the reason for such a particular attention given to this part of
the plant. In the amine loop we recover H2 that is sent then to the recycle in order to keep
the reaction at its optimal value. Therefore it‟s evident that if we have a correct workable
unit in this section we can achieve best results in all the plant.
In order to design a correct project for this loop, the starting point has been taken from
tutorials available for this commercial software 35 . With this raw scheme it has been
possible realize a workable unit useful to simulate the correct behavior of this loop.
For what concerns the amine used in the amine-washing unit, it has been selected DEA
instead MEA for its major absorption capability; while operative conditions adopted for the
simulation have been selected in order to achieve the best result 36 . These value are
comparable with the literature ones used for the plant project.
Stripper column, instead, has been substituted in the simulation with a simple flash whose
duty has supplied by an external source. The main reason why this solution has been
selected is found in the potential instability during calculations introduced by a stripper
column. Generally besides problems dealing with slowness of every iteration, we have to
face the possibility of some numerical inconsistencies and subsequently a non feasible
solution.
Before doing this radical simplification, different results in specification has been taken in
consideration and compared in order to see if it was acceptable or it could be result in
something completely wrong. For what concerns the concentration in sulfur compounds in
gas oil desulfurized stream we have the same results, within a less than 1% of tolerance. In
this way it has been possible to build a realistic and faster model than if we adopted the
real configuration.
Global operative conditions are the same seen for the previous models.

35
HYSYS 3.0, HYPROTECH LTD., 2002.
36
LUKE ADDINGTON AND CHRIS NESS, An Evaluation of General “Rules of Thumb” in Amine Sweetening Unit
Design and Operation, GPA 89th Annual Convention March 21-24, 2010.

53
FRANCESCO MORANDI SELF-OPTIMIZING CONTROL

3.4.5 SETTING PRESSURES and TEMPERATURES

Figure 21: HDS unit with the optimal configuration of the equipment

At this point, it has been useful to introduce some small details in order to simplify the
procedure of optimization of the plant. As it can be possible observe, among the main
stream, it has been inserted different set instruments. Thank to this approach different line
where controlled setting a value of reference that is then responsible of the optimization
value we will obtain from the optimization system.

Before doing this, trough the “workbook” command in Aspen HYSYS®, all over specified
values has been deleted in order to make the program evaluate them. In this way it has
been possible avoid some unwanted error due to inconsistencies.

Figure 22: Aspen HYSYS® workbook

54
FRANCESCO MORANDI SELF-OPTIMIZING CONTROL

Having complete this preliminary part some test changing feed flow and condition of
pressure as been tested to verify the feasibility of the system and if there weren‟t some
numerical problem of evaluation.

3.4.6 THE LAST MODEL

Figure 23: final HDS unit layout

This last version of the scheme plant is the outcome of a discussion with professor
Skogestad about the possibility of making more reliable and credible the project. In the
make-up line it has been inserted a compressor to simulate the real supply of H2 in a real
plant: the line before the plant is typically at about 6 [bar], while the system is working at
about 100 [bar].

With this modification, also the flash column at the end of treatment plant has been
removed to make the plant easier to work and faster than before.

Beyond this all throttling valves used to flash the mix inside the different vessels (e.g. HP
separator and scrubber) have been substituted by direct expansion inside the equipment
cited before.

The number of fixed variables have been lowered more than the previous model, allowing
the simulation program evaluate them. The reason of this procedure is the same expressed
before.

A “spreadsheet” has been introduced in order to monitor more accurately the operative
parameters and working conditions.

55
FRANCESCO MORANDI SELF-OPTIMIZING CONTROL

What is possible to observe directly on the PFD is how clearer is compared with the
previous models; but even if the equipment has been modified or changed in order to make
calculations easier and maybe without any numerical error, the model follows as best as its
possibility (that is optimally as well) a real hydrodesulfurization unit.

At the end of this preliminary part professor Skogestad has verified the real feasibility and
rightness of the system. The model was proved to be functional and able to reproduce
different situation of loading and at various operative conditions.

3.5 KINETICS

In order to evaluate the right behavior of the reactor and subsequently its performance on
sulfur compounds‟ abatement, a series of different kinetics have been tested. Here it will be
shown simple graphs to give an idea how they work and why the choice has been done on
one instead of another one.

The best way to test them has been to build some models in C++ code and make it to plot
the graphs used to compare with the example taken in literature and with our goal. We
have to note that among the different possibilities offered by AspenHYSYS® simulation
program there is the possibility to insert and adopt different kinetic scheme from
equilibrium to heterogeneous catalytic provided that the right parameters are used.

Figure 24: how to select the reaction set in Aspen HYSYS®

56
FRANCESCO MORANDI SELF-OPTIMIZING CONTROL

3.5.1 FIRST KINETICS: FORMENT et al.

In order to make it possible the right operation of the reactor, as stated before, it‟s
necessary to introduce a right kinetic model and see if it‟s capable to represent every
situation that could be verify during the normal activity of a plant. First kinetic scheme is
that proposed by Froment et al. 37.

In this paper the HDS reactions was analyzed one by one considering all the possible way
through which the reaction could happen. The main compounds treated were
benzothiophene and dibenzohiophene those are the major contribute for what concerns the
total sulfur presence in a typical diesel feedstock.

This scheme was proved only for T = 533 [K] as it was in isothermal conditions even if it
is unreal as hypothesis, but that was the only temperature available: all kinetic expressions
are patented and “that is now part of a commercial software” (G.Froment, 12/03/2013).

The rate of H2 consumption and sulfur compounds, with the production of hydrogen
sulfide too were in agreement with literature data and also in test with the corresponding
graphs of the article from which the scheme was taken there was good respect.

But the impossibility to verify the workability of a system on a range of temperature


different from the one used before, has made necessary the goal to find a new kinetic
scheme which it could be fully defined and completely functional.

3.5.2 SECOND KINETICS: D. TSAMATSOULIS and N. PAPAYANNAKOS

In this second case the possibility to choose a kinetic model with an expression rate unique
for all the compounds taken in consideration, has been verified. For this analysis a paper of
D. Tsamatsoulis et al. 38 has been used.

In the article the hydrodesulfurization of a VGO has been taken into consideration and
there have been estimated kinetic equations in order to evaluate the behavior of the system;
two kinetics have been analyzed:

37
GILBERT F. FROMENT, GUY A. DEPAUW, AND VALÉRIE VANRYSSELBERGHE, Kinetic Modeling and Reactor Simulation in
Hydrodesulfurization of Oil Fractions, Amercan Chemical Society 1994.
38
D. TSAMATSOULIS AND N. PAPAYANNAKOS, Investigation of intrinsic hydrodesulphurization of a VGO in a trickle
bed reactor with backmixing effects, Chemical Engineering Science, Vol. 53, No. 19, pp. 3449-3458, 1998.

57
FRANCESCO MORANDI SELF-OPTIMIZING CONTROL

with and

The research has brought to find some different parameters based on the type of catalyst;
and these parameters have been evaluated for the catalyst used in the process and they are
found to equal to:

Table 6: HDS kinetic parameters of D. TSAMATSOULIS AND N. PAPAYANNAKOS

HDS kinetic parameters VALUE


n 2.13
np 0.92
kHDS0 1.5x108
EHDS 142,950
KH2S 0.36x10-5

3.5.3 THIRD KINETICS: AYSAR T. JARULLAH, IQBAL M. MUJTABA, ALASTAIR S.


WOOD
Having noted that a kinetic equation able to describe the overall behavior of this unit is
more appreciable also in the results it gives us; the analysis of a “near optimal” kinetics
have brought to the paper of Aysar T. Jarullah et al 39. The HDS I described by an equation
based on Langmuir-Hinshelwood model:

Where

39
AYSAR T. JARULLAH, IQBAL M. MUJTABA, ALASTAIR S. WOOD, Kinetic parameter estimation and simulation of a
trickle-bed reactor for hydrodesulfurization of crude oil, Chemica Engineering Science, vol. 66, pp. 859-871,
2011.

58
FRANCESCO MORANDI SELF-OPTIMIZING CONTROL

While the other parameters are summarized in table below:

Table 7: HDS kinetic parameters of AYSAR T. JARULLAH, IQBAL M. MUJTABA, ALASTAIR S. WOOD

HDS kinetic parameters VALUE


n 1.147
m 0.4749
AHDS0 2,026.23
EAHDS 50,264.10

Seeing in comparison the simulation results (evaluated by Aspen HYSYS®, on the basis of
the above mentioned kinetics40) and the paper used as reference to study the process 41, with
the feedstock to treat, we can observe a good prediction of the performance of this unit:

Figure 25: results from Aspen HYSYS®

40
AYSAR T. JARULLAH, IQBAL M. MUJTABA, ALASTAIR S. WOOD, Kinetic parameter estimation and simulation of a
trickle-bed reactor for hydrodesulfurization of crude oil, Chemica Engineering Science, vol. 66, pp. 859-871,
2011.
41
GILBERT F. FROMENT, GUY A. DEPAUW, AND VALÉRIE VANRYSSELBERGHE, Kinetic Modeling and Reactor Simulation in
Hydrodesulfurization of Oil Fractions, Amercan Chemical Society 1994.

59
FRANCESCO MORANDI SELF-OPTIMIZING CONTROL

Figure 26: Gilbert F. Froment results

With a percentage of desulfurization equal to 99.04%, value that is comparable with the
literature previous one.

3.6 THE FEEDSTOCK

As seen previously, there are different kind of HDS unit, everyone optimal and
“specialized” for a certain type of feedstock. In particular for a certain raw feed to treat.
It‟s quite clear that the characteristics of the feed, in terms of sulfur and nitrogen content,
depend on first of all the geographical position; then obviously it‟s tied with the type of the
lump taken in consideration.

For this simulation the focus was pointed on a synthetic diesel feed 42. Although there are
various possibilities cause the extreme variation also in composition and the different ratio
between PAN in the same raw feed; the choice has been done in order to make some direct
considerations on this fuel that in these last years has been put under the direct observation
of the various environmental agencies in order to control the worldwide toxic emission,
among those: SOx and NOx. The composition used is reported in the table below:

42
GILBERT F. FROMENT, GUY A. DEPAUW, AND VALÉRIE VANRYSSELBERGHE, Kinetic Modeling and Reactor Simulation in
Hydrodesulfurization of Oil Fractions, Amercan Chemical Society 1994.

60
FRANCESCO MORANDI SELF-OPTIMIZING CONTROL

Figure 27: Aspen HYSYS® composition

As you can note in the table there is a compound over star-signed: the n-nonylcyclohexane
has been introduced in the simulation building the molecule through UNIFAC structure; all
data for this compound are those from the same theory and in order to have a good
approach to simulation these data have been compared with those coming from literature
matching perfectly.

For what concerns the conditions of this stream are those one:

Figure 28: Aspen HYSYS® feed conditions

61
FRANCESCO MORANDI SELF-OPTIMIZING CONTROL

Chapter 4
Steady State simulation: theory, analysis and
results

“Ready Steady Go”


Paul Oakenfold
4.1 INTRODUCTION
The previous chapter has been useful to introduce and specify the scheme-basis on which it
will be possible to apply the procedure of control that is the so-called Self Optimizing
Control based on a more general approach of the Economic Plantwide Control. This
procedure found its basis in the theory developed by Professor Sigurd Skogestad.
We must consider that a chemical plant has thousands of measurements and control loops.
If we thought to adjust and control each of these measurements maybe we couldn‟t have
reason of what a plant is and how it acts during its work. So the term plantwide control
“does not imply the tune and behavior of these loops, but rather the CONTROL
PHILOSOPHY of the overall plant with emphasis on the structural decisions” 43.
The analysis brings to a control system divided in several layers each for a different
timescale level, as you can see in Figure 1.

43
SIGURD SKOGESTAD, Plantwide Control, Recent Developments and Applications, Wiley 2012

62
FRANCESCO MORANDI SELF-OPTIMIZING CONTROL

Figure 29: hierarchical control

This theory was born in 1983 when for the first time Skogestad started to search a correct
structure for a column control. Taking his inspiration from a famous article by Foss
(1973)44 in which he stated that the determination of control system structure is the “central
issue to be resolved in control” he went on this way. In 2004 he published his work 45
where he explained the basis and the rules of the procedure; procedure that is split in two
parts: a top-down, based on economics and a bottom-up which aims to reach a simple and
robust “stabilizing” or “regulatory” control.
This philosophy aims to reach control as simplest as possible, “PhD Control” (Figure 2):

Figure 30: PhD control

44
FOSS, A. S., Critique of Chemical Process Control Theory, IEEE Transactions on Automatic Control, AC-18(6),
646-652, 1973.
45
SIGURD SKOGESTAD, Control structure design for complete chemical plants, Computers & Chemical
Engineering, Volume 28, Issues 1–2, 15 January 2004, Pages 219-234.

63
FRANCESCO MORANDI SELF-OPTIMIZING CONTROL

that is an idealized way to see the control structure in a process; because what we have in
reality is something more complex and sometimes unforeseeable, something like this (a
classical “PID control”):

Figure 31: Tennessee Eastman challenge problem (Downs, 1991)

where the choice of decide what and where to put the different controllers becomes really
difficult, especially if our intent is reach an intelligent way to control, taking into account
of the fastness of response of the diverse controllers and the possibility to avoid redundant
measurements.
The right question become how do we get from a PID to a PhD control? Is it possible to
achieve a systematic procedure to pursue this goal?
To these questions we can find an initial response if we consider the previous referring to
Alan Foss46:

The central issue to be resolved ... is the determination of control system


structure. Which variables should be measured, which inputs should be
manipulated and which links should be made between the two sets? There is
more than a suspicion that the work of a genius is needed here, for without it
the control configuration problem will likely remain in a primitive, hazily

46
FOSS, A. S., Critique of Chemical Process Control Theory, IEEE Transactions on Automatic Control, AC-18(6),
646-652, 1973.

64
FRANCESCO MORANDI SELF-OPTIMIZING CONTROL

stated and wholly unmanageable form. The gap is present indeed, but contrary
to the views of many, it is the theoretician who must close it.
Or as stated by Carl Nett47:
Minimize control system complexity subject to the achievement of accuracy
specifications in the face of uncertainty.
So when the purpose is to control a process the objective is not the tuning and behavior of
each control loop, rather the control philosophy of the overall plant with emphasis on the
structural decisions:
– Selection of controlled variables (“outputs”)
– Selection of manipulated variables (“inputs”)
– Selection of (extra) measurements
– Selection of control configuration (structure of overall controller that
interconnects the controlled, manipulated and measured variables)
– Selection of controller type (LQG, H-infinity, PID, decoupler, MPC etc.).
If we proceed in this way we can easily achieve the previous goal: to get from ``PID
control‟‟ to “PhD” control.
Summarizing all the previous idea the concept on which work in order to reach the control
of a process is that: plantwide control is control structure design for complete chemical
plant.
Now we can begin with the analysis of Skogestad theory.

4.2 THE THEORY


When we thin to the term plantwide control we have to focus on a definition like this:
control structure design applied to chemical plants. If we concentrate on this assertion we
can highlight how this definition doesn‟t imply directly neither the tuning of the control
system nor the behavior of the various loops naturally present in a plant. What is expressed
in reality is a “control philosophy” whose horizon is the overall plant; and the fulcrum of
all this dissertation are the structural decision.
In a schematic view of the procedure:
- Selection of Controlled Variables (CVs), i.e. OUTPUTS
- Selection of Manipulated Variables (MVs), i.e. INPUTS
- Selection of extra measurements
47
HADDAD, WASSIM M., BERNSTEIN, DENNIS S., NETT, CARL N., Decentralized H2/H∞ controller design: The
discrete-time case, Proceedings of the IEEE Conference on Decision and Control, 2, pp. 932-933, 1989

65
FRANCESCO MORANDI SELF-OPTIMIZING CONTROL

- Selection of control configuration (that is overall controller)


- Selection of controller type (P, PI, PID, MPC, LQC etc.)
When we speak of plant-wide control, we mean all the decisions necessary to understand
and make us possible to represent the plant in a diagram: from the simplest possible PF to
the most complete one P&I. If we consider this problem from the mathematical point of
view, we could abandon the hope to find a solution, because it‟s a very huge system made
by a large number of discrete decision variables; problem that has its objective in keeping
the CVs near to the set point. But the real question is what must we control? That is said
more precisely and in a different manner: What are the CVs?
If we consider AS WE MUST DO the overall plant, the answer is minimize cost or
maximize profit satisfying at the same time the constraints imposed by the market or
equipment.
Speaking in a mathematical manner this is a simple way to formulate the problem: we have
a constrained problem whose solution could be found in some particular cases, with the
right assumptions (this has been possible thanks to the recent development in technology
and in particular in the numerical calculus). In practice what it‟s necessary to do is obtain a
complete plant, detailed in both dynamic and steady-state model, defining:
- Operational constraints
- All measurements and manipulations
- All expected disturbances
- All expected, allowed or desirable ranges for variables
and finally design a controller to keep variables close to set points. That is a controller that
in any time collects all the information coming from the plant and consequently provides
an optimizations of the Manipulated Variables.
That could be possible I some cases (very simple one), but practically it‟s being considered
impossible. Moreover if we consider that acceptable control is achieved with simple
structures and that the same structures made possible a good operation for most real plants,
then the right conclusion is that we need to use a simple, but effective, control strategy.
How could be possible to control a real system? Sometimes as usual a decomposition of
the global problem into many simpler problems could be useful to highlight and find the
solution. During the decomposition we take into account two different phases:
1- Decentralized control: it‟s the “horizontal decomposition”, where the advantage we
obtain is based on spatial separation, i.e. the control of individual process units.

66
FRANCESCO MORANDI SELF-OPTIMIZING CONTROL

2- Hierarchical control: it‟s the “vertical decomposition”, whose subdivision is made


analyzing the timescale of the process (Figure 1); in order to have an idea of what
concerns with the timescale of a real chemical plant, you can observe that:
 SCHEDULING is in the order of the weeks;
 SITE-WIDE OPTIMIZATION takes days;
 LOCAL OPTIMIZATION, we talk about ours;
 SUPERVISORY CONTROL, just minutes;
 REGULATORY CONTROL, whose characteristic time is about seconds;
Such a cascade decomposition that operates on different timescale is at the basis of the
control of all complex systems, e.g. biological systems, airplanes, plants and so on.
Considering the previous picture, and referring to the precedent bulleted list in particular to
the first three points, we have to precise that those are part of an economic optimization of
the plant or structure we want to control. What is the real interest of this work is focused
on the last two points, or layers, i.e. the supervisory control and the regulatory control one,
with the goal to track the set points coming from the previous layers. And before this, a
very important decision is the choice of the Controlled Variables (CVs), for which the set
points are given: in the supervisory control we select the CVs useful for an economic
vision of the plant. For what concerns the selection of the TYPE of the controller, there‟s
no a unique choice, the most used one is the PID controller, able to grant rapidity,
precision and robustness to the control action.
Independently from the choice, it‟s necessary work and think about the sequent decisions:
 Decision 1: we must individuate the primary CVs (CV1) based on an economic
analysis of the plant  supervisory control. Their set points are the bridge between
the optimization layer and the control one;
 Decision 2: we must select the secondary CVs (CV2), i.e. the stabilizing variables
 regulatory control. Their set points link the two control layers;
 Decision 3: we need to locate the throughput manipulator (TPM);
 Decision 4: we have to select pairings for the stabilizing layer, i.e. pair inputs
(valves) and controlled variables, those coming from Decision 2.
The first and the second decision we can see in Figure 2, where CV 1, CV2 are matrices and
they are defined by H1 y and H2 y (both H are matrices themselves) while y are the
measurements available.

67
FRANCESCO MORANDI SELF-OPTIMIZING CONTROL

So that‟s the previous picture with the respect of the choice made for the different CVs and
how they act on the plant:

Figure 32: different control levels

4.3 THE PROCEDURE


This project has been made possible thanks to the previous work by Sigurd Skogestad. One
of the first approach to the plantwide control was made by Buckley 48 (1964), then what we
can obtain from literature is relatively scarce.
A recent review of the literature on plantwide control can be found in Skogestad and
Larsson in 2000 49 who proposed a systematic subdivision and a good review of the
previous works and approaches proposed.
On the basis of their work it‟s possible to distinguish and join the different articles into two
classes:
 the process-oriented approaches (that are engineering or simulation-based), among
those we can find the ones proposed by Buckley50, Douglas51, Downs52, Luyben et

48
PAGE S. BUCKLEY, Techniques of Process Control, Wiley, New York (1964)
49
TRULS LARSSON, SIGURD SKOGESTAD, Plantwide control - A review and a new design procedure, Modeling,
Identification and Control, 21 (4), pp 209 – 240 (July 1, 2000)
50
PAGE S. BUCKLEY, Techniques of Process Control, Wiley, New York (1964)
51
JAMES MERRIL DOUGLAS, Conceptual Design of Chemical Processes, New York: McGraw-Hill (1988)

68
FRANCESCO MORANDI SELF-OPTIMIZING CONTROL

al. (1997, 1998) and Konda et al. (2005). They are characterized by a non sufficient
systematic procedure and their goal is reached without giving importance to
economics.
 the optimization or mathematically oriented approaches (academic in a certain
way), with works of Narraway and Perkins53, Hansen et al. (1998), Kookos and
Perkins54, Chen and McAvoy (2003) and Engell (2007). With the term of academic
here, we mean that these approaches lead to optimization problems those are
insoluble for what concerns the plantwide application, i.e. their use is purely
academic without any correspondence in real life.
So the right way to move on is a hybrid approaches between the previous ones: and in this
sense we find the works of Zheng et al.(1999), Larsson and Skogestad55, Vasbinder and
Hoo (2003), Skogestad56, Ward et al. (2006).

Process-oriented approaches Mathematically oriented approaches


Buckley Downs Narraway and Perkins
Skogestad
Douglas Larsson Hansen Kookos
Zeng
Luyben Vasbinder and Hoo Chen and McAvoy
Ward
Konda Engell

The procedure that will be shown in the next pages is the one developed by Skogestad; it
was inspired by Luyben, but we‟ll able to note that the former is divided in two different
parts: one concerned with steady-state economics, while the second one is involved in the
stabilization loops.

52
J.J. DOWNS, Distillation Control in a Plantwide Control Environment, In: Practical Distillation Control, W.L
Luyben (ed.), Van Nostrand Reinhold, 413-439 (1992)
53
L. T. NARRAWAY, J. D. PERKINS, Selection of process control structures based on Economics, Comp. Chem. Eng.
18, S511, (1993a)
54
KOOKOS, I.K., PERKINS, J.D., An algorithmic method for the selection of multivariable process control
structures, Journal of Process Control, 12 (1) , pp. 85-99, 2002
55
TRULS LARSSON, SIGURD SKOGESTAD, Plantwide control - A review and a new design procedure, Modeling,
Identification and Control, 21 (4), pp 209 – 240 (July 1, 2000)
56
SIGURD SKOGESTAD, Control structure design for complete chemical plants, Computers & Chemical
Engineering, Volume 28, Issues 1–2, 15 January 2004, Pages 219-234.

69
FRANCESCO MORANDI SELF-OPTIMIZING CONTROL

In order to evaluate the differences between these two configurations we start analyzing
Luyben‟s procedure and then we‟ll compare with Skogestad‟s one.

4.3.1 LUYBEN’S PROCEDURE


• Step L1. Establish control objectives
• Step L2. Determine control degrees of freedom
• Step L3. Establish energy management system
• Step L4. Set the production rate (Decision 3)
• Step L5. Control product quality and handle safety, environmental and operational
constraints
• Step L6. Fix a flow in every recycle loop and control inventories
• Step L7. Check component balances
• Step L8. Control individual unit operations
• Step L9. Optimize economics and improve dynamic controllability

Now we have to note some things:


 “Establish control objectives” in step L1 doesn‟t lead directly to the choice of
controlled variables (i.e. what we find in Decisions 1 and 2). Thus, in Luyben‟s
procedure, Decisions 1, 2 and 4 aren‟t explicit, but are included implicitly in most
of the steps.
 Even though the procedure is systematic, we have a cascade of different steps, it is
still heuristic and ad hoc: as it is not clear how the authors arrived at the steps or
their order.
 A major weakness is that the procedure doesn‟t include economics, except as an
afterthought in step L9 where we find the expression “Optimize economics”, but
it‟s loosely implied in all the procedure, or better there‟s no a clear link between
economics and the optimization of a plant.

70
FRANCESCO MORANDI SELF-OPTIMIZING CONTROL

4.3.2 SKOGESTAD’S PROCEDURE


I Top Down (focused on steady-state optimal operation)
• Step S1: Define operational objectives (optimal operation)
– Cost function J (to be minimized)
– Operational constraints
• Step S2 (optimization):
– (a) Identify degrees of freedom (MVs).
– (b) Optimize for expected disturbances and find regions of
active constraints
• Step S3 (implementation): Select primary controlled variables c = y1
(CVs) (Decision 1).
• Step S4: Where set the production rate? Where do I collocate the
TPM? (Inventory control) (Decision 3)
II Bottom Up (focused on the control layer structure)
• Step S5: Regulatory / stabilizing control (PID layer)
– What more to control (y2 ; local CVs)? y (Decision 2)
– Pairing of inputs and outputs y (Decision 4)
• Step S6: Supervisory control (MPC layer)
• Step S7: Real-time optimization (Is it necessary?)
As we can observe from the very first point of the overall procedure we have to note that
economics is at the basis of all the process: in other words we have collocate economics
into basic control layer.

Table 8: top-down/bottom up procedure

Comments, analysis tools and model


STEP requirements
I. TOP-DOWN ANALYSIS:
1. MANIPULATED VARIABLES
May need extra equipment if analysis
Select manipulated variables MV (valves and
shows there are too few DOFs.
actuators) for control.
2. DEGREE OF FREEDOM ANALYSIS
Identify dynamic and steady-state degrees of
freedom (DOF)
3. PRIMARY CONTROLLED VARIABLES: Steady-state economic analysis:
Which (primary) variables c should we control?  Define cost and constraints
 Control active constraints  Optimization w.r.t. steady-state DOFs
 Remaining DOFs: Control variables for which for various disturbances (gives active

71
FRANCESCO MORANDI SELF-OPTIMIZING CONTROL

constant setpoints give small (economic) loss constraints)


when disturbances occur.  Evaluation of loss with constant
setpoints

4. PRODUCTION RATE:
Optimal location follows from steady-state
Where should the production rate be set?
optimization (step 3), but may move
(Very important choice as it determines the
depending on operating conditions.
structure of remaining inventory control system.)

Controllability analysis: Compute zeros,


II. BOTTOM-UP DESIGN:
poles, pole vectors, relative gain array,
(With given controlled and manipulated variables)
minimum singular values, etc.
5. REGULATORY CONTROL LAYER.
5.1 Stabilization
5.2 Local disturbance rejection 5.1 Pole vector analysis (Havre and
Purpose: “Stabilize” the plant using single-loop PID Skogestad, 1997) for selecting measured
controllers to enable manual operation (by the variables and manipulated inputs for
operators) stabilizing control.
Main structural issue: What more should we 5.2 Partially controlled plant analysis.
control? Control secondary measurements (v) so
 Select secondary controlled variables that the layer above (or the operators) can
(measurements) y2 handle the effect of disturbances on the
 Pair these with manipulated variables m, primary outputs (c).
avoiding m’s that saturate (reach constraints) Model: Linear multivariable dynamic
model. Steady state usually not important.

6. SUPERVISORY CONTROL LAYER.


6a. Decentralized:
Purpose: Keep (primary) controlled outputs c at
Preferred for noninteracting process and
optimal setpoints cs, using unused manipulated
cases where active constraints remain
variables and setpoints vs for regulatory layer as
constant.
degrees of freedom (inputs).
Pairing analysis: Pair on RGA close to
Main structural issue: Decentralized or
identity matrix at crossover frequency,
multivariable control?
provided not negative at steady state. Use
6a. Decentralized (single-loop) control
CLDG for more detailed analysis
Possibly with addition of feed-forward and ratio
6b. Multivariable:
control.
1. Use for interacting processes and for
 May use simple PI or PID controllers.
easy handling of feedforward control
 Structural issue: choose input-output pairing
2. Use MPC with constraints handling for
moving smoothly between changing
6b. Multivariable control
active constraints (avoids logic needed
Usually with explicit handling of constraints (MPC)
in decentralized scheme 5a)
 Structural issue: Size of each multivariable
Model: see 5
application

7. OPTIMIZATION LAYER
Purpose: Identify active constraints and compute
Model: Nonlinear steady-state model, plus
optimal setpoints cs for controlled variables.
costs and constraints.
Main structural issue: Do we need real-time
optimization (RTO)?

72
FRANCESCO MORANDI SELF-OPTIMIZING CONTROL

Nonlinear dynamic simulation of critical


8. VALIDATION
parts
Now we will analyze the procedure in each of the above point (steps), in order to give the
right visual of what they mean and to make possible understanding the overall project.

4.4 SKOGESTAD’S PROCEDURE


Before the analysis of each points previously stated, it‟s necessary to introduce the concept
of degrees of freedom for operation.

4.4.1 DEGREES OF FREEDOM


The idea of degrees of freedom for an operation is often not as simple as it would be
expected. First of all when we consider an operation we have to think that the equipment is
assumed fixed; and then that the degrees of freedom (our MVs or DOFs, and from this
point on u in the sequent explanation) change during the passage through the hierarchy (as
you can see in Figure 1and 4). We can have three different kind of degrees of freedom:
 DOFs for optimization (i.e. steady-state DOFs, u), MVopt=CV1s
 DOFs for supervisory control, MV1 = CV2s + unused valves
 DOFs for stabilizing control (physical), MV2 = valves (for the dynamic process)
For this application, the economics of overall the plant is determined and evaluated at
steady-state behaviour (Morari et al. 1980), and the former steady-state degrees of freedom
are the same of the economic DOFs.
A brief explanation of what we consider as valves: this term is adopted here because this
is their nature when we speak of process control. The stabilizing control layer usually
doesn‟t use all the valves: e.g. some valves may not be used dynamically because they are
optimally constant (for example a relief valve); while other valves may not be needed for
stabilizing control and they can be used by the supervisory control layer to improve the
dynamic performance of the control system itself.

4.4.1a Steady-State DOFs


This is the first class of DOFs. And while the individuation of the physical degrees of
freedom in quite absolute (in the sense that they represent the valves and in this sense it‟s
easier to collocate their position and number inside the process); it is more difficult to find
the steady-state DOFs (our u, as seen before).

73
FRANCESCO MORANDI SELF-OPTIMIZING CONTROL

Yet what is useful to find is the number of the DOFs, here nu and not really the variables
themselves, because this is the number that individuate the CV 1 that will be used in Step 3.
In this sense it‟s important to note and keep in mind that with this counting we will be able
to understand and identify the number of primary controlled variables, i.e.:
N° of primary CVs = N° of steady-state DOFs, nu
It‟s important to find a independent and simple way to find nu because it‟s useful to check
and then to reduce the time spent on optimizing the process.
Actually there are three ways to identify and evaluate the number of them:
 Equation-counting
 Valve counting
 Potential degrees of freedom

EQUATION-COUNTING
Among all the possibilities this is certainly the most difficult to apply without encountering
in an error. It‟s the most “brutal” method and concerns with the evaluation of all equations
in the model; but if we think that is really difficult have a correct model for the process, the
idea of identify all the equations seems to become unreal and almost fantastic.
In practice we have:
nu = n° of variables – n° of equations/specifications
and it‟s difficult and not used in everyday practice.

VALVE COUNTING
The first approach is identify all physical degrees of freedom (the so called dynamic),
valves (“valves” also includes adjustable compressor power for example, in other words
anything we can manipulate). But as the economics is usually depending on only steady-
state, we have to take into account valves with steady-state effect and avoid inserting
variables without or with negligible effect on economics, i.e. variables with dynamic effect.
So:
#steady-state DOFs, nu = #valves - #variables without steady-state effect
Or if we want to be more precise:
nu = Nvalves – N0ss – Nspecs
where:
Nvalves = n° of dynamic (control) DOFs (valves)

74
FRANCESCO MORANDI SELF-OPTIMIZING CONTROL

N0ss = N0y + N0,valves = n° of variables with no steady-state effect


N0,valves = n° purely dynamic control DOFs
N0y = n° controlled variables (e.g. liquid levels) with no steady-state effect
Nspecs = n° of equality specifications (e.g. given pressure)

Let‟s now see some examples in order to understand better the meaning of the above
mentioned method usable to identify the different steady-state degrees of freedom. In
particular we analyze a typical simple equipment (heat exchanger) and more complex one,
very important in whatever plant we are concerning with studying, that is the distillation
column.

Examples:

Heat exchanger = 3 valves (bypass valves on both hot and cold side and a valve on
cooling water) – 2 bypass valves = 1 nu, i.e. the amount of heat transferred.

Figure 33: a heat exchanger

Distillation columns = 6 valves (feed, F + bottom product, B + distillate product, D +


cooling, C + reflux, R + heat input Q  it determines the boil-up V) – 2 specifications on
product (B + D) = 4 nu.
It‟s usual to exclude liquid levels because they have no steady-state.

75
FRANCESCO MORANDI SELF-OPTIMIZING CONTROL

Figure 34: DOFS of a distillation column

POTENTIAL STEADY-STATE DOFs


Another way to identify the different degrees of freedom of an equipment is to adopt the
sequent table (Table 1). When you use the word “potential” you have to take into account
that there isn‟t always a valve to adjust the DOF: in the previous example when we speak
about the heat exchanger we have said that it has 1 DOF, but if there‟s not a bypass then it
cannot be used.

Table 9: Potential steady-state degrees of freedom

Equipment nu

Each external feed stream 1 Feed rate


Splitter n-1 Split fractions
Mixer 0 [-]
Compressor, turbine, pump 1 Work/speed
Adiabatic flash tank 0* [-]
Heat exchanger 1 Bypass/flow
Liquid phase reactor 1 Holdup reactant
Gas phase reactor 0* [-]
Column (no heat exchanger) 0*+n° side streams [-]
Note: pressure*: we need to add 1 DOF at each extra place you set the pressure (that means
to use an extra valve, compressor or pump).

76
FRANCESCO MORANDI SELF-OPTIMIZING CONTROL

Considering again the previous examples we obtain:

Examples:

Heat exchanger = 1 nu, as we can evidence from Table 1.

Distillation column = the column itself has 0 DOFs (steady-state), but I‟s necessary to
take into account other potential DOFs: feed rate count for 1, splitter (reflux) is 1 again and
we have 2 DOFs coming from heat exchangers (condenser and reboiler). The total is
another time 4 nu as previously obtained from valves counting.

We can start now the analysis of Sigurd Skogestad approach, going to analyze all the steps
more in detail.

4.4.2 TOP-DOWN
This part as said before is particularly concerned with economics and economics is the
basis on which building the control structure. For this part of the analysis a steady-state of
the model is sufficient for a complete study and selection of the CVs.
The model built previously it would be the canvas for our project.

4.4.2a Step 1: Define Operational Objectives (Cost J and constraints)


If our intent is approaching a plantwide control we need first of all to quantify and evaluate
the operational objectives as scalar cost function J ([money/time] [€/h]) which has to be
minimized or, that is the same thing, evaluate the profit function, which has to be
maximized (P= -J).
For this point all we have to consider is:
J = cost feed + cost utilities (i.e. energy) – value products [money/time]
But what are we going to use our degrees of freedom u (MVs), previously determined, for?
The answer is just here. The scalar function we define is a function defined in this way:
J = J(u, x, d)

77
FRANCESCO MORANDI SELF-OPTIMIZING CONTROL

where:
u: degrees of freedom (usually steady-state)
x: states (internal variables)
d: disturbances

We have to pay attention to some critical points:


 Fixed costs and capital costs aren‟t included: indeed they aren‟t affected by plant
operation on timescale we consider (i.e. usually 1 [h]).
 We want to minimize J, subject to satisfying the operational constraints (safety,
environmental constraints and also some physical measurements whose negative
value would be a nonsense, e.g. flows and concentration, and so on).

4.4.2b Step 2: Determine the steady-state Optimal Operation


In order to achieve a reliable and efficient control system we have to find the optimal way
of operating process, and this has to be done before design the control system itself.
Consider that we can find that a valve should be closed (e.g. a bypass valve); then this
device couldn‟t be used for stabilizing control unless we accept the loss of “back-off”
coming from the optimal operating conditions. And to determine this plant condition we
have to work on a steady-state model, then identify DOFs (as explained previously) and
optimize it for expected disturbances.
Summarizing:
1- Identify steady-state DOFs u: we must identify them as seen before this
explanation of the first two steps. In this case actually we are concerned into the
number of u and what‟s the nature of this variables. Indeed they make an
independent set.
2- Identify important disturbances d and their expected range: they are usually
concerned into the feed rate (throughput) and feed composition, for what regards
internal disturbances; and they are related to temperature and pressure when we
consider external variables. We have to add also disturbances from the possible
change in specifications and active constraints (e.g. purity specification) and in
parameters (equilibrium constants or rate constants for example); moreover we
have to include the expected range in prices of products, feed and energy.

78
FRANCESCO MORANDI SELF-OPTIMIZING CONTROL

3- Optimize the operation for the expected disturbances: in this passage we specify
the disturbances d and we vary the degrees of freedom in the optimal way as
possible uopt (d) in order to make it possible the minimization of the cost function J,
with the respect of the constraints. Our goal is the achievement of the individuation
of the constraint regions and the optimal set-points in regions themselves.
With the scalar function we define:
J = J(u, x, d)
we optimize operation with respect to u for given d (usually steady-state):
minu J(u, x, d)
subject to:
 Model equations: f(u, x, d) = 0
 Operational constraints: g(u, x, d) ≤ 0

Example:

Figure 35: a distillation column

We proceed now analyzing a distillation column at steady-state with given p (i.e. pressure)
and F (that is the feed fed to the column): we have nu = 2, e.g. L and V (u).

What must be minimized is the cost:


J=-P
where P = pD·D + pB·B – pF·F – pV·V
with pD, pB, pF, pV = prices of Distillate, Bottom, Feed and Energy (p V)

79
FRANCESCO MORANDI SELF-OPTIMIZING CONTROL

and the following constraints:


Purity D: xD, impurity ≤ max
Purity B: xB, impurity ≤ max
Flow constraints: min ≤ D, B, L etc. ≤ max
Column capacity (flooding): V ≤ Vmax
Pressure: 1) p given (d)
2) p free (u): pmin ≤ p ≤ pmax
Feed: 1) F given (d)
2) F free (u): F ≤ Fmax
Our goal is to minimize J with respect to steady-state DOFs (u)

That is the mathematical expression of the steady-state optimization problem.


Remembering what has been explained previously, we have u those are our degrees of
freedom, x are the internal variables (states) and d are the disturbances cited before.
When we put the equality f = 0 it‟s our intent to express the mathematical model equations
and some equality constraints (e.g. the feed flow or a pressure, whatever has the necessity
to be set with a certain value); on the other hand when we state g ≤ 0, we want to represent
the operational constraints (for example a product composition or maybe a liquid/vapour
ratio).
When we speak about active constraints 57, we refer to the expression used in mathematical
optimization, where we have a problem defined using an objective function that has to
minimize or maximize, and in parallel we have a set of constraints, usually some equalities
and inequalities:
g1 (x) ≥ 0
g2 (x) ≥ 0
g3 (x) ≥ 0

gn (x) ≥ 0
that define the feasible region, that is, the set of all x to search for the optimal solution. So
given a point x in the feasible region, one of the previous constraints,
gi (x) ≥ 0
is called active if we have the following equality:
57
NOCEDAL, JORGE; WRIGHT, STEPHEN J., Numerical Optimization, Berlin, New York: Springer-Verlag, 2nd ed.
2006

80
FRANCESCO MORANDI SELF-OPTIMIZING CONTROL

gi (x) = 0
One of the most important results we want to obtain from the optimization is to find the
region of active constraints; this is the most time-consuming step of the entire procedure
both obtaining the model from literature or building up from a previous P&I diagram and
the optimization itself with the mathematical numerical resolution.
In case of necessity (when the model may not be available or there no material time to
operate the optimization) it could be possible perform a simplified version of the Step 1
and Step 3 using a process insight to operate the individuation of active constraints
themselves and the possible self-optimizing variable (CV1) for the remaining DOFs.
It‟s important to understand that there is not a unique control structure defined as optimal
and that is due to the fact that the set of active constraints will vary and depend on
disturbances and on prices of market. Generally the control system should be able to
supply reliability in case of future variations. What should be done in order to increase the
performance of the control system, it would be the offline analysis and optimization to
identify expected regions of active constraints.
As we stated before the optimal active constraints change depending on disturbances (feed,
composition, temperature and so on) and on market conditions (i.e. price). For this latter
cause there are two mode of operation, function of market condition: given
throughput/feed-rate (Buyer‟s Market) or maximum throughput/product (Seller‟s Market).

The former (given throughput/feed-rate) is usually the nominal mode for which the control
system is set up to handle. It‟s the typical situation tending to maximize the efficiency,
where we have some tradeoff among the utilities (consumption of energy) and the recovery
of the product. It a situation providing an unconstrained optimum.
What we have is the sequent situation:
J = cost feed – value product + cost energy
where the first two terms of the previous equality (cost feed – value product) are
considered often constant; and usually, as said before, optimal operation is usually
unconstrained. The control is going to operate at optimal trade-off, but this is not obvious
what to control to achieve this.

81
FRANCESCO MORANDI SELF-OPTIMIZING CONTROL

Figure 36: objective function J versus cost energy

The latter (maximum throughput/product) could be applied when the product price is
higher if compared to price of raw material (feed) and energy (utilities) and tends to
increase the throughput as much as possible. In contrast with the previous situation here
the efficiency falls and it usually happen that constraints reach the bottleneck and in this
state any further increase results in an infeasible solution.
The situation is similar to the previous one:
J = cost feed – value product + cost energy
but in this case the assumptions regard the feederate that is now a degree of freedom and
that the product is much more valuable than the feed: the optimal operation is then as
stated before to maximize product rate (and always as seen before, remaining in the
feasible region). Here what has to be controlled is the “bottleneck” and it‟s obvious for
sure.

Figure 37: objective function J versus DOFs

4.4.2c Step 3: Select Economic (Primary) Controlled Variables, CV 1


In this point we will deal with the implementation of optimal points found in the previous
two steps. Our goal is tend to organize them in a robust and simple manner in order to
make it possible to create a praxis useful to anyone is going to apply this procedure.
What we have obtained from the previous analysis is the economic degrees of freedom (u),
and now what we want to reach is identify the economic controlled variables (CV 1) those

82
FRANCESCO MORANDI SELF-OPTIMIZING CONTROL

are responsible for the first controlling layer of the hierarchical structure we have see at the
beginning of this chapter. Remember that the number of first controlled variables is equal
to steady-state degrees of freedom one.
The question we have to answer is: What should we control? And how should we adjust
the degrees of freedom (u)?
So the optimal operation for given d *:
minu J(u, x, d)
subject to:
Model equations: f(u, x, d) = 0
Operational constraints: g(u, x, d) ≤ 0
The result of this study is uopt, whose values however usually cannot be kept constant
because disturbances d change, and in this way the optimal results change too.
By the way for the economical optimal operation, we have two rules:
 CV1 – rule 1, that is control active constraints
 CV1 – rule 2, i.e. for the remaining unconstrained DOFs, we must control self-
optimizing variables, where the ideal self-optimizing variable c is the gradient is
c =  J/ u = Ju
As seen previously for each steady-state degree of freedom, u, there is one CV1. There is a
unique match between the variables and the DOFs. For what regards the so-called “self-
optimizing” variables we mean variables “for which close-to-optimal operation with
constant setpoints can be achieved, even when there are disturbances” 58.
Thinking about active constraints, instead, they are self-optimizing variables in the wider
sense, because operation is optimized keeping their values constant (usually they are
referred as unconstrained self-optimizing variable, highlighting the difficulty to find them).

CV1 – rule 1
As anticipated in the previous section of this paragraph we have to control active
constraints for first, because they are the obvious self-optimizing variables to be kept
constant. Active constraints are evaluated by analysis in Step 2. It‟s important to identify if
an active constraints is an input constraints or an output one. What‟s the difference? The
former are trivial to implement cause we could only set the minimum or the maximum
value for this variable and because this reason no control is then needed.
58
TRULS LARSSON, SIGURD SKOGESTAD, Plantwide control - A review and a new design procedure, Modeling,
Identification and Control, 21 (4), pp 209 – 240 (July 1, 2000)

83
FRANCESCO MORANDI SELF-OPTIMIZING CONTROL

For example, if we are dealing with an optimal operation of a runner (a sprinter on 100
[m]), assuming that the optimal operation itself is time (J = T), we could achieved the
solution for this problem with maximum speed for the athlete (no thinking is then required).
In this case the active constrained “controlled” is the speed of the athlete or better the
power he‟s able to use to run that distance.
Instead, if we are dealing with an output constraints (the latter of the previous distinction)
we need a controller, where it‟s often sufficient a simple single-loop to achieve the desired
result. Take into consideration the previous example: if now we have to face a marathon
(42.195 [km]), it‟s impossible to require an athlete to run at his maximum speed for such a
long race. Even if time is our optimal operation that has to be minimized we can achieve
the result in a better way. The question here is which self-optimizing variable c do we need
to control at constant setpoint?
• c1 = distance to leader of race
• c2 = speed
• c3 = heart rate
• c4 = level of lactate in muscles

Figure 38: self-optimizing control of a marathon runner

84
FRANCESCO MORANDI SELF-OPTIMIZING CONTROL

Thinking in this way the problem is now reduced into a simple and robust implementation,
where disturbances are indirectly handled by keeping a constant heart rate (our variable
controlled) which may have infrequent adjustment of setpoint (heart rate itself).

BACK –OFF and LOSS


Let‟s now consider another example in reference to the previous one about marathon. In
this case think about a car and consider that the maximum speed limit is like an active
constraint (suppose that the maximum speed limit is 80 [km/h]) and because of this nature
we should be selected as controlled variable (CV1). In order to reach this control we may
use the cruise controller, which is able to adjust the engine power, keeping the car velocity
near to the setpoint. But either in the case there is a steady-state measurement error (ny = 5
[km/h]) or there is a dynamic control error (3 [km/h]), we have to recede from the speed
limit to avoid exceeding the speed limit itself (and for this reason put the setpoint at 72
[km/h]). So in general we want to minimize the back-off because larger is the back-off
itself greater is the loss (i.e. larger J = T).
The back-off is the limit, the safety margin from the active constraints and it can be defined
as the difference between the active constraints value and the chosen setpoint, i.e.:
Back-off = |Constraint - Setpoint|
In our example we have a back-off equal to 8 [km/h] (|80 [km/h] – 72 [km/h]|) if we must
not exceed the speed limit (i.e. a hard constraint); while it can be reduced to 5 [km/h] if we
consider that the dynamic error will average out if control has an integral action (80 [km/h]
– {72 [km/h] + 3 [km/h]}), i.e. soft constraint.
Why do we have to control the active constraints? They must be selected as CVs because
the optimum doesn‟t reach a flat point with the respect of this variables.
Moreover, as mentioned before, there is an economic penalty if we recede from an active
constraint: if a constrained optimization method is used for the optimization, then we can
quantify the loss using the Lagrange multiplier , associated to the constraint itself:
Loss =  x back-off
Because of the direct dependence (linearity) from the Lagrange multiplier, we can note that
bigger is the multiplier, greater is the economic loss unless the back-off is very small.
What‟s the right value for a back-off?

85
FRANCESCO MORANDI SELF-OPTIMIZING CONTROL

In case of input constraints, it‟s obvious we do not have any back-off unless the goal is to
use the input for stabilization in the lower regulatory layer (the stabilizing one); in that case
then we need a range in order to make possible the control.
If, instead, we deal with output constraints, we have to face two different cases:
 soft output constraints (only average value matters):
back-off = measurement error (bias, ny)
 hard output constraints (has to be satisfied at all time):
back-off = measurement error (bias, ny) + control error (dynamic)

Figure 39: active output constraints

It‟s quite clear that a method to reduce the back-off is to have the possibility using accurate
measurements of output constraints; then for hard output constraints we need:
 tight control with a small dynamic control error
 “squeeze and shift rule”:by squeezing the output variation (with control), we can
shift the setpoint closer to its limit value.

86
FRANCESCO MORANDI SELF-OPTIMIZING CONTROL

Figure 40: “squeeze and shift” rule, Richalet©

In summary and other simple thumb rules:


1. Always control active constraints
2. Purity constraint on expensive product are always active (we don‟t need
overpurification), i.e.:
(a) "Avoid product give away" (e.g. selling water as expensive product)
(b) Save energy (costs energy to overpurify)
3. In case of an unconstrained optimum, we NEVER try to control a variable that
reaches maximum or minimum at the optimum (never try to control directly the
cost J).
4. Exception to CV1 – rule 1: if the Lagrange multiplier is small for some active
constraint, then it may be better to select some other unconstrained CV s, because of
simplicity, but one then needs to back off from optimal point to guarantee
feasibility.

CV1 – rule 2
This second rule is concerning the selection of the unconstrained self-optimizing variables.
In order to accomplish this goal, we have to follow the next steps:

87
FRANCESCO MORANDI SELF-OPTIMIZING CONTROL

1. Identify candidate measurements y: first of all we must identify all the candidate
measurements y, with their expected static measurements error (bias, ny). In y we
take into account both the inputs and the measurements used to control the active
constraints.
2. Select primary (economic) controlled variables CV1 = H · y (Decision 1): in the
most optimistic vision we would like to control all single measurements for
simplicity and (in this case) H is a selection matrix. More generally we may control
measurement combination and in this case H is the “full” matrix. This step must be
repeated for each constraint region. In order to reduce the necessity to switch
between regions, it could be possible to use the same CV1s, although it‟d be
responsible of infeasibility.
As we can understand it‟s easy to identify and control the active constraints; the most
difficult thing is to work on and with the other degrees of freedom, and if our question is if
it makes a difference what we control with the remaining DOFs, we have to expect that the
answer is clearly “yes”. In order to understand better we can take as example the previous
one about a marathon runner: that is a clear problem that it couldn‟t be solved as the
sprinter one. We have to use a self-optimizing approach.

It‟s now useful introducing some qualitative approaches to identify a good c = H · y,


most of all if our intention is to control a single measurement (i.e. c = selected y):
1- The optimal value of our controlled variable c is insensitive to disturbances, that is
the derivative dcopt/dd should be small, but it doesn‟t mean that the sensitivity of c
to disturbances is small: dc/dd must be not too large, because in this case the
control would be difficult. On the other hand the same ratio doesn‟t have to be too
small (that is dc/dd tends to 0), because in this case it would be impossible to
control acting on the variable c and the disturbance cannot be deleted. In summa we
weant dcopt/dd to small.
2- The variable c has to be easy to measure and control in an accurate manner.
3- The value of c is insensitive to disturbance (see point 1), but is sensitive to
manipulated variable u: the gain G = dc/du from u to c is large (i.e. a large error in
controlled variable c results in a small change of u). we could say equivalently that
the optimum with respect to variable c should be “flat”.

88
FRANCESCO MORANDI SELF-OPTIMIZING CONTROL

4- In case of two or more controlled variables c, they don‟t have to be loosely


correlated in order to avoid impracticability of control due to dependencies among
them.
Note that all four point above mentioned should be satisfied.
How can we find the matrix H? In practical there are two main approaches for finding it,
i.e. to identify self-optimizing CV1s associated with the unconstrained DOFs: the “brute
force” approach and the “local” approaches.
1- “Brute force” approach: given a set of controlled variables CV1 = c = H · y, we
compute the cost J (c, d) when we keep c constant (c = cs + H· ny) for various
disturbances d and measurements error ny. With this procedure we want to cover
the expected future operation. Generally expected extreme values in the parameter
space (d and ny) are used to compute the cost for alternative choice in the
controlled variable (matrix H). this method although is simple to understand and
apply and it works for non-linear plants and for changes in active constraints too
(only one nominal optimization is required to find the setpoints); it‟s very time
consuming and it cannot guarantee that every case is covered (there are infinite
choices for matrix H and we are not sure that the best cs are found).
2- “Local” approaches: the basis is on quadratic approximation of the cost 59, and the
main ones are pointed out here:
a. Maximum gain rule: this is a quantitative version of the previous
requirements 1 and 3, i.e. that “sensitive” variables should be controlled
with a large gain, |G| from u to c = H· y. With this rule we can have a clear
insight of our variables. In words we want to maximize (Gs), where:

In words, select controlled variables c for which the gain G (= “controllable


range”) is large compared to its span (= sum of optimal variation and
control error).

59
ALSTAD, V., SKOGESTAD, S., HORI, E.S., Optimal measurement combinations as controlled variables, Journal
of Process Control, 19 (1), pp. 138-148 (2009)

89
FRANCESCO MORANDI SELF-OPTIMIZING CONTROL

b. Nullspace method: we know that the optimal measurement sensitivity,


defined as F = dyopt/dd, is obtained when we select matrix H in order to
have HF = 0. This method reach its best results and optimal measurement
combinations in case of no noise, that is ny = 0. Each column of F express
the optimal change in y when the unconstrained variable u is adjusted in
order to have a system remaining optimal with respect to disturbances d.
Active constraints are assumed to be constant. With a model of the process
is straightforward to obtain F numerically. Assuming that we have at least
as many independent measurements y as the sum of the number of
independent inputs u and disturbances d, the optimal c = H · y is selected to
have HF = 0, that doesn‟t require that H = 0, because H is a non-square
matrix, but rather that H is in the Nullspace of FT.
The assumption is that we want optimal value of c to be independent of
disturbances d, that is:

so, we find optimal solution as a function of d: uopt(d), yopt(d) firstly, then


we linearize this relationship and in this sense we obtain
yopt = F ·d
where, as stated before F is the optimal sensitivity matrix. In order to
achieve this we want then:

That is to have . It is always possible if the number of


measurements is greater than the sum of disturbances and the number of
DOFs, i.e. #y ≥ #u + #d60.
Moreover if we consider the derivative of the cost function with respect to
the degrees of freedom, Ju, what we obtain is that this is equal to 0; indeed if
we consider the following simplification of a transfer function of the
system:

Figure 41: simple scheme for transfer function

60
V. ALSTAD AND S. SKOGESTAD, Null Space Method for Selecting Optimal Measurement Combinations as
Controlled Variables, Ind.Eng.Chem.Res, 46 (3), 846-853 (2007)

90
FRANCESCO MORANDI SELF-OPTIMIZING CONTROL

we have:

d
d
And because we state that H is in the Nullspace of FT, we have HF = 0, and
then Ju = 0 for any disturbances, d.

c. Exact local method: called also loss method, it extends the Nullspace
method to the case with noise and to any number of measurements61, 62.

Figure 42: control scheme for the exact local method

In this case for any disturbance d, we have to choose u in such a way that

where cs is constant, equal to 0 nominally. Our intent with this approach is


to find the optimal H in order to minimize the magnitude of the loss, L, for
the expected d and ny, i.e.:
,
Where:

61
ALSTAD, V., SKOGESTAD, S., HORI, E.S., Optimal measurement combinations as controlled variables, Journal of
Process Control, 19 (1), pp. 138-148 (2009)
62
KARIWALA, V., Optimal measurement combination for local self-optimizing control, Industrial and
Engineering Chemistry Research, 46 (11), pp. 3629-3634 (2007)

91
FRANCESCO MORANDI SELF-OPTIMIZING CONTROL

,
,

In order to evaluate the magnitude there are two solutions:


1- Worst case loss, i.e. Lwc
2- Average loss, i.e. Lavg
Considering only the second case as example, we have:
,

with:
Y = [FWd Wn]
F = GyJuu-1Jud – Gdy
So the optimal H is such a solution that minimize the following expression:

The general analytical solution is the following (indeed “full” H):


H = GyT (YYT)-1,
While in case of no disturbances (i.e. Wd = 0) and with the same noise for
all measurements (Wny = 1), we obtain that H = GyT; and if we have no
noise we achieve again the solution of Nullspace method, i.e. HF = 0. This
is always true only if we can have a sufficient number of measurements.

NOTE: as explained before, the new self-optimizing variables must be identified (offline)
for each region, and when you find a new region (online) it‟s necessary to switch
controlled variables. So it‟s easy to identify when to switch when a constraints is
encountered. It‟s less clear when to switch out a constraint, but monitoring the value of the
unconstrained CVs from the neighboring regions and then switch out of the constraint
region when the unconstrained CV reaches its setpoint.

92
FRANCESCO MORANDI SELF-OPTIMIZING CONTROL

4.4.2d Step 4: Select the Location of TPM (Decision 3)


As basis idea a process plant aim to transform some raw materials into products whose
value is higher than the feedstock, and to do this it involves the passage of mass through
the plant. This amount is expressed by the feed rate or product rate and it‟s determined by
specifying one DOF which we are usually to refer as the throughput manipulator (TPM).
Where does the TPM have to be collocated? This is an important decision and it‟s the link
between the TOP DOWN and the BOTTTOM UP of the procedure. In a certain sense it‟s
the “gas pedal” of the process deciding how much matter has to be transformed per hour,
it‟s usually a flow and it‟s set by the manual control.
A TPM is a degree of freedom that affects the network flow and which is not
directly or indirectly determined by the control of the individual units,
including their inventory control.63
The TPM is not unique: some plant with parallel units may have more than one TPM. It‟s
usually placed at the feed of the plant and this is due to the fact that most of the control
structure decision are made during the design stage, i.e. before the plant is build, and when
the feed rate is considered constant and fixed. But focusing on the future operation of the
plant it‟s probable that we want to maximize the feed in order to achieve better
performance and it could be a reasonable choice to move the TPM.
The question now is where should the TPM collocated? Generally the TPM may be
collocated anywhere in the plant, even if operators prefer to have at the feed and this is
usually the default choice. But what doesn‟t affect from the steady-state point of view, it‟s
instead very important if we consider the dynamic behavior of the plant. We have to
consider two principles to set it:
1- Economics, as usual, covers a great importance in all decisions: in fact there could
be a possible back-off if active constraints are not tightly controlled. The economic
loss, moreover, could be large if the bottleneck unit is not tightly controlled. If this
would be the case, then the TPM should be located close to the bottleneck in order
to reduce the back-off from the active constraint (largest effect on the production
rate).
2- Structure of regulatory control system: because of the radiation rule 64 , the
location of the TPM has a great importance on the structure of the regulatory

63
ibidem
64
PRICE, R.M., GEORGAKIS, C., Plantwide regulatory control design procedure using a tiered framework,
Industrial and Engineering Chemistry Research, 32 (11), pp. 2693-2705 (1993)

93
FRANCESCO MORANDI SELF-OPTIMIZING CONTROL

control structure for the entire plant. Inside this assumption there is the will of
“local” consistency of the inventory control system 65. This rule may not be adhered
to by allowing for “long” inventory loops: but this is not common for operational
reasons (i.e. emptying or overfilling tanks, startup, tuning and increasing
complexity).
Some TPM collocation‟s examples, we can find in the following pictures (Figure 13, 14,
15):

Figure 43: TPM at inlet (feed), inventory control in direction of flow

Figure 44: TPM at outlet, inventory control in direction opposite the flow

Figure 45: general case with TPM inside the plant, Radiation Inventory Control

If we consider only a part of the plant, there could be the possibility that this part doesn‟t
have a TPM. There will be instead a given flow that acts as a disturbance which the control
system has to set up and handle. In other words it is as having the TPM at a fixed location.
And this is the preferable solution: first of all in fact it makes it simpler for operators (they
are responsible for adjusting the TPM); then it avoids switching the inventory structure,
which should be “radiating” around the TPM. Anyway because there is not a specific
location (or better, just common preferable location), the tempting to consider its
collocation as degree of freedom, and the moving of it in order to improve performance,

65
ALSTAD, V., SKOGESTAD, S., HORI, E.S., Optimal measurement combinations as controlled variables, Journal of
Process Control, 19 (1), pp. 138-148 (2009)

94
FRANCESCO MORANDI SELF-OPTIMIZING CONTROL

reduce back-off as disturbances cause the optimal constraints to change. Skogestad


proposed this rule66:
To get tight control of the new active constraint and achieve simple switching,
locate the TPM “close” to the next active constraint (such that the TPM can be
used to achieve tight control of the constraint when it becomes active).
The basis to this rule can be found in economics and it aims to simplify the required
switching when the next capacity constraint becomes active. We have to note, however,
that when we move the TPM, we may have to switch regulatory loops, and this is not a
desirable thing.
Come at this point we pass through the hierarchical structure and we proceed to the bottom
up control procedure. We are now going to deal with the real control layer: regulatory and
supervisory layers. They are responsible of the action on the plant devices set up, and they
are “children” of all the previous analysis done till this point.

4.4.3 BOTTOM-UP
4.4.3a Step 5: Select the Structure of Regulatory (Stabilizing) Control Layer
As it is clear from the name itself, the main purpose of the regulatory layer is to “stabilize”
the plant, preferring a simple control structure with single-loop PID controller (reliable of a
more robust control). In this case stabilize means that the process doesn‟t drift too far away
from acceptable operation even if there are some disturbances (indeed it‟s this one the goal
of a regulatory layer).
Among all the control layer the regulatory is the fastest one and it‟s used to control
variables which need a fast and a tight control (i.e. the economically important active
constraints). Note that this layer uses setpoint coming from the Step 3, but it should follow
the ones given by the supervisory layer.
In this step there are two faced two main decision:
1- Select controlled variables (CV2), i.e. Decision 2
2- Select inputs (that is valves) and “pairings” in order to control the above mentioned
variables (CV2), i.e. Decision 4
It‟s worth of noting that while the selection of controlled variable can be done on steady-
state considerations, dynamics is the primary concern when we select inputs and pairings.

66
SIGURD SKOGESTAD, Control structure design for complete chemical plants, Computers & Chemical
Engineering, Volume 28, Issues 1–2, 15 January 2004, Pages 219-234.

95
FRANCESCO MORANDI SELF-OPTIMIZING CONTROL

We have not to add any degrees of freedom because the setpoints CV 2s are left as
manipulated variables (MVs) for the supervisory layer.
If we consider to allow for cascade loops, then the stabilization layer may be designed
independent of the supervisory control layer (i.e. the economic one). But when we use
cascade loops, we add complexity and when closing a stabilization loop, we use up some
of the time window as given by the closed-loop time response (bandwidth) of the
stabilization loop. This is why it‟d be better to simplify and reduce the need of cascade
loops.

SELECT STABILIZING CV2 (Decision 2)


Firstly we have to stabilize the process by controlling drifting variables such level and
pressure (i.e. inventories), reactor temperature and temperature profile in the distillation
column. More precisely we want to control:
1. Levels (inventory liquid)
2. Pressures (inventory gas/vapor) (note: some pressures may be left floating)
3. Inventories of components that may accumulate/deplete inside plant
• E.g., amount of amine/water (deplete) in recycle loop in CO 2 capture
plant
• E.g., amount of butanol (accumulates) in methanol-water distillation
column
• E.g., amount of inert N2 (accumulates) in ammonia reactor recycle
4. Reactor temperature
5. Distillation column profile (one temperature inside column)
• Stripper/absorber profile does not generally need to be stabilized

This simplifies the supervisory control layer and its tasks because it provides for local/fast
disturbances rejection and reducing the non-linearity in the model. Then we should include
active constraints (CV1) those need a tight control (they are usually hard output
constraints) in CV2 for the regulatory layer. In this way we reduce the required back-off.
Because of all these reasons it is usually not necessary for tight control of unconstrained
CV1, indeed the optimum is quite “flat”.
We have to take into consideration two main objectives of the regulatory layer, in order to
select systematically select the stabilizing CV2 = H2 · y:

96
FRANCESCO MORANDI SELF-OPTIMIZING CONTROL

 Local disturbance rejection (indirect control of primary variables CV1). If we


control variables CV2, the effect of any disturbance on the primary CV1 should be
small. This is to get fast control of the CV1, which are important to reduce the
control error (and then the back-off) for some variables such as active output
constraints.
 Stabilization (minimize state drift). Because more generally the objective is to
minimize the effect of the disturbances on the internal variables (states, x). Why?
The main reason is to keep the process in the linear region and in this way (we want
it close to the nominal steady-state) to avoid the process itself drift into a region
where it‟s impossible to control it. The advantage of considering some
measurements of all the states x, is that the regulatory control layer isn‟t tied to a
particular control objectives (CV1), that could be change with time on the basis of
disturbances and prices.
When we consider disturbance rejection and stabilization in this layer of control (i.e.
regulatory), it‟s the behaviour at the closed-loop time constant of the above supervisory
control layer which is of main interest; moreover the supervisory layer is usually relatively
slow and for this reason is sufficient consider the steady-state behaviour when we select
CV2.

SELECT INPUTS (i.e. valves) FOR CONTROLLING CV2 (Decision 4)


Now we have to find inputs (valves) to use to control CV2. Usually and in practice single-
loop, decentralized, controllers are used in the regulatory layer and in this way the
objective is to identify pairings. The main rule is to “pair close” so that the dynamic
controllability is good and at the same time affecting in small measure the delay (i.e. small
interactions between the loops). Then we have to note that:
 Local consistency for the inventory control67; this means that the inventory control
system is radiating around the given flow (remember the radiation rule 68 mentioned
before)
 Tight control of important active constraints (we want to avoid the back-off)

67
E.M.B. ASKE AND S. SKOGESTAD, “Consistent inventory control”, Ind.Eng.Chem.Res, 48 (44), 10892-10902
(2009)
68
PRICE, R.M., GEORGAKIS, C., Plantwide regulatory control design procedure using a tiered framework,
Industrial and Engineering Chemistry Research, 32 (11), pp. 2693-2705 (1993)

97
FRANCESCO MORANDI SELF-OPTIMIZING CONTROL

 Avoid selecting variables that could be optimally saturate (steady-state) as MVs in


this layer, because this would require either reassignment of regulatory loop or
back-off for the same manipulated variables
 Derivation from the previous note is that we need to avoid the reassignments in the
regulatory layer
In a more schematic visual of the problem we may divide this second step (selection of
inputs  valves), in to two different passages, those are:
 Structure of inventory control layer (we find some analogies with Step 4)
 Structure of remaining regulatory control system
Because of its nature, this layer of control should be independent from economic control
objectives. This means that CV1 and CV2 are independent each other. But to simplify and
reduce the need for cascade loops, it‟s an advantage if we have CV 1 = CV2, or at least
some of their variables.

4.4.3b Step 6: Select Structure of Supervisory Control Layer


This control layer, often called “advanced” control layer, has three main tasks:
1- It must control the primary (economic) CV1 using the setpoints to the regulatory
layer plus any remained unused valves as MVs.
Considering that at this timescale interactions are significant, then MPC (i.e. a
kind of multivariable control) should be taken into consideration. To improve
dynamic response this layer uses level setpoints or other additional dynamic
degrees of freedom. At steady-state condition these variables should stay to
their ideal value. It could use also feedforward control (i.e. the use of
disturbances).
For the estimation of the CV1 not measured, there could be used “soft sensors”
for their estimation. These sensors are static, although dynamic estimators. But
their use is quite small cause the slowness in response time of this control layer.
2- Supervise the performance of the regulatory layer: this layer should take action to
avoid saturation of MVs used for regulatory control. In fact we must note that if a
MV in the regulatory layer saturates, then the control of the corresponding CV2 is
resulting in a large drift away from the desired operating point.

98
FRANCESCO MORANDI SELF-OPTIMIZING CONTROL

3- Switch controlled variables and control strategies when we have disturbances on


prices that could cause the process to change the region of active constraints.

The controller those are possible to use in this layer are essentially two:
 Advanced single loop control, i.e. PID control, with some additional “fixes”, for
example feedforward ratio, decouplers, logic, selectors and split range control.
With this kind of control it‟s extremely important to select the pairings. Moreover
we must to note that finding the right pairings is more difficult because slower the
timescale stronger are the interactions.
 Multivariable control (MPC). Although the use of MPC can reduce the use of
switching and logic, this cannot be completely avoided. Generally when we switch
regions it may be necessary to change the performance objective of the MPC
controllers.

4.4.3c Step 7: Structure of Optimization Layer (RTO) (related to Decision 1)


Now, the goal of the RTO layer is to update the setpoints for CV 1 and to detect changes in
active constraints regions that should be required the switching of the same CV 1.
We have to note that in most of cases with a self-optimizing choice for the choice of the
primary controlled variables, what we can obtain from this kind of system is too low to
justify the cost of creating and then sustaining the detailed steady-state model required for
this layer.
If we add the numerical issues related to the optimization (difficult) and then the offline
optimization (difficult too), the use of this last layer can be reasonably avoided. That said
in other word present the possibility to reach the same result of a RTO with an only stable,
simple and robust control, not always, but for sure in those cases for which the costs for
this solution are not comparable with the advantages we can obtain.
This is an appendix of all the procedure, that could be useful but it doesn‟t have to be a
“must”. It‟s for this reason that the last word on this chapter should be left another time to
economics, which rules from the very for step till the last one.

After the analysis of the procedure proposed and applied by Sigurd Skogestad, it will be
interesting verify this systematic approach to a process which for a long time has been used

99
FRANCESCO MORANDI SELF-OPTIMIZING CONTROL

and exploited, but relatively to its performance hasn‟t been optimized and searched for an
optimal solution: the hydrodesulfurization process.
Starting from the very first step of the above descript procedure, it will developed in order
to reach and develop a simple and robust control structure. It‟s interesting to note that
although the process is dated and it started working online in the refining process industry
during the „20s of the last century, neither it has received nor it has been modified a lot
during this lat period. Till the very recent years. In this last period cause the more severe
and stringent rules on the environment pollution has given the right propulsion to an
improvement for this process and under some points of view a new “life” for an, otherwise,
outdated plant unit.

4.5 ANALYSIS AND RESULTS


It‟s time to analyze the previous process plant and then operate on this one in order to have
the possibility to search for a stable and robust control structure. The procedure is followed
from the beginning till the last point; leaving at the end of all the systematic procedure the
possibility to apply or not to, a RTO.
In this part, the looking for application of the previous steps will follow. We can now start.
4.5.1 Step 1:Define Operational Objectives
Standing on the basis of the previous description of this step, we have to evaluate the
operational objectives in terms of a “scalar cost function” J, in this case [€/h], that has to
be minimized. To build this function it has been conducted a search in different ways
(meaning in that case the different possibilities used to calculate it) in order to find a
correct evaluation for this goal. First of all after having build a reliable model of the plant it
has been used a spreadsheet in which has been collected the main factors those are
responsible for a certain cost in the plant. In this selection has been taken into
consideration all the contributes those affected in a certain way the cost of a plant. We
must note, as it highlighted previously, that in this case and for this economic evaluation,
all the costs of devices and instrumentation (pipes, vessels, reactors, columns and tanks
too) are not of our interest cause they can be considered as sunk costs and for this reason
not worth for an analysis that aims to model a structure around something that has subject
to variation due to changes of economic nature.

100
FRANCESCO MORANDI SELF-OPTIMIZING CONTROL

Figure 46: spreadsheet with all contributes to the final cost function

As possible to individuate in the previous shot of the spreadsheet, the different


contributions of all elements in the plant has been individuate and selected weighting the
different contribution in terms of power required to work properly. It means that for this
passage has been done a research on the different costs and it has been possible evaluating
all the different values.
First of all it has been done a selection of the different contributions on the basis of their
impact on real costs. Utilities are studied and analyzed to see how strong was their
importance in this first step. For example not all pumps have been considered for the
overall count of costs because some of them had a very small impact on this balance; they
have been considered instead:
 Compressor (K-100 – K-101 on the process flow diagram)
 Heaters (E-102 – E-104)
 Pump (P-100)
In order to have the same calculation basis, the work necessary to make possible the right
working of this device has been converted in cost per hour: that means the work evaluate
by the simulation program, expressed in [kW] has been multiplied per the cost of electric
energy for what regards compressors and pumps; while for what is dealing with heaters,
their power has been estimated multiplying this value for the cost of natural gas used to
feed to furnace. This has been done without considering effectiveness factors or other
different corrections because the importance in this economical analysis of the plant what

101
FRANCESCO MORANDI SELF-OPTIMIZING CONTROL

it is really important is a global visual of all costs and how much they affect plant
economics. This assumption however does not compromise the effectiveness of the results.
So electricity and natural gas prices have been found on Europe’s Energy Portal69, and in
order to obtain a unique value it, it has been necessary to average the different value.
On the same site in has possible then to evaluate the costs for crude to treat and the value
for the product, considering diesel price without VAT. Hydrogen cost has been estimated
on the basis of industrial gases market, referring to values reported by industries dealing
with selling gases with different specification in accordance to their use (medical,
industrial or laboratory) 70 . The different values for energy utilities are reported in the
following tables.
Table 10: costs of electricity and natural gas

Country € per kWh Natural Country € per kWh


Gas Electricity
Austria 0.04021 Austria 0.11845
Belgium 0.04022 Belgium 0.09899
Bulgaria 0.03294 Bulgaria 0.06326
Czech Rep. 0.03774 Cyprus 0.16953
Denmark 0.08427 Czech Rep. 0.09993
Estonia 0.0316 Denmark 0.09774
Finland 0.04622 Estonia 0.07686
France 0.04137 Finland 0.07443
Germany 0.04966 France 0.07768
Hungary 0.04122 Germany 0.11417
Italy 0.0417 Greece 0.08994
Latvia 0.03363 Hungary 0.09593
Lithuania 0.03832 Ireland 0.09618
Luxembourg 0.04769 Italy 0.14438
Netherlands 0.04434 Latvia 0.09716
Poland 0.03428 Lithuania 0.10696
Portugal 0.04193 Luxembourg 0.07482
Romania 0.02386 Malta 0.16313
Slovakia 0.04011 Netherlands 0.09138
Slovenia 0.05574 Poland 0.08768
Spain 0.03675 Portugal 0.10086
United Kingdom 0.0286 Romania 0.07348
AVERAGE [€/kWh] 0.041472727 Slovakia 0.12163
Slovenia 0.08673
Spain 0.09927
Sweden 0.07764
United Kingdom 0.09895
AVERAGE [€/kWh] 0.099894815

69
http://www.energy.eu/fuelprices/, accessed on February 2013
70
PRICE SCHEDULE –INDUSTRIAL GASES 49009.XLS, accessed on February 2013

102
FRANCESCO MORANDI SELF-OPTIMIZING CONTROL

While for the raw material prices and for the product value in the next table there is the
summa of what the above mentioned survey:
Table 11: price and value of raw materials

EU [$/bbl] CRUDE OIL [€/bbl] [€/l]* [€/m3]


160.6 75.764 0.476504 476.5035

PRODUCT VALUE [€/bbl] *€/l+ *€/m3]


with VAT 230.550 1.45 1450
without VAT 190.323 1.197 1197

HYDROGEN COST [€/m3]


INDUSTRIAL 8.8937665
*note that all prices have been converted on the basis of actual $/€ change, February 2013

Because we want to evaluate:


J=-P
in order to minimize it, we have to determinate
J = cost feed + cost utilities (i.e. energy) – value products [money/time]
in this case71:
cost feed = actual volume flow feeds [m3/h] * cost [€/m3]
= Gas Oil Feed [m3/h] * Crude Oil [€/m3] + H2 [m3/h] * Hydrogen Cost [€/m3]
= 72.27 [m3/h] * 476.50 [€/m3] + 70.62 [m3/h] * 8.89 [€/m3]
= 34516.17 [€/h]
value products = actual volume flow product72 [m3/h] * cost [€/m3]

= Desulfurized Gas Oil [m3/h] * Product Value [€/m3]


= 74.32 [m3/h] * 1197.00 [€/m3]
= 88961.04 [€/h]
cost utilities = compressors‟ power [kW] * Electric Energy cost [€/kWh] + heaters‟ power
[kW] * Natural Gas cost [€/kWh] + pump‟s power [kW] * Electric Energy cost [€/kWh]

71
Note that all reference are taken from Aspen HYSYS® flow sheet, i.e. values for floods, properties of
feedstock and so on
72
Take into consideration that for product in the overall discussion it will be considered only the
desulfurized charge, while for the main other by product we are not concerning in, i.e. hydrogen sulfide.
Note that in case of more precision and maybe an improvement to this simulation, this value could be, for
example, evaluated in order to feed it to a Klaus process to recover sulfur.

103
FRANCESCO MORANDI SELF-OPTIMIZING CONTROL

= 1287.11 [kW] * 0.10 [€/kWh] + 33020.10 [kW] * 0.042 [€/kWh] + 328.10


[kW] * 0.10 [€/kWh]
= 1548.35 [€/h]

J = -51896.52 [€/h]
In this way it has been possible to evaluate simply and clearly our objective function. If
under some points of view it could be seemed too much simple, we have to consider that
for an economical evaluation all the more important contributes have been taken into
consideration and the important in this step is not to lose sight of the general goal: to
optimize a complete unit.
At this point we have completed the first step of the procedure: all the important terms
have been collected and it has been evaluated the objective function.

4.5.2 Step 2: Determine the steady-state Optimal Operation


As stated before, “in order to achieve a reliable and efficient control system we have to
find the optimal way of operating process”.
To determine the optimal plant condition we have to work on a steady-state model (the one
determined and built in Aspen HYSYS®), then identify DOFs and finally optimize it for
expected disturbances (having considered previously what are the most important ones and
their range of variation). That is:
J = J(u, x, d)
we optimize operation with respect to u for given d (usually steady-state):
minu J(u, x, d)
subject to:
 Model equations: f(u, x, d) = 0
 Operational constraints: g(u, x, d) ≤ 0

So in order to achieve this result we have first of all to identify the steady-state degrees of
freedom. Actually we are concerned into the number of u and what‟s the nature of this
variables. It‟s important to find a independent and simple way to find nu because it‟s useful
to check and then to reduce the time spent on optimizing the process. Even if there are the
above mentioned three methods to individuate them, we will eventually find them when we
perform optimization, i.e. analyzing the system and its behavior it is possible to highlight

104
FRANCESCO MORANDI SELF-OPTIMIZING CONTROL

some variables that have a great impact on all the structure, while for other there is not the
same emphasis. This is useful to understand which are the “real” degrees of freedom to
take into account. With respect to methods explained before, because the dimension of the
plant is reasonably big, it is useful to evaluate the DOFs counting with the “Potential
Degrees of Freedom”. That has been done because the idea to consider the equations
counting was impossible, practically because working on the Aspen HYSYS® we are not
able to evaluate all the ones the program is using to simulate the unit. Because the other
two methods are quite similar, in order to reach the easiest way to evaluate them we have
done it considering the Table 2 (here reported in order to make it easy the evaluation of the
next step), that is perfectly analogue to the valve counting method:

Equipment nu

Each external feed stream 1 Feed rate


Splitter n-1 Split fractions
Mixer 0 [-]
Compressor, turbine, pump 1 Work/speed
Adiabatic flash tank 0* [-]
Heat exchanger 1 Bypass/flow
Liquid phase reactor 1 Holdup reactant
Gas phase reactor 0* [-]
Column (no heat exchanger) 0*+n° side streams [-]
Note: pressure*: we need to add 1 DOF at each extra place you set the pressure (that means
to use an extra valve, compressor or pump).
Figure 47: Table 2

Analyzing the plant scheme of the hydrodesulfurization unit, we can start to evaluate all

Figure 48: the plant

the device and equipment, we have the following degrees of freedom:

105
FRANCESCO MORANDI SELF-OPTIMIZING CONTROL

Table 6: DOFs counting

Equipment nu name DOFs

Each external feed stream 1 Gas Oil Feed, H2 source, 3


H2O makeup
Mixer 0 MIX-100/101/102 0
Compressor, turbine, pump 1 K-100/101, P100/101 4
Adiabatic flash tank 0 HP Separator, V100 0
Heat exchanger 1 E-101/102/104/108/109 5
Liquid phase reactor 1 PFR 101 1
Column 2 T-100/101 4
TOTAL 17

It could seemed finished here the counting step, but it is necessary to make some
observations useful to reduce, without losing precision and effectiveness, the number of
these degrees of freedom. First of all is important as explained before having clear in mind
the purpose of all the project: the optimization of an HDS unit. In order to achieve this goal
is fundamental knowing the plant: the key issue is understand how it works and which are
the DOFs responsible for great changing in plant working. For this reason the previous
number of DOFs has been reduced and modified, till the value of 11 degrees of freedom,
considering and taking into consideration the importance of the different variable per each
equipment.
Table 7: list of degrees of freedom

DOFs’ name in nu

Inlet Pressure P-100 1


Reactor Temperature PFR-100 1
Cooling Temperature E-101 1
HP Separator Pressure HP Separator 1
Absorption Pressure T-100 1
H2/Gas Oil Ratio MIX-100 1
Molar Flow Absorption T-100 1
Reboiler Duty T-101 1
Inlet Temperature Regeneration E-104 1
Outlet Temperature Regeneration E-109 1
Regeneration Column Pressure T-101 1
TOTAL 11

After the identification of the degrees of freedom it has been possible to recognize and then
select the important disturbances, with their expected range.

106
FRANCESCO MORANDI SELF-OPTIMIZING CONTROL

Because the unit is designed to work with an expected feed, the most important disturbs
will come from it; while without losing in generality the disturbances in feed temperature
or pressure could be neglected because of their not so important impact.
For the optimization the disturbances have been selected as follows:
- Changing in feed flood :
50000 [kg/h] < 60600 [kg/h] < 80000 [kg/h]
- Changing in composition, meaning the total sulfur content:
4.01 % [mol/mol] < 9.55 % [mol/mol] < 14.97 % [mol/mol]
- Changing in one compound composition, double DBZT content:
2.33 % [mol/mol] < 4.5 % [mol/mol]
At this point it has been possible to start the optimization of the system, firstly optimizing
the nominal point of working (the “normal” operative conditions), then disturbing the
system and optimizing it for the expected disturbances. In this way it has been possible
evaluate the optimal sensitivity of the system.
In order to solve this problem it has been built a program using Matlab®: in particular
the .m file it would be result in an optimization code, which calling inputs directly from
Aspen HYSYS® tries to optimize the previous objective function acting on the above
mentioned DOFs; the values find in this way are then sent again to Aspen HYSYS® which
evaluates another time the objective function. All this procedure is repeated till a feasible
solution is found with the respect of the operative constraints and equalities set for the
model.
In this case the operational constraints has been set to stream:
Table 12: operational constraints set to the system

Lower bound value ≤ Constraints ≤ Upper bound value


1 Delta Pressure [kPa] 10
300 Inlet Temperature Reactor [°C] 500
10 Cooling Temperature [°C] 110
1.2959E+04 Inlet Pressure [kPa] 2.5E+04
1.919 H2/Gas Oil Ratio [-] 3
5.10E+03 Absorption Pressure [Pa] 7.0E+03
1E+07 Reboiler Duty [kJ/h] 1E+08
0.278 Molar flow ABS column [kgmol/h] 0.556
40 Inlet Temperature Regeneration [°C] 99
30 Outlet Temperature Regeneration [°C] 60
130 Regeneration Column Pressure [kPa] 190
the total sulfur content “Desulfurized Gas Oil” *% wt/wt+ ≤ 0.007
title of stream “H2 to Compressor” = 1

107
FRANCESCO MORANDI SELF-OPTIMIZING CONTROL

Note that title of “H2 to Compressor” = 1 is fixed to be equal to 1(i.e. in order to avoid the
compressor‟s collapse; because for this purpose it is usually selected an axial-flow
compressor, whose fan-like airfoils are susceptible of damage and failing in case of
presence of liquid fluids) and the non correct working of the plant; and the total sulfur
content “Desulfurized Gas Oil” [% wt/wt] ≤ 0.007 (at the exit of the unit it has to be in
accordance with the environmental specification).
The function called by Matlab® for this purpose is fmincon73, which attempts to find a
constrained minimum of a scalar function of several variables starting at an initial estimate
(the initial point). This is generally referred to as constrained nonlinear optimization or
nonlinear programming.

Figure 49: fmincon function, MATHWORKS, fmincon, http://www.mathworks.it/it/help/optim/ug/fmincon.html,


accessed on March 2014

What it is worth of noting is that this function uses optimizers which are not very robust.
This results in a not unique solution, which is instead function of the initial estimate of it.
For this reason to obviate to this problem the optimization has been repeated again and
again, with the disturbances selected in order to see what was the solution or the solutions
able to “absorb” better them and with a greater value of optimum. Anyway this is not a
huge problem, because we have to consider another particular i.e. that we are looking for a
solution, an optimal solution, which won‟t be the absolute optimum. This will be a relative
optimum that is in turn function of disturbances, and it will be the control of the system
which will have the duty to move it to the optimum. So for this reason it is important to
find the optimal point, but not so necessary that it has to be the absolute one.

4.5.2a Results of the Optimization


The simulations have been done on a laptop with Windows 7, operative system 32 bit,
RAM 2GB and Intel® Pentium® Dual CPU T2370 @ 1.73[GHz] 1.73 [GHz].
73
MATHWORKS, fmincon, http://www.mathworks.it/it/help/optim/ug/fmincon.html, accessed on March
2014

108
FRANCESCO MORANDI SELF-OPTIMIZING CONTROL

Matlab® version was R2011b (7.13.0.564).


Here it follows the result of the optimization. Note that have been found 6 different
“configurations”, that means six possible combinations of different value for the degrees of
freedom that result in an optimal solution starting from different initial values.
Each configuration is composed by the following “parameters”:
Table 13: DOFs with their unit of measurement

DOFs’ name [Unit of Measurement]


Inlet Pressure [kPa]
Reactor Temperature [°C]
Cooling Temperature [°C]
HP Separator Pressure [kPa]
Absorption Pressure [kPa]
H2/Gas Oil Ratio [-]
Molar Flow Absorption [kgmole/h]
Reboiler Duty [kJ/h]
Inlet Temperature Regeneration [°C]
Outlet Temperature Regeneration [°C]
Regeneration Column Pressure [kPa]

So when there is the reference to “configuration” we have to understand that it is a


generalized way to consider the different optimal values found by the optimization
program.

Cost Function Optimized


-51000
1 2 3 4 5 6
-51500

-52000
Cost [€/h]

-52500

-53000

-53500

-54000
Configurations

Figure 50: cost function optimized, no disturbances

109
FRANCESCO MORANDI SELF-OPTIMIZING CONTROL

Perturbation I: lower feed


-41000
-41500 1 2 3 4 5 6
-42000
-42500
Cost[€/h]

-43000
-43500
-44000
-44500
-45000
-45500
-46000
Configuration

Figure 51: changing in feed flood

Perturbation II: higher feed


-66000
1 2 3 4 5 6
-67000

-68000
Cost[€/h]

-69000

-70000

-71000

-72000

-73000
Configuration

Figure 52:changing in feed flood

110
FRANCESCO MORANDI SELF-OPTIMIZING CONTROL

Perturbation III: lower sulfur content


-51500
-52000 1 2 3 4 5 6
-52500
-53000
-53500
Cost[€/h]

-54000
-54500
-55000
-55500
-56000
-56500
-57000
Configuration

Figure 53: changing in feed composition

Perturbation IV: higher sulfur content


-49000
-49500 1 2 3 4 5 6
-50000
-50500
Cost[€/h]

-51000
-51500
-52000
-52500
-53000
-53500
-54000
Configuration

Figure 54: changing in composition

111
FRANCESCO MORANDI SELF-OPTIMIZING CONTROL

Perturbation V: double DBZT content


-50500
1 2 3 4 5 6
-51000
-51500
-52000
Cost[€/h]

-52500
-53000
-53500
-54000
-54500
Configuration

Figure 55: changing in one feed’s compound’s composition

After this re-optimization due to disturbances, all results have been collected in order to
see what was the best solution, here meant as the best combination of different degrees of
freedom‟s values. It has been a quite empirical method on the way of the “brute force”
system, but it has responded in a good way.
Time consuming of each optimization was about three hours and a half; time that could be
reduce significantly maybe changing the machine or finding a best optimizer. Anyway the
problem remains that those are very time consuming simulations.

112
FRANCESCO MORANDI SELF-OPTIMIZING CONTROL

Comparison
-40000

-45000

Optimal Normal Operative


-50000 Conditions
Perturbation I
-55000
Cost[€/h]

Perturbation II
-60000
Perturbation III
-65000
Perturbation IV
-70000
Perturbation V
-75000

-80000
Configuration

Figure 56: comparison between the different optimizations with the respect of disturbances and their expected range

As you can see there is not a great difference among all the simulation/configuration, that
is due to the fact that all are relative optima of the objective function. But analyzing with
more attention configuration 3 and 4 we can note that those are the best in terms of less
loss from the optimal normal operative conditions.
At this point we have completed the second step.
And here we report the results of this simulation:
Table 14: results of the third configuration

RUN
Initial Value Optimal result DOFs Cost Function, J
3 1,123185427 Delta Pressure [kPa] -53274,30429
350 350,0028264 Inlet Temperature Reactor [°C]
50 64,51189265 Cooling Temperature [°C]
1,30E+04 12958,68577 Inlet Pressure [kPa]
2,43 2,400512383 H2/Gas Oil Ratio [-]
6,10E+03 6100,019689 Absorption Pressure [Pa]
13888,88889 50000015,72 Reboiler Duty [kJ/h]
0,430555556 1159,504081 Molar flow ABS column [kgmol/h]
60 50,00011585 Inlet Temperature Regeneration [°C]
35 30,93073656 Outlet Temperature Regeneration [°C]
150 150,294351 Regeneration Column Pressure [kPa]

113
FRANCESCO MORANDI SELF-OPTIMIZING CONTROL

4.5.3 Step 3: Select Economic (Primary) Controlled Variables, CV1


This is the part which is dealt with the implementation of optimal points found in the
previous two steps and their organization in a robust and simple manner.
What we have obtained from the previous analysis is the economic degrees of freedom (u),
and now what we want to reach is identify the economic controlled variables (CV1). Or
better in this case we want to verify also if those are the right variables for the degrees of
freedom.
Remember that the number of first controlled variables is equal to steady-state degrees of
freedom one. But what is needed to control?
So the optimal operation for given d:
minu J(u, x, d)
subject to:
Model equations: f(u, x, d) = 0
Operational constraints: g(u, x, d) ≤ 0

The result of this study has been the uopt, whose values however usually cannot be kept
constant because disturbances d change, and in this way the optimal results change too.
However for the economical optimal operation, we have two rules, the ones mentioned
before:
 CV1 – rule 1, that is control active constraints
 CV1 – rule 2, i.e. for the remaining unconstrained DOFs, we must control self-
optimizing variables.
From the analysis previously conducted, we have found that inlet pressure is an active
constraint of the plant, that means that it is not susceptible for changing and that it has to
be set to its maximum (in this case, but it depends on the nature of the variable) value.
Together with this degree of freedom we have to take into account, seen from the previous
analysis of the results of the optimization the temperature the H2/Gas Oil ratio and the
temperature out from the regeneration column. These considerations make possible to
reduce the remaining degrees of freedom for which is then necessary to select a controlled
variable. So as stated by rule 1, we must control active constraints; the reason is quite
obvious because we have to understand that usually the optimum is not “flat” with respect
to these variables: this means that there is often an economic penalty if we back off from
those variables and we must add a tight control of them.

114
FRANCESCO MORANDI SELF-OPTIMIZING CONTROL

In summa our active constraints could be assumed as (observing the precedent constraints
when they become active):
- Inlet pressure
- Outlet temperature regeneration
- H2/Gas Oil ratio
So we have assigned three CVs to the same number of degrees of freedom. And for the
other ones? The solution lies in controlling of the self-optimizing variables. At this step 10
degrees of freedom remain to be assigned. It is now necessary identify the candidate
measurements and then select them on the basis of CV 1 = H · y, as explained before.
In order to accomplish this goal we have to follow the procedure seen previously.
From the analysis of the system we can use the previous measurements associated with the
DOFs to see if they were good candidate to be self-optimizing variables. To achieve this it
has been used one of the main approaches developed to find the matrix H; having excluded
the “brute force” approach, the problem has been solved with the “local” approach: the so
called Nullspace method74,75. Let we see a brief theoretical explanation.
4.5.3a The Nullspace Method
We consider the unconstrained optimization problem as given in the following form, the
same seen previously:
minu J(u, d)
that is, we assume “active constraint control”, where all optimally constrained variables
are assumed to be kept constant at their optimal values. Our goal is to find a linear
measurement combination c = Hy to be kept at constant setpoints cs. Here, H is a constant
nu x ny matrix and y is a subset of the available measurements.
To make possible to apply this procedure we have to make some assumptions:
1- Steady state: We consider only steady-state operation. The justification for this is
that the economics of operation is primarily determined by the steady state; directly
from this assumption is that a control system in place that can quickly bring the
plant to its new steady state.

74
See it on pages 27-28, and VIDAR ALSTAD, SIGURD SKOGESTAD, EDUARDO S. HORI, Optimal measurement
combinations as controlled variables, Journal of Process Control 19, 138–148, Elsevier Ltd., 2009
75
V. ALSTAD AND S. SKOGESTAD, Null Space Method for Selecting Optimal Measurement Combinations as
Controlled Variables, Ind.Eng.Chem.Res, 46 (3), 846-853 (2007)

115
FRANCESCO MORANDI SELF-OPTIMIZING CONTROL

2- Disturbances: Only disturbances that affect the steady-state operation are included
(in this case we assume the disturbances seen before  Feed flow and feed
composition).
3- Active constraint control: the same active constraints remain active for all values of
the disturbances and that we control these constraints (see CV1 – rule 1).
4- No implementation error: The implementation error is the sum of the control error
and the effect of the measurement error.
It is a more serious assumption to neglect the measurement error, so the method implicitly
assumes that the measurements have been carefully selected (this is the limitation that the
exact local method tries to eliminate).
So firstly has been obtained the optimal measurement sensitivity F, defined as F = dyopt/dd

However, in practice, it is usually easier to obtain F directly, by optimizing for the selected
disturbances using a nonlinear steady-state model of the plant: because we are going to
find F numerically we use a commercial steady-state process simulators, i.e. Aspen
HYSYS®. Just to note that in theory, one may even obtain F from experiments on a real
operating plant, but this method is sufficiently accurate.
So numerically, the ny x nd matrix F may be obtained by perturbing the disturbances d and
re-solving the optimization problem with the active constraints being constant, that is:
1- Under nominal conditions (d = d*), use the steady-state model to obtain the
nominal optimum yopt(d*) and identify the active constraints (finding the nominal
optimum may be difficult, because the optimization problem is generally
nonconvex)  Step 2.
2- For each of the nd disturbances, make a small perturbation and resolve the
optimization with the constant active constraints to obtain yopt(d).
3- Compute yopt = yopt(d) - yopt(d*) and obtain F numerically (note that a minimum
of nd + 1 optimization runs are required.
The next step is to obtain H. Numerically, H may be obtained from a singular value
decomposition of FT. We have HF = 0 or, equivalently, FTHT = 0. Thus, selecting HT as
the input singular vectors of FT, corresponding to zero singular values in FT, gives an

116
FRANCESCO MORANDI SELF-OPTIMIZING CONTROL

orthogonal basis: then the matrix H is find because it has to be respected the condition HF
= 0. Note that, as stated before, because of H is a non-square matrix, it is not necessary that
H is equal to 0, rather that H is in the Nullspace of FT.

The corresponding optimal change in the controlled variables is


copt(d) - copt(d*) = H(yopt(d) - yopt(d*)
and by the previous explanation we get that
copt(d) - copt(d*) = HF(d - d*)
From which we know that the constant setpoint policy is optimal if:
copt(d) - copt(d*) = 0
As usual remember that:
 This method reach its best results and optimal measurement combinations in case
of no noise: ny = 0.
 Each column of F express the optimal change in y when the unconstrained variable
u is adjusted in order to have a system remaining optimal with respect to
disturbances d.
In this sense it has been calculated the optimal sensitivity first of all optimizing the system
(in order to obtain the “nominal point”, ynominal) then with the re-optimizing the system
after the disturbances, it has possible to evaluate the new working value of the variables ,
yddi. Considering the step as ddi = ddinominal + step, where ddi is the value of the disturbance
and ddinominal the value of the disturbed variable associated with the nominal point (ynominal);
it is possible define the optimal sensitivity, F approximating the differentiation as ratio
between difference of the two optimal measurements (ynominal and yddi) and the value of the
step. This is a valid numerical approximation which can give optimal results in short time.
That is:

117
FRANCESCO MORANDI SELF-OPTIMIZING CONTROL

This matrix will be then transposed in order to have the desired H T matrix; numerically the
operation has been done with Matlab®, using null76 function.
In order to relevance, first of all it has been necessary to identify the candidate
measurements in a number that has to be at least equal to the sum of disturbances and
DOFs (unconstrained, because the active constraints have been set to be constant),
#y ≥ #d + #u
#y ≥ #2 + #8
#y ≥ 10
So a number of minimum 10 measurements are required, but we have to take into
consideration, in general, the inputs (for example the flow rates), with those used to control
active constraints; remember that to make possible to use this method we have to use a
number equal to the sum of disturbances and unconstrained degrees of freedom (that will
result in a matrix of the same dimensions). That is quite clear if we think that this control
has to avoid the system drift away from the optimal nominal point. We consider two sets of
measurements in order to see what are the best candidates to be the self-optimizing
variables.
Table 15: first candidate measurements

y
gas oil to heat temperature [°C]
reactor inlet mass flow [kg/h]
reactor product temperature [K]
H2O make up mass flow [kg/h]
flash VAP mass flow [kg/h]
rich MDEA to flash temperature [°C]
sweet gas temperature [°C]
regeneration bottom temperature [°C]
H2 to compressor temperature [°C]
desulfurized gas oil total sulfur content [% wt/wt]

76
Z = null (A) is an orthonormal basis for the null space of A obtained from the singular value
decomposition, MATHWORKS, null, http://www.mathworks.it/it/help/matlab/ref/null.html, accessed on
March 2014

118
FRANCESCO MORANDI SELF-OPTIMIZING CONTROL

Table 16: second candidate measurements

y
reactor inlet temperature [K]
cooling temperature [°C]
inlet pressure [kPa]
absorption pressure [kPa]
reboiler duty [kJ/h]
molar flow ABS column [kgmol/h]
inlet temperature regeneration [°C]
regeneration pressure [kPa]
H2 actual volume flow [m3/h]
k-100 power [kW]

Then the system has been disturbed taking into consideration the main two disturbances
associated with this kind of process, that are: the mass flow of the feed and the total sulfur
content in the feedstock that has to be removed during the processing of the charge.
Table 17: disturbances and their value after small perturbation

Disturbance
Mass Flow [kg/h] 60,600
3000 63600
-3000 57600
Composition [% wt/wt] 0.077
0.005 0.082
-0.005 0.072

And it has been evaluated the F matrix in order to have an idea of the “impact” of
disturbances on the process unit:
Table 18: positive disturbance in mass flow, first candidate measurements

Mass Flow +
y(d*) y(d) y y
45.29123785 45.2912378501 0.0000000000 gas oil to heat temperature
62737.70802687 65758.63656 3,020.9285334023 reactor inlet mass flow
630.01178930 636.2049418 6.1931525538 reactor product temperature
40.71074626 36.94205737 -3.7686888891 H2O make up mass flow
0.32926267 0.309765347 -0.0194973191 flash VAP mass flow
33.63208048 33.55132885 -0.0807516310 rich MDEA to flash temperature
35.05421281 35.03825373 -0.0159590843 sweet gas temperature
122.89776011 122.8977601 0.0000000000 regen bottom temperature
124.46109664 123.3321516 -1.1289450488 H2 to compressor temperature
0.00145463 0.001170 -0.0002849289 desulfurized gas oil total sulfur
content

119
FRANCESCO MORANDI SELF-OPTIMIZING CONTROL

Table 19: negative disturbance in mass flow, first candidate measurements

Mass Flow -
y(d*) y(d) y y
45.29123785 45.2912378501 0.0000000000 gas oil to heat temperature
62737.70802687 59667.63157 -3,070.07646175 reactor inlet mass flow
630.01178930 682.0213114 52.0095221169 reactor product temperature
40.71074626 42.72250998 2.0117637245 H2O make up mass flow
0.32926267 0.339320942 0.0100582768 flash VAP mass flow
33.63208048 33.6740639 0.0419834128 rich MDEA to flash temperature
35.05421281 35.02923462 -0.0249781930 sweet gas temperature
122.89776011 122.8977601 0.0000000000 regen bottom temperature
124.46109664 126.012413 1.5513163739 H2 to compressor temperature
0.00145463 0.000437 -0.0010180343 desulfurized gas oil total sulfur
content

Table 20: positive disturbance in composition, first candidate measurements

Composition +
y(d*) y(d) y y
45.29123785 45.29108239 -0.0001554604 gas oil to heat temperature
62737.70802687 62721.48632 -16.2217034520 reactor inlet mass flow
630.01178930 558.7717489 -71.2400404310 reactor product temperature
40.71074626 39.87849367 -0.8322525828 H2O make up mass flow
0.32926267 0.32499145 -0.0042712157 flash VAP mass flow
33.63208048 33.60282563 -0.0292548581 rich MDEA to flash temperature
35.05421281 35.03474995 -0.0194628602 sweet gas temperature
122.89776011 122.8977601 0.0000000000 regen bottom temperature
124.46109664 123.8306725 -0.6304240880 H2 to compressor temperature
0.00145463 0.005795 0.0043403701 desulfurized gas oil total sulfur
content

Table 21: negative disturbance in composition, first candidate measurements

Composition -
y(d*) y(d) y y
45.29123785 45.2914444714 0.0002066212 gas oil to heat temperature
62737.70802687 62717.13668 -20.5713466485 reactor inlet mass flow
630.01178930 631.0971491 1.0853598516 reactor product temperature
40.71074626 39.70022117 -1.0105250844 H2O make up mass flow
0.32926267 0.325021184 -0.0042414817 flash VAP mass flow
33.63208048 33.59593921 -0.0361412732 rich MDEA to flash temperature
35.05421281 35.03684367 -0.0173691445 sweet gas temperature
122.89776011 122.8977601 0.0000000000 regen bottom temperature
124.46109664 121.4552614 -3.0058352328 H2 to compressor temperature
0.00145463 0.001087 -0.0003674347 desulfurized gas oil total sulfur
content

120
FRANCESCO MORANDI SELF-OPTIMIZING CONTROL

Table 22: positive disturbance in mass flow, second candidate measurements

Mass Flow +
y(d*) y(d) y y
613.00282639 620.207308 7.2044816892 reactor inlet temperature [K]
64.51189265 49.308238 -15.20365467 cooling temperature [°C]
12958.68577127 12958.685771 0.0000000000 inlet pressure [kPa]
6100.01968930 6099.996326 -0.02336339 absorption pressure [kPa]
50000015.715747 49999995.8515 -19.8642330 reboiler duty [kJ/h]
1159.50408148 1159.504081 0.00000000 molar flow ABS column [kgmol/h]
50.00011585 59.718520 9.71840443 inlet temperature regeneration [°C]
150.29435100 150.503542 0.20919095 regeneration pressure [kPa]
44.78198883 72.413558 27.63156914 H2 actual volume flow [m3/h]
775.33290728 880.861136 105.528229 k-100 power [kW]

Table 23: negative disturbance in mass flow, second candidate measurements

Mass Flow -
y(d*) y(d) y y
613.00282639 666.4425025 53.4396761375 reactor inlet temperature [K]
64.51189265 71.21896752 6.7070748660 cooling temperature [°C]
12958.68577127 12958.68577 0.0000000000 inlet pressure [kPa]
6100.01968930 5747.715073 -352.3046161195 absorption pressure [kPa]
50000015.715747 50615149.98 615,134.267744 reboiler duty [kJ/h]
1159.50408148 1159.504081 0.0000000000 molar flow ABS column [kgmol/h]
50.00011585 50.00000093 -0.0001149250 inlet temperature regeneration [°C]
150.29435100 150.0000029 -0.2943480889 regeneration pressure [kPa]
44.78198883 67.63310187 22.8511130392 H2 actual volume flow [m3/h]
775.33290728 873.2714304 97.9385231524 k-100 power [kW]

Table 24: positive disturbance in composition, second candidate measurements

Composition +
y(d*) y(d) y y
613.00282639 542.038878 -70.9639485548 reactor inlet temperature [K]
64.51189265 61.480053 -3.0318398951 cooling temperature [°C]
12958.685771 12958.685771 0.0000000000 inlet pressure [kPa]
6100.0196893 6097.764008 -2.2556810765 absorption pressure [kPa]
50000015.7157 50011629.9247 11,614.20903 reboiler duty [kJ/h]
1159.504081 1159.504081 0.0000000000 molar flow ABS column [kgmol/h]
50.00011585 51.725190 1.7250743287 inlet temperature regeneration [°C]
150.29435100 150.000311 -0.2940404841 regeneration pressure [kPa]
44.78198883 69.501676 24.7196875204 H2 actual volume flow [m3/h]
775.332907 841.512105 66.1791973661 k-100 power [kW]

121
FRANCESCO MORANDI SELF-OPTIMIZING CONTROL

Table 25: negative disturbance in composition, second candidate measurements

Composition -
y(d*) y(d) y y
613.00282639 615.7160074 2.71318098 reactor inlet temperature [K]
64.51189265 66.05463659 1.54274394 cooling temperature [°C]
12958.68577127 12958.68577 0.00000000 inlet pressure [kPa]
6100.01968930 6104.307174 4.2874843925 absorption pressure [kPa]
50000015.71574 50008574.84 8,559.128937 reboiler duty [kJ/h]
1159.50408148 1159.504081 0.0000000000 molar flow ABS column [kgmol/h]
50.00011585 50.01475998 0.0146441283 inlet temperature regeneration [°C]
150.29435100 150.0047448 -0.2896062309 regeneration pressure [kPa]
44.78198883 67.52268825 22.7406994 H2 actual volume flow [m3/h]
775.33290728 833.7486652 58.4157578 k-100 power [kW]

Figure 57: results of the Nullspace method applied to the previous optimal configuration, first two matrices

Where H_T1 and H_T2 are the matrices dealing with positive (the first one) disturbance in
feed flow and composition too, while the second one has undergone to a negative
disturbance in composition and feed flow. Both are the H matrix in which we could “see”
how the remaining unconstrained DOFs are connected to some variables. They are 10x8
matrices that means we have in row the different measurements and in column the

122
FRANCESCO MORANDI SELF-OPTIMIZING CONTROL

unconstrained degrees of freedom, where the self-optimizing variables are meant as


possible combination of measurements. An increasing prediction could be done trying to
increasing the measurements or selecting them with more accuracy.

Figure 58: results of the Nullspace method applied to the previous optimal configuration, second two matrices

In these two second matrices the value represented are in H_T3 and H_T4 a positive (the
first one) disturbance in feed flow and a negative disturbance in composition, while the
second one has undergone to a negative disturbance in feed flow and a positive disturbance
in composition (we have to remember that each H matrix is considering the possible
disturbances and so each of them has a feed and composition disturbance). The same work
has been done with the second candidate measurements (we report them in order to see the
most significant changing in their value).

123
FRANCESCO MORANDI SELF-OPTIMIZING CONTROL

Figure 59: results of the Nullspace method applied to the previous optimal configuration, with the second set of
measurements

124
FRANCESCO MORANDI SELF-OPTIMIZING CONTROL

As we can see from the previous results (all H matrices) we can note how values of the
first row are higher (absolute value) with respect to the other one and that is because it is
dealing with the variation of reactor inlet temperature and gas oil to heat temperature, those
affect the behavior of all the unit. So they are to be measured the former and control the
latter one variable. Same speech could be done about the fourth row, “absorption pressure”,
because it significantly affect the grade of purification of the hydrogen to be recycled to
the plant. All these variables have to be considered self-optimizing and in this sense they
have to be controlled or in a different way measured and used to make more robust the
control.

4.5.4 Step 4: Select the Location of TPM (Decision 3)


For this point as discussed previously, there are different approaches and solution on which
could be the best location for the TPM. In this case, because the basis idea of the process is
to transform some raw materials into products whose value is higher than the feedstock, in
an amount that could be justify the expense of the same plant, it is clear that the easiest and
the most reliable position of the throughput manipulator is at the beginning of the plant on
the feeding line.
It is reasonable acting in this way instead moving it through the plant because in this case
as in almost all the other the easiest solution is quite always the best.

Figure 60: location of the TPM, select at the beginning of the plant

125
FRANCESCO MORANDI SELF-OPTIMIZING CONTROL

4.5.5 Step 5: Select the Structure of Regulatory (Stabilizing) Control Layer


With this step we aim to stabilize the plant, preferring a simple control structure with
single-loop PID controller (reliable of a more robust control). As seen in the previous
explanation “stabilize” means that the process doesn‟t drift too far away from acceptable
operation even if there are some disturbances.
In this step there are two faced two main decision:
1- Select controlled variables (CV2), i.e. Decision 2
2- Select inputs (that is valves) and “pairings” in order to control the above mentioned
variables (CV2), i.e. Decision 4
We have not to add any degrees of freedom because the setpoints CV 2s are left as
manipulated variables (MVs) for the supervisory layer.
If we consider to allow for cascade loops, then the stabilization layer may be designed
independent of the supervisory control layer (i.e. the economic one).
SELECT STABILIZING CV2 (Decision 2)
Firstly we have to stabilize the process by controlling drifting variables such level and
pressure, reactor temperature and temperature profile in the distillation column. In this case
we want to control:
1- Levels: the HP Separator, in the Absorption Column (T-100) and in V-100.
2- Pressures/Temperatures, in case of necessity
This simplifies the supervisory control layer and its tasks because it provides for local/fast
disturbances rejection and reducing the non-linearity in the model. Then we should include
active constraints (CV1) those need a tight control (they are usually hard output
constraints) in CV2 for the regulatory layer. In this way we reduce the required back-off.
Because of all these reasons it is usually not necessary for tight control of unconstrained
CV1, indeed the optimum is quite “flat”.

4.5.6 Step 6: Select Structure of Supervisory Control Layer


At this point of the explanation we remember that this control layer has three principal
tasks: firstly it has to control the CV1 in such a way that it is dealing with the exploitation
of the set points to the regulatory layer and using the remained unused valves (meant as
MVs); then it has to supervise the performance of the previous layer of control, i.e. the
regulatory one, and finally it is involved in switching controlled variables and control

126
FRANCESCO MORANDI SELF-OPTIMIZING CONTROL

strategies when we have some disturbances on price: this means that in some cases we
have the possibility to have a switch from a region of active constraints to another one.
Among the two possibilities of controller, advanced single loop control and multivariable
control, we will privilege the first one (PID) instead of the second (MPC); and the reason is
quite intuitable: we want to grant robustness and at the same time to improve the response
of the dynamic system; so when it could be possible we must select and adopt the easiest
solution, that is not always so banal to be refused.
We now analyze the control structure proposed to obtain the no-drifting away of the unit
during its working. The first structure was proposed by Professor Skogestand and it has
been then modified with Professor Manenti in order to be more significant of the real
meaning of this process, taking into consideration all the previous dissertations and trying
to apply all of them.
In red we can see the CV1 and the regulatory control acting on them, while in green the
remaining CV2 those are the stabilizing CV2.

127
FRANCESCO MORANDI SELF-OPTIMIZING CONTROL

Figure 61: HDS control structure

128
FRANCESCO MORANDI SELF-OPTIMIZING CONTROL

This is the structure proposed in order to compensate the disturbances and operate always
near to the optimal working point, granting stability and at the same time the less loss
possible. To the complete explanation of all the system, please refer to Chapter 5.

4.5.7 Step 7: Structure of Optimization Layer (RTO) (related to Decision 1)


For this last point we have to understand that it could be the normal way to conclude such a
kind of optimization, nut it has not to be the only reason to guide such an analysis. This is
due because the goal of the RTO layer is to update the setpoints for CV 1 and to detect
changes in active constraints regions that should be required the switching of the same
CV1; but in most of cases with a self-optimizing choice for the choice of the primary
controlled variables, what we can obtain from this kind of system is too low to justify the
cost of creating and then sustaining the detailed steady-state model required for this layer.
For this project this choice to make use of the RTO layer has been refused instead a
dynamical simulation based on the response to different disturbances.
Moreover for this system it was quite impossible to propose, having seen the slowness of
the optimization step. It could be useless such a kind of device if it could not be used
properly because the system did not permit such a rapid optimization. If we add the
numerical issues related to the optimization (difficult) and then the offline optimization
(difficult too), the use of this last layer can be reasonably avoided.

129
FRANCESCO MORANDI SELF-OPTIMIZING CONTROL

Chapter 5
Dynamic Simulation: tuning and scenarios

“Run”
Snow Patrol
5.1 INTRODUCTION
It is now time to face the control structure derived from the previous analysis and the
previous assumptions. For this project it has been assumed the possibility to control ll with
a PID control. That means for the supervisory control layer the choice between MPC and
PID it has been guided to the second one; for simplicity and for the robustness it is at the
same time to grant.
In this sense we have to control the CV1 in such a way that it is dealing with the
exploitation of the set points to the regulatory layer and using the remained unused valves
(meant as MVs). We now proceed with a brief explanation of what does it mean to use a
PID control and then the result of the dynamic simulation will be presented.

5.2 PID77
Commonly used in industrial control systems the PID is a control loop feedback
mechanism (controller). This device of control works because it is able to calculate an
"error" value as the difference between a measured process variable and a desired setpoint.
This controller aims to minimize the error by adjusting the process control outputs.
In this way the controller exercises three actions:
 the proportional one (P), where the signal emitted by the regulator is
“proportional” to the error and we have such an action that:

where Kc is the static proportional gain, cs is the bias (signal when  = 0)

77
PIO FORZATTI, LUCA LIETTI, Strumentazione chimica industriale, Volume 1, Milano: CUSL, 2003

130
FRANCESCO MORANDI SELF-OPTIMIZING CONTROL

 the integral action (I), because the only proportional action involves a residual
error (i.e. the offset), this could be deleted with the integral action:

Usually these two actions are combined to reduce completely the error and avoid its
persistence as offset (which means that the regulating signal is proportional to the
error and to its integral in time), that is the PI control:

 the derivative action (D), it is requested when we have to prevent and anticipate
the error effect, we introduce the derivative action, where the action is proportional
to its derivative. Derivative action predicts system behavior and thus improves
settling time and stability of the system, even if it is rarely used in the practice.

More often its action is combined with the above mentioned control inputs, the so
called PID control:

In a more general sense we can state that the above mentioned actions are able to grant a
particular response of the system; in particular the readiness and quickness of the
regulation operated by the proportional action, while the integrative one is responsible of
the precision and the accuracy. The derivative action instead is the term able to grant
robustness to the response.
Even if all of them seems desired actions in all kind of control (in this sense in every
variable to be controlled), more often we prefer the control acts with one or maybe a
combination of them, without using them all in an aprioristic way.

5.3 TYPICAL REGULATOR CHOICE


We have usually to consider that the choice is made on the basis of the characteristics of
the control action (in this sense, P, PI, PID, seldom PD) and on the dynamics of the process.
Here reported there are some considerations about the kind of regulator.

131
FRANCESCO MORANDI SELF-OPTIMIZING CONTROL

In some cases it is preferable to use a different control and we could take into consideration
MPC controller, or some adaptive controller. But in order to show the easiness of this
control procedure it will be used a simple controller too.

5.3.1 FLOOD
This is a variable with a rapid response, with some disturbances due to turbulence and
pressure‟s pulsations (linked to pumps‟ work). In this case it is preferable to use a
proportional regulators with the integral action too. That is suddenly explained: we have to
grant a rapid action (proportional contribute, P) and at the same time, because we have to
deal with disturbances that would remain uncontrolled otherwise, we have an accurate
response (integral contribute, I). This coupled action is reliable of rapidity and precision
without leaving offset at transient spent, PI.
5.3.2 PRESSURE
In this case we have to make a distinction among the phase of the fluid has to be
controlled; in particular:
 Liquid: because liquid pressure is going to change with the square of the flood (i.e.
P ∝ Q2) and as stated before noise due to pumping machines are always present we
have to control them with only PI control.
 Gas: we can consider as general rule the use of the only proportional, P, action,
that is due to the fact that usually it is necessary “only” to keep this variable
between two different values (upstream and downstream) at a constant value.
 Vapor: in this last case we have to consider the processes involved and more
precisely we have that for a tank with vapor to storage or a simple pipeline we
have to use a proportional and integral action (PI), to grant precision and a certain
rapidity of response; if we have, instead, to control the pressure of a partial
condenser it is preferable to have a PID controller: that is because the process of
heat transmission is slow and we have to grant, and most of all, increase the
response fastness.
5.3.3 LEVEL
Also in this case we have to make some considerations in order to give some thumb rule to
grant a reliable response of the controller. In particular we have to deal with the
dimensions of the vessel, that for practice are considered as cylinder:

132
FRANCESCO MORANDI SELF-OPTIMIZING CONTROL

 Great diameter: in this case is more than sufficient a proportional, P, controller (we
have a high process time constant, i.e. slow dynamics).
 Small diameter: because we have a smaller process time constant and we need
more precision it is usually accepted the use of a proportional and integral action,
PI.
The derivative action is rarely used and it has to be refused if we have to treat with
boiling or agitated liquids.
5.3.4 TEMPERATURE
As mentioned before for pressure control, we have to consider first of all the process at the
basis of what we are going to control. In this case because we are dealing with the control
of temperature and the process we have to consider is heat transmission, typical of systems
with slow response, we need (or better it is preferable to use) the combination of all the
three above mentioned actions, that is PID controller. That is resulting an robustness of the
regulation without too many oscillations among the set point.
5.3.5 COMPOSITION
At the basis of these processes we have usually mixing or separation processes. But
because this kind of analyzers are dealing with dead time and their regulation is sometimes
critic, their use is limited to supervisor function instead of a direct control. For all these
reasons the main and more sensate control it is normal to use a combination of PID action,
granting all the necessary actions involved with this kind of regulators.

5.4 THE PLANT CONTROL


After having disserted about the different action of a PID regulator, we can now see the
control applied to the process with all the previous assumptions derived by the self-
optimizing control procedure; and In particular we have the optimal set points coming
from the optimization of the system unit (those are fixed and for whose we have to grant an
accurate control and in the case of the active constraints a tight control too). Remembering
the previous control diagram we can see that all active constraints (the one which have
been found to be active) are controlled tightly. That is possible in practice fixing a short
range for PV (i.e. process variable) and controlling with accuracy the different PID
parameters, each of the three actions usable. The control has been done in order to control
all the DOFs encountered during the analysis (useful for CV 1 layer) and then introducing
more control in order to improve and avoid the drifting away of the system.

133
FRANCESCO MORANDI SELF-OPTIMIZING CONTROL

Here the control structure:

Figure 62: the HDS control

134
FRANCESCO MORANDI SELF-OPTIMIZING CONTROL

As it is possible to note all streams have been controlled, with a flood control (FC) in the
previous control scheme (see Chapter 4). That give reason to the main disturbances take
into considerations, and make possible a good control in order to achieve the steady state
operative conditions. A point that has to be highlighted is the fact that to store the feed and
flash it at right conditions (i.e. to atmospheric pressure and reasonable temperature) it has
been introduced a tank out of the line of processed gas oil charge. It is only a little
modification that does not affect the results obtained previously, but it is more correct
under the perspective of a real processing unit (with this shrewdness it is possible to better
control the level of the treated charge).
Anyway, the control of the system has been introduced in order to smooth disturbances and
make it possible the right working in conditions as nearest as possible to the optimal point
of the unit.
For the optimal values found during the optimization of the system have been selected as
optimal set points, that have to be linked during the normal working of the process unit; in
this way the best solution is desirable to be found. So first of all the active constraints have
been controlled with a tight control and in this sense we have the control on “gas oil ratio”,
realized by a ratio control between the “gas oil feed” and the H 2 coming from the
regeneration loop; the inlet pressure given by a pressure controller set on line of the
regenerating mixtures, but no one pressure controller is present on the feed line acting on
the pump, P-100, because the inlet pressure is already set here. The last active constraint,
the temperature out from the regenerating column, has been doubly controlled with an
indicator on line and a controller acting on the coolant flood.
This means reason for a robust control on the active constraints. Then the remaining
degrees of freedom has been set under control with some shrewdness: in particular having
notice that, from the previous search for self-optimizing control, the H2O is influencing the
behavior of the plant, particular attention has been give to this variable and in particular a
composition control (CC) has been put in order to evaluate the mass fraction of water in
regenerating product, and then link to the set-point given by this controller a flood control
has been disposed to adjust that value replacing water lost by evaporating during the
process of regeneration; in order to have some negative values in H 2O make-up line
another flood control has been put in act.

135
FRANCESCO MORANDI SELF-OPTIMIZING CONTROL

Moreover every “external” feed and in this case it is necessary to take into account the “H2
source”, the feed and the make-up water itself, has been controlled by a flood control (FC).
Then it was time for pressures in the different lines, and in particular in the absorption and
in the regenerating columns. The former is controlled by flashing the feed at the entrance
of the column (4th tray), while the latter is performed controlling the acid gas pressure
which is linked to condenser pressure. Another pressure controller has been put before the
“HP separator” to provide the necessary difference of pressure for the flashing operation.
Another issue has been the temperature control: first of all on the reactor, that aims to
avoid some runaway reaction and then on the coolers in order to grant the best optimal
working conditions; and in this sense all heaters/coolers have been put under the control of
a temperature controllers (TC), verifying in parallel the result with other
controllers/indicators. That means reason of a good control on the main self-optimizing
variables too, identified as “gas oil to heat” temperature, its pressure, “H2O make-up”
mass flow, and linked to this the water mass fraction. In order to face in a first
approximation these variables have been selected because of their greater impact on the
disturbances‟ smoothing. It is reasonable to believe that more alternatives could be adopted
and maybe they could be more effective, but in this case with this control good results have
been obtained. Here the main and relevant out coming from the dynamic simulation with
disturbances acting on the feed flow and on its composition, as studied before.
In order to accomplish the result the control structure has been developed in cascade,
adopting a controller on-line, reliable to measure the CV and acting on the relative DOF,
whose set-point is given by the master controller which measure the same variable in a
critical point maybe more susceptible of variance, i.e. in a more sensitive point; this
method has been applied in reactor temperature control, in pressure control (for line
coming from amine loop) and for all the temperature measured (and in particular on the 4 th
tray of both the absorption column and the regenerating column, because it was the most
susceptible of changing) and controlled in the unit process plant.
Moreover we have to note that for simulation the integration time has been reduced in
order to make it possible the integration itself, because the pre-chosen span-time was too
big and it was not possible to achieve a correct simulation.
5.5 RESULTS
As seen previously the control acted on flood has been selected with only proportional
action (P), in this sense it is possible to note the presence of the offset. Same thought is

136
FRANCESCO MORANDI SELF-OPTIMIZING CONTROL

valid for pressure control for which it is avoid the use of integrative action because of the
noise caused by control valves themselves.

Figure 63: valve signal on feed control valve

Figure 64: feed flood control and its disturb with increase in flood value from 60600 [kg/h] to 63600[kg/h]

137
FRANCESCO MORANDI SELF-OPTIMIZING CONTROL

Figure 65: inlet pressure and its control

Figure 66: reactor temperature control (slow dynamics) and its response after the disturbance

138
FRANCESCO MORANDI SELF-OPTIMIZING CONTROL

Figure 67: absorption column pressure control and the disturbance effect

Figure 68: absorption column pressure control with autotuning results

139
FRANCESCO MORANDI SELF-OPTIMIZING CONTROL

Figure 69: desulfurized conditions

Figure 70: level control of the desulfurized charge, with the result after the disturb in feed flood

As it is possible to observe the system is stopped drifting away from its optimal point.
Values are kept almost constant, even if the control it could seem too simple. With some

140
FRANCESCO MORANDI SELF-OPTIMIZING CONTROL

shrewdness there is the possibility to imply a better control maybe using a autotuning
controller. After disturbances in composition too, it is possible to note how the system is
pointing to its normal working point.

Figure 71: composition after disturb in total sulfur content

141
FRANCESCO MORANDI SELF-OPTIMIZING CONTROL

Chapter 6
Conclusions and future perspectives

“The end of this chapter ”


Sonata Arctica
6.1 CONCLUSIONS
At the end of this procedure it is now time to draw the conclusions and trace a general
guideline for next future applications. In this sense it is important to have found it possible
to control in a systematic manner a complex process plant, like a HDS unit; the systematic
of all the method is however based on the economics, and in this way it is important
understand how correct ideas and assumptions are able to grant good result in accordance
with the reality.
During path, ideally recognized in the step-by-step procedure and realistic meant as
passage through the Northern Lands, it has been possible to me understand how simplicity
and accuracy can be the best companions for simulations. All this procedure, even if young
and really complete, found its basis in an incredibly long gestation and its hermeneutics
has been responsible of a great innovation in chemical engineering plantwide control.
Through the overall project, assumptions have been done and suppositions have found
reasons to be introduced, in order to improve the feasibility of the system without, at the
same time, losing in reliability for results obtained.
In this project the procedure above mentioned has found its reason to be applied: even if, it
is important to show, this systematic approach has been seldom applied on chemical plant
(instead of single equipments); positive results have been obtained.
What has been done in this project has been to face a real problem with very modern and
felt close implications, like environmental pollution, and working on a system set up to
solve it partially (we have to remember that refineries at standing conditions are not able to
overcome 15 [ppm wt/wt] sulfur content limit, in this case we can reach the level of 70
[ppm wt/wt], and so this system is reliable if in parallel there are downstream processes to
accomplish the previous request); trying to find its optimal way of work without basing the
control operation on the classical way to operate this, but on the basis of economics

142
FRANCESCO MORANDI SELF-OPTIMIZING CONTROL

building a robust and simple control structure. The model have been manipulated,
transformed and the tested. There did not have been a unique project: a lot of tries have
been studied with different kinetics. After the choice of the model it has been time for
optimization: this part of the project has been developed and performed by Matlab®.
While the writing (meant literally) of the program has been simple, the different
optimization‟s runs have been long and not always correct: there have been tested some
cost functions (changing the different contributes given by the equipment and taking into
consideration different DOFs) in order to give reason to the really important contributes in
this process, trying to find the real key parameters able to influence it either positively or
negatively.
After the optimization there has been the necessity to find the self-optimizing variables,
key role in this procedure, and with the “Nullspace method” it has been possible to verify
what variables have a concrete impact on process‟ operation. Results have given different
possibilities to apply a control structure, and so the easiest one has been chosen. No doubt
there could be more and significantly different choice, but the structure at the basis had to
be controlled in this way.
As mentioned before the project has been “subjected to” different simplifications and
assumptions those have made possible the outcome of this project. The most important
simplifications, in this sense, have dealt with the kinetics and the reaction involved. So it
has been possible to work with a realistic model with such a fastness that could work in
acceptable time. The kinetics itself has demonstrated to be in accordance with practical
results; but it is right to say that, in this sense, all these simplifications can be removed
introducing some more complex models both for kinetics and for the simulation of the
catalyst‟s particle (omitted in this case because the difference in the “order of magnitude”
of the process scale; all of this would be result in a perfect model with at the same time
really heavy calculation because this is a typical scale up problem: catalyst dimensions vs.
plant extension). So a better simulation could introduce longer time to optimize the system
which could be now described more accurately with the problem to have more heavier
calculation load. Following the same guideline, for what concerns the reaction path
involved, for kinetics a single rate has been selected in order to have rapidity and
promptness of the system simulated while for typical sulfur compounds it has been chosen
the main two present in a common gas oil feedstock (thionaphtene and dibenzothiophene).

143
FRANCESCO MORANDI SELF-OPTIMIZING CONTROL

On the same direction we can find the considerations about plant‟s control typology: PID
vs. MPC or RTO. The choice has been made on the previous one because there were the
right conditions to be applied and then because a receiving horizon approach could result
impossible cause the too slowness of the system itself.
There no doubt that an improving effort on calculations (meant as power to solve them)
and maybe a more detailed model with other assumptions on some small shrewdness, that
is why a case study parallel with a real plant could be very useful, can solve this problem
and grant more precision and more reliability.
Standing on calculations it is worth to note that great problems are dealing with the
optimizer used to solve the constrained problem associated with the control procedure: it
has been demonstrated less robust than it was thought. In this sense it could be useful to
introduce a more robust one and then trying to find a more detailed optima, even if the
result obtained in this work are not affected, because we have to consider that the optimum
is always relative in the condition we are working. It is worth to note that it would be better
to have an optimum as nearest as possible to the absolute one (i.e. find one solution more
stable and defined with more accuracy the region of active constraints and their actual
values); a try has been done in order to verify if some changes there could be obtained, and
in this sense it has been selected another criterion of optimization (i.e. the algorithm): the
switch has been done between interior point and active-set, both of fmincon.
The result of this choice it has been an increasing velocity per each iteration step to the
optimal solution (the “main” difference between the two algorithms is the step used to find
the optimum), with a small precision on the nature of the optimum itself (even if a huge
number of iteration admitted, it is reached without a real solution found).
Then during the analysis of this unit it has been possible to suppose some problems tied to
bottleneck conditions and in this sense this is been thought to be when we overpass the
sulfur content limits posed by environmental laws. The model has overcome this point with
an increasing temperature that for the kinetics is feasible, but that in reality could be useful
to verify specially in parallel with a real process data.
Anyway what it is important to note till the first moments the extreme simplicity and
reliability of the procedure described in the previous chapters and followed during all this
analysis of a process. The results highlighted during this analysis, beyond the control itself,
are the schematization and the linearity of the way to control a plant: we started with a

144
FRANCESCO MORANDI SELF-OPTIMIZING CONTROL

procedure and we end with a method. It is a ductile system and in this sense very
accessible.
It is important to understand that this is and remain a complex system in which better
results can be achieved with time and maybe implementing different strategies of control.
In this sense it has to be though that self-optimizing control has shown its best results in
controlling single equipment of a process plant and single unit operation. Some optimal
results have been proved to be reached when the control is applied to single processes and
with more data available 78.
Moreover it is worth to note that this particular approach directed to plantwide control is
today too, subject of lot of publications both in chemical engineering papers 79 and
innovative control strategies of MSc Thesis and PhD thesis too. As example we can
consider that in this date, while I‟m concluding this project Springer is receiving overview
for other publications. There no reason to doubt that this could be the next way to approach
to the chemical process plant.
If we consider that before the last two decades the control was made with some thumb
rules, this systematic procedure is able to improve results that have been obtained
previously and understand better the real effect and importance of a clear and methodic
approach. That means in the very next future the possibility to increase precision and
introduce new strategies of control will be real and in this sense a kind of control
philosophy could be introduced as new way to think and to approach to problems.

78
ANTONIO ARAÚJO, SIGURD SKOGESTAD, Control structure design for the ammonia synthesis process, Computers
and Chemical Engineering 32, 2920–2932, Elsevier Ltd. 2008
79
http://www.scopus.com/, the search with “self-optimizing control” gives as matches 573 results with
already 5 papers only in 2014, while we get the total correspondence in the last five years; 19 th March 2014.

145
FRANCESCO MORANDI SELF-OPTIMIZING CONTROL

Bibliography
Volume:
[1] SIGURD SKOGESTAD, Plantwide Control, Recent Developments and Applications, Wiley
2012
[2] SURINDER PARKASH , Refining Processes Handbook, Gulf Professional Publishing,
2003
[3] DONALD ACKELSON, UOP UNICRACKING PROCESS FOR HYDROCRACKING, in
Robert A. Meyers (ed.), Handbook of Petroleum Refining Processes, 3rd ed. (2003)
[4] MCGRAW-HILL, SYBIL P. PARKER, McGraw-Hill Dictionary of Scientific & Technical
Terms, 6th ed. (2002), Copyright © 2003 by The McGraw-Hill Companies, Inc
[5] JAMES H. GARY, GLENN E. HANDWERK, Petroleum Refining, Institut Français du
Pétrole, 4th ed., CRC Press, 2001
[6] OZREN OCIC, Oil Refineries in the 21st Century: Energy Efficient, Cost Effective,
Environmentally Benign, Wiley 2005
[7] DAVID S. J. JONES, PETER R. PUJADÓ, Handbook of Petroleum Processing, Edited by
DAVID S. J. “STAN”, 2006
[8] GIULIO NATTA, ITALO PASQUON, Pricipi della Chimica Industriale, Milano: Tamburini,
1966
[9] ROBERT H. PERRY, DOW W. GREEN, Perry's Chemical Engineers' Handbook Seventh
Edition, 7th ed., McGraw-Hill, 1999
[10] RUNE PRESTVIK , KJELL MOLJORD , KNUT GRANDE , AND ANDERS HOLMEN, Catalytic
Naphtha Reforming, Revised and Expanded, Edited by George J . Antos and Abdullah M .
Aitani, CRC Press 2004
[11] ROBERT A. MEYERS, Handbook of Petroleum Refining Processes, McGraw-Hill, 3rd
ed. (2003)
[12] LEE, S., Encyclopedia of Chemical Processing, Taylor & Francis, 2006
[13] MAPLES, R.E., Petroleum Refinery Process Economics, PennWell Corporation, 2nd ed.,
2000
[14] James H. Gary, Glenn E. Handwerk, Petroleum Refining, Technology and Economics,
Marcel Dekker, Inc., 4th ed., 2001

146
FRANCESCO MORANDI SELF-OPTIMIZING CONTROL

[15] ANCHEYTA, JORGE, Modeling and simulation of catalytic reactors for petroleum
refining, John Wiley & Sons, Inc., Hoboken, New Jerse, 1st ed., 2011
[16] S.S.E.H. ELNASHAIE AND S.S. ELSHISHINI, Modelling, simulation, and optimization of
industrial fixed bed catalytic reactors, Yverdon, Switzerland ; Langhorne, Pa. : Gordon
and Breach Science Publishers, 1993
[17] LEVENSPIEL, O., Chemical reaction engineering, John Wiley & Sons, Inc., 3rd ed.
1972
[18] WILLIAM L. LUYBEN, Chemical Reactor, Design and Control, John Wiley & Sons,
Inc., 2007
[19] HEINZ A PREISIG, The ABC of Process Modelling, Lecture Notes, Version: 1.01, 20-
08-2012
[20] NOCEDAL, JORGE; WRIGHT, STEPHEN J., Numerical Optimization, Berlin, New York:
Springer-Verlag, 2nd ed. 2006
[21] PIO FORZATTI, LUCA LIETTI, Strumentazione chimica industriale, Volume 1, Milano:
CUSL, 2003

Paper:
[22] GILBERT F. FROMENT, GUY A. DEPAUW, AND VALÉRIE VANRYSSELBERGHE, Kinetic
Modeling and Reactor Simulation in Hydrodesulfurization of Oil Fractions, Amercan
Chemical Society 1994
[23] AYSAR T. JARULLAH, IQBAL M. MUJTABA, ALASTAIR S. WOOD, Kinetic parameter
estimation and simulation of trickle-bed reactor for hydrodesulfurization of crude oil,
Chemical Engineering Science 201, vol. 66, pp. 859-871, 2011
[24] LUKE ADDINGTON, CHRIS NESS, An Evaluation of General “Rules of Thumb” in
Amine Sweetening Unit Design and Operation, GPA 89th Annual Convention March 21-24,
2010
[25] D.TSAMATSOULIS, N. PAPAYANNAKOS, Investigation of intrinsic
hydrodesulphurization of a VGO in a trickle bed reactor with backmixing effects, Chemical
Engineering Science, Vol. 53, No. 19, pp. 3449-3458, 1998
[26] FOSS, A. S., Critique of Chemical Process Control Theory, IEEE Transactions on
Automatic Control, AC-18(6), 646-652, 1973

147
FRANCESCO MORANDI SELF-OPTIMIZING CONTROL

[27] SIGURD SKOGESTAD, Control structure design for complete chemical plants,
Computers & Chemical Engineering, Volume 28, Issues 1–2, 15 January 2004, Pages 219-
234
[28] HADDAD, WASSIM M., BERNSTEIN, DENNIS S., NETT, CARL N., Decentralized H2/H∞
controller design: The discrete-time case, Proceedings of the IEEE Conference on
Decision and Control, 2, pp. 932-933, 1989
[29] PAGE S. BUCKLEY, Techniques of Process Control, Wiley, New York (1964)
[30] TRULS LARSSON, SIGURD SKOGESTAD, Plantwide control - A review and a new design
procedure, Modeling, Identification and Control, 21 (4), pp 209 – 240 (July 1, 2000)
[31] JAMES MERRIL DOUGLAS, Conceptual Design of Chemical Processes, New York:
McGraw-Hill (1988)
[32] J.J. DOWNS, Distillation Control in a Plantwide Control Environment, In: Practical
Distillation Control, W.L Luyben (ed.), Van Nostrand Reinhold, 413-439 (1992)
[33] L. T. NARRAWAY, J. D. PERKINS, Selection of process control structures based on
Economics, Comp. Chem. Eng. 18, S511, (1993a)
[34] KOOKOS, I.K., PERKINS, J.D., An algorithmic method for the selection of
multivariable process control structures, Journal of Process Control, 12 (1) , pp. 85-99,
2002
[35] ALSTAD, V., SKOGESTAD, S., HORI, E.S., Optimal measurement combinations as
controlled variables, Journal of Process Control, 19 (1), pp. 138-148 (2009)
[36] PRICE, R.M., GEORGAKIS, C., Plantwide regulatory control design procedure using a
tiered framework, Industrial and Engineering Chemistry Research, 32 (11), pp. 2693-2705
(1993)
[37] E.M.B. ASKE AND S. SKOGESTAD, “Consistent inventory control”, Ind.Eng.Chem.Res,
48 (44), 10892-10902 (2009)
[38] SATTERFIELD, C. N., Trickle-bed reactors, AIChE journal [0001-1541], year:1975
volume: 21, issue:2, pages: 209 -228
[39] UNITED STATES ENVIRONMENTAL PROTECTION AGENCY, OFFICE OF ENVIRONMENTAL
INFORMATION, Emergency Planning and Community Right-to-Know Act – Section 313:
Guidance for Reporting Toxic Chemicals: Polycyclic Aromatic Compounds Category, EPA
260-B-01-03, Washington, DC, August 2001

148
FRANCESCO MORANDI SELF-OPTIMIZING CONTROL

[40] M. A. RODRÍGUEZ AND J. ANCHEYTA, Modeling of Hydrodesulfurization (HDS),


Hydrodenitrogenation (HDN), and the Hydrogenation of Aromatics (HDA) in a Vacuum
Gas Oil Hydrotreater, Energy & Fuels 2004, 18, 789-79, 2004
[41] KIM G. KNUDSEN, BARRY H. COOPER, HENRIK TOPSØE, Catalyst and process
technologies for ultra low sulfur diesel, Applied Catalysis A: General 189 (1999) 205–215,
Elsevier Science B.V 1999
[42] G.D. BELLOS, N.G. PAPAYANNAKOS, The use of a three phase microreactor to
investigate HDS kinetics, Catalysis Today 79–80 (2003) 349–355, Elsevier Science B.V.
2003
[43] MIGUEL A. RODRÍGUEZ, IGNACIO ELIZALDE, JORGE ANCHEYTA, Comparison of kinetic
and reactor models to simulate a trickle-bed bench-scale reactor for hydrodesulfurization
of VGO, Fuel 100 (2012) 91–99, Elsevier Ltd., 2012
[44] G.D. BELLOS, L.E. KALLINIKOS, C.E. GOUNARIS, N.G. PAPAYANNAKOS, Modelling of
the performance of industrial HDS reactors using a hybrid neural network approach,
Chemical Engineering and Processing 44 (2005) 505–515, Elsevier B.V. 2004
[45] BANDAR ALSOLAMI, JOANA T. CARNEIRO, JACOB A. MOULIJN, MICHIEL MAKKEE, On-
site low-pressure diesel HDS for fuel cell applications: Deepening the sulfur content to ≤1
ppm, Fuel 90 (2011) 3021–3027, Elsevier Ltd. 2011
[46] BJØRN DONNIS, RASMUS GOTTSCHALCK EGEBERG, PEDER BLOM, KIM GRØN KNUDSEN,
Hydroprocessing of Bio-Oils and Oxygenates to Hydrocarbons. Understanding the
Reaction Routes, Top Catal (2009) 52:229–240, Springer Science+Business Media, LLC
2009
[47] BARBARA PAWELEC, RUFINO M. NAVARRO, JOSE´ MIGUEL CAMPOS-MARTIN AND

JOSE´ L. G. FIERRO, Towards near zero-sulfur liquid fuels: a perspective review, Catal. Sci.
Technol., 2011, 1, 23–42, The Royal Society of Chemistry 2011
[48] RASMUS G. EGEBERG, NIELS H. MICHAELSEN, LARS SKYUM, Novel hydrotreating
technology for production of green diesel, Haldor Topsøe
[49] ELENA GÓMEZ SAYALERO, SIMPLIFIED MODELING AND VALIDATION OF AN
INDUSTRIAL DIESEL HYDRODESULFURIZATION PLANT, Dpt. of System Engineering
and Automatic Control University of Valladolid, Spain, July 2009
[50] A.S. FERREIRA, M.C. NICOLETTI, J.R. BERTINI JR., R.C. GIORDANO, Methodology for
inferring kinetic parameters of diesel oil HDS reactions based on scarce experimental data,
Computers and Chemical Engineering 48 (2013) 58– 73, Elsevier Ltd., 2012

149
FRANCESCO MORANDI SELF-OPTIMIZING CONTROL

[51] HASAN SILDIR, YAMAN ARKUNA, BERNA CAKALB, DILA GOKCEB, EMRE KUZUB, A
dynamic non-isothermal model for a hydrocracking reactor: Model development by the
method of continuous lumping and application to an industrial unit, Journal of Process
Control xxx (2012) xxx– xxx, Elsevier Ltd., 2012
[52] FABIAN S. MEDEROSA, JORGE ANCHEYTA, IGNACIO ELIZALDE, Dynamic modeling and
simulation of hydrotreating of gas oil obtained from heavy crude oil, Applied Catalysis A:
General 425– 426 (2012) 13– 27, Elsevier B.V., 2012
[53] PEYMAN MAHINSA, MOHAMMAD TAGHI SADEGHI, HAMID GANJI, SAEED SHOKRI, Two
Dimensional Dynamic Modeling of Hydrodesulphurization Reactor, Journal of Petroleum
Science Research (JPSR), Vol. 1, Issue. 2, PP. 32-35, Jul. 2012
[54] FABIÁN S. MEDEROS, JORGE ANCHEYTA, Mathematical modeling and simulation of
hydrotreating reactors: Cocurrent versus countercurrent operations, Applied Catalysis A:
General 332, 8–21, Elsevier B.V., 2007
[55] KNUT WIIG MATHISEN, Integrated Design and Control of Heat Exchanger Network,
University of Trondheim, The Norwegian Institute of Technology, Submitted April, 1994
[56] ANTONIO ARAÚJO, SIGURD SKOGESTAD, Control structure design for the ammonia
synthesis process, Computers and Chemical Engineering 32, 2920–2932, Elsevier Ltd.
2008
[57] JAMES J. DOWNS, SIGURD SKOGESTAD, An industrial and academic perspective on
plantwide control, Annual Reviews in Control 35 (2011) 99–110, Elsevier Ltd. 2011
[58] LUCA LIETTI, Material of “Catalisi e Processi Catalitici – 12 – Reattori Industriali”,
2011-2012
[59] G. C. A. SCHUIT, B. C. GATES, Chemistry and Engineering of Catalytic
Hydrodesulfurization, AlChE Journal (Vol. 19, No. 3) May, 1973
[60] LARS ERIK ØI, Aspen HYSYS Simulation of CO2 Removal by Amine Absorption from a
Gas Based Power Plant, SIMS2007 Conference, Gøteborg, October 30-31, 2007
[61] B. BALOOCHY, S.SHOKRI, M.AHMADI MARVAST, Design and Implementation of a
Simulator for the Hydro Treating Reactor in RTO Development, International Journal of
Chemical Engineering and Applications, Vol. 1, No. 4, December 2010
[62] GAEL D. ULRICH, PALLIGARNAI T. VASUDEVAN, How to Estimate Utility Costs,
CHEMICAL ENGINEERING, WWW.CHE.COM, APRIL 2006

150
FRANCESCO MORANDI SELF-OPTIMIZING CONTROL

[63] R.M. COTTA, M.R. WOLF-MACIEL, R. MACIEL FILHO, A cape of HDT industrial
reactor for middle distillates, Computers and Chemical Engineering 24, 173 I- 1735,
Elsevier Science Ltd., 2000
[64] E. GÓMEZ, C. DE PRADA, D. SARABIA, C. A. MÉNDEZ, S. CRISTEA, JOSE M. SOLA, E.
UNZUETA, Grey-box modelling of an industrial hydrodesulphurization process, 18th
European Symposium on Computer Aided Process Engineering – ESCAPE 18 Bertrand
Braunschweig and Xavier Joulia (Editors), Elsevier B.V./Ltd. 2008
[65] KAZUO MATSUDA, KENICHI KAWAZUISHI, YOSHIICHI HIROCHI, REI SATO, YASUKI
KANSHA, CHIHIRO FUSHIMI, YUTAKA SHIKATANI, HIROSHI KUNIKIYO, ATSUSHI TSUTSUMI,
Advanced energy saving in the reaction section of the hydro-desulfurization process with
self-heat recuperation technology, Applied Thermal Engineering 30, 2300 - 2305, Elsevier
Ltd., 2010
[66] ATSUSHI ISHIHARA, FRANCK DUMEIGNIL, JEAYOUNG LEE, KOUHEI MITSUHASHI, EIKA
W. QIAN, TOSHIAKI KABE, Hydrodesulfurization of sulfur-containing polyaromatic
compounds in light gas oil using noble metal catalysts, Applied Catalysis A: General 289,
163–173, Elsevier B.V., 2005
[67] SON-KI IHM, SANG-JIN MOON, AND HYUNG-JOON CHOI, Hydrodesulfurization of
Thiophene over CoMo, NiMo, and NiW/A203, Catalysts: Kinetics and Adsorption, Ind. Eng.
Chem. Res.,29, 1147-1152, 1990
[68] IGNACE A. VAN PARLJS, LUCLEN H. HOSTEN, AND GILBERT F. FROMENT, Kinetics of
Hydrodesulfurization on a CoMo/-Al2O3 Catalyst, Kinetics of the Hydrogenolysis of
Benzothiophene, Ind. Eng. Chem. Prod. Res. Dev., 25, 437-443, 1986
[69] KH. MOHAMADBIGY, M. BAZMI, R. BEHRADI, R. BINESH, AMINE ABSORPTION
COLUMN DESIGN USING MASS TRANSFER RATE SIMULATION, Petroleum & Coal
47 (2), 39-46, 2005
[70] HUAMIN WANG, ROEL PRINS, HDS of benzothiophene and dihydrobenzothiophene
over sulfided Mo/-Al2O3, Applied Catalysis A: General 350, 191–196, Elsevier B.V., 2008
[71] SYLVETTE BRUNET, DAMIEN MEY, GUY PÉROT, CHRISTOPHE BOUCHY, FABRICE DIEHL,
On the hydrodesulfurization of FCC gasoline: a review, Applied Catalysis A: General 278,
143–172, Elsevier B.V., 2005
[72] S. A. ALI, Thermodynamics of Hydroprocessing Reactions. Chapter 4 in
Hydroprocessing of Heavy Oils and Residue, J. Ancheyta and J. Speight (Eds.), CRC Press,
N.Y., USA, pp. 51-70, 2007

151
FRANCESCO MORANDI SELF-OPTIMIZING CONTROL

[73] M. BHASKAR, G. VALAVARASU, B. SAIRAM, K. S. BALARAMAN, K. BALU, Three-


Phase Reactor Model to Simulate the Performance of Pilot-Plant and Industrial Trickle-
Bed Reactors Sustaining Hydrotreating Reactions, Ind. Eng. Chem. Res., 43, 6654-6669,
2004
[74] SAEID SHOKRI, MAHDI AHMADI MARVAST, MORTEZATAJERIAN, PRODUCTION OF
ULTRA LOW SULFUR DIESEL: SIMULATION AND SOFTWARE DEVELOPMENT,
Petroleum & Coal 49 (2), 48-59, 2007
[75] FAWZI M. ELFGHI, N. A. S. AMIN, Parametric Study of Hydrodesulfurization and
Hydrodearomatization of Gasoil in Hydrotreating Process over CoMo – S Catalyst using a
Pilot Plant Integral Reactor, Journal Teknologi, 56 (Sains & Kej.) Keluaran Khas (1), Dis.
2011: 53–73
[76] MICHEL CAMES, ECKARD HELMERS, Critical evaluation of the European diesel car
boom - global comparison, environmental effects and various national strategies,
Environmental Sciences Europe, 25:15, Springer 2013
[77] KIM G. KNUDSEN AND BARRY H. COOPER, Ultra Deep Desulfurization of Diesel: How
an understanding of the underlying kinetics can reduce investment costs, Practical
Advances in Petroleum Processing, pp 297-316, Springer 2006
[78] V. ALSTAD AND S. SKOGESTAD, Null Space Method for Selecting Optimal
Measurement Combinations as Controlled Variables, Ind.Eng.Chem.Res, 46 (3), 846-853
(2007)
[79] KARIWALA, V., Optimal measurement combination for local self-optimizing control,
Industrial and Engineering Chemistry Research, 46 (11), pp. 3629-3634 (2007)
[80] EUROPEAN COMMISSION, Consumer Prices of Petroleum Products (inclusive of duties
and taxes) – Automotive Gas Oil (Diesel Oil), prices at: 11-02-2013, Eurogeographic for
the administrative boundaries

Site:
[81] ENCYCLOPÆDIA BRITANNICA ONLINE, Thiol,
www.britannica.com/EBchecked/topic/592252/thiol, accessed on February 2014
[82] ENCYCLOPÆDIA BRITANNICA ONLINE, Sulfide,
www.britannica.com/EBchecked/topic/572504/sulfide, accessed on February 2014
[83] ENCYCLOPÆDIA BRITANNICA ONLINE, Thiophene,

152
FRANCESCO MORANDI SELF-OPTIMIZING CONTROL

www.britannica.com/EBchecked/topic/592288/thiophene, accessed on February 2014


[84] ENCYCLOPÆDIA BRITANNICA ONLINE, Pyrrole,
www.britannica.com/EBchecked/topic/485130/pyrrole, accessed on February 2014
[85] ENCYCLOPÆDIA BRITANNICA ONLINE, Pyridine,
www.britannica.com/EBchecked/topic/484880/pyridine, accessed on February 2014
[86] ENCYCLOPÆDIA BRITANNICA ONLINE, Quinoline,
www.britannica.com/EBchecked/topic/487446/quinoline, accessed on February 2014
[87] ENCYCLOPÆDIA BRITANNICA ONLINE, Catalyst,
www.britannica.com/EBchecked/topic/99128/catalyst, accessed on February 2014
[88] ENCYCLOPÆDIA BRITANNICA ONLINE, Friedrich Bergius,
www.britannica.com/EBchecked/topic/61797/Friedrich-Bergius, February 2014
[89] THE MERCK INDEX ONLINE, Benzothiophene,
www.rsc.org/Merck-Index/monograph/mono1500009455/thianaphthene?q=authorize,
accessed on February 2014
[90] EUROPEAN COMMISSION,
http://ec.europa.eu/environment/air/quality/legislation/existing_leg.htm, seen on March
2013
[91] NATIONAL SERVICE CENTER FOR ENVIRONMENTAL PUBLICATIONS,
http://nepis.epa.gov/Exe/ZyNET.exe/900B0H00.TXT?ZyActionD=ZyDocument&Client=
EPA&Index=2000+Thru+2005&Docs=&Query=&Time=&EndTime=&SearchMethod=1
&TocRestrict=n&Toc=&TocEntry=&QField=&QFieldYear=&QFieldMonth=&QFieldDa
y=&IntQFieldOp=0&ExtQFieldOp=0&XmlQuery=&File=D%3A\zyfiles\Index%20Data\0
0thru05\Txt\00000011\900B0H00.txt&User=ANONYMOUS&Password=anonymous&So
rtMethod=h|-
&MaximumDocuments=1&FuzzyDegree=0&ImageQuality=r75g8/r75g8/x150y150g16/i4
25&Display=p|f&DefSeekPage=x&SearchBack=ZyActionL&Back=ZyActionS&BackDes
c=Results%20page&MaximumPages=1&ZyEntry=1&SeekPage=x&ZyPURL, accessed on
March 2013
[92] WIKIPEDIA, THE FREE ENCYCLOPEDIA, Halide, http://en.wikipedia.org/wiki/Halide,
accessed on February 2014
[93] TRECCANI.IT, L‟ENCICLOPEDIA ITALIANA, VOLUME II - RAFFINAZIONE E
PETROLCHIMICA,

153
FRANCESCO MORANDI SELF-OPTIMIZING CONTROL

www.treccani.it/enciclopedia/enciclopedia-degli-idrocarburi_%28altro%29/, accessed on
March 2013
[94] FIAT, Common Rail, http://www.fiat.it/fiatpedia/glossario/common-rail, accessed on
March 2014
[95] IARC, IARC: DIESEL ENGINE EXHAUST CARCINOGENIC,
http://www.iarc.fr/en/media-centre/pr/2012/pdfs/pr213_E.pdf, accessed on March 2014
[96] EUROPEAN COMMISSION, Verso standard più rigidi per i carburanti,
http://ec.europa.eu/italia/attualita/archivio/trasporti_energia/11078bab270_it.htm
[97] EUROPE‟S ENERGY PORTAL, Fuel Price, http://www.energy.eu/fuelprices/, accessed on
February 2013
[98] Australian Government, Price Schedule – Industrial Gases 49009, accessed on
February 2013
[99] MATHWORKS, fmincon, http://www.mathworks.it/it/help/optim/ug/fmincon.html,
accessed on March 2014
[100] MATHWORKS, null, http://www.mathworks.it/it/help/matlab/ref/null.html, accessed
on March 2014

Guide:
[101] HYPROTECH LTD, HYSYS 3.0, 2002
[102] ASPENTECH, Advanced Process Modeling Using Aspen HYSYS®, AspenTech 2007
[103] ASPENTECH, Advanced Process Modeling Using Aspen HYSYS®, AspenTech 2008
[104] ASPENTECH, Aspen HYSYS® Customization Guide, AspenTech October 2007
[105] ASPENTECH, Aspen HYSYS® Customization Guide, AspenTech, Version Number:
7.3 March 2011
[106] ASPENTECH, HYSYS®2004.2 Tutorials & Applications, AspenTech, October 2005
[107] ASPENTECH, HYSYS® 3.2 Simulation Basis, Hyprotech 2003
[108] ASPENTECH, HYSYS® 3.2 User Guide, Hyprotech 2003

154

You might also like