Index

1
Index

• 1. UPC and my Thesis work presentation
• 2. Complex distillation columns with energy
  savings
• 3. The work
• 3.1 Design
• 3.2 Dynamic aspects
• 3.3 Control
• 4. Conclusions and future work

2
Universitat Politècnica de Catalunya (UPC).

• Founded in 1971, it has
• 9 schools and faculties (Industrial Engineering)
• 8 technical colleges
• 7 associate schools
• 38 departments (Chemical Engineering)
• 21 diplomas, 8 degrees 30.000 students last year
• 44 Ph.D. programs 149 thesis during 1996-1997
• budget 1998 260,00 Mcan

3
The Chemical Engineering Department

• 90 teachers and researchers
• 95 Ph.D. students
• Main goals
• chemical process optimisation, security and
  accident modelisation, reactors, water
  technology, fluid-particle systems, alimentary
  technology, waste treatment, contaminants
  analysis, environmental studies, molecular
  engineering, polymer synthesis and structure.

4
The thesis work

• Title Energy optimisation in complex
  distillation columns
• Objective study complex designs for energy
  savings already described to bring them closer to
  implementation
• design, operation and control
• Status
• Petlyuk Column centre of my studies till now
• some design, some control, some operation
• 60 of work done

5
The Petlyuk Column origin

• Wright (1949) proposed a promising design
  alternative for separating ternary mixtures
• Petlyuk (1965) studied the scheme theoretically
• Most important literature since Petlyuk
  Fidkowski and Krolikowski / Glinos and Malone /
  Triantafyllou and Smith / Kaibel / Wolf and
  Skogestad

6
The Petlyuk Column structure
7
Conventional designs
INDIRECT TRAIN
DIRECT TRAIN
8
Distillation process in a Petlyuk Column
9
Petlyuk Column features

• No more than one component is stripped out in
  each section, key components A and C
• reversibility during mixing of streams in feed
  location (pinch zone)
• no remixing effect
• Thermal coupling
• no thermodynamic losses in heat exchanges of
  prefractionator reboiler and condenser
• reversibility during mixing of streams at ends of
  columns

Reported 30 of energy savings
10
The Divided Wall Column
Thermodynamical equivalence in only one shell
11
Extension to other multicomponent distillations
A
B
A B C D
C
D
12
Distinguishing features

• n(n-1) sections required for an n-component
  separation
• Only one condenser and one reboiler
• Key components in each column are not two
  adjacent ones, but the ones with extreme
  volatility

13
Design of the Petlyuk Column
Work presented at AIChE Meeting, Los Angeles, 1997

• Degrees of freedom
• design number of trays per section and feed
  trays
• operation flowrates or flowrate ratios. Two
  extra DOF used to optimise the process
• Main design decision separation to be carried
  out by the prefractionator.
• Two levels of specification
• two specified variables
• three specified variables

14
Short-cut methods facing multicomponent systems
Most of numerical correlations used by short-cut
methods solve distillation columns based on
required recoveries of just key components
Ability to play only with two recoveries
Importance of all three prefractionator
recoveries over the global economic performance
of a complex distillation column
15
Proposed design heuristic method
Balance between prefractionator and main column
and between upper and down main column

• Decision of A and C recoveries. Design following
  short-cut indications (simplified model).
  Rigorous simulations.
• Change of feed tray to minimise the larger vapour
  flow between flows at COL2 bottom and COL3 top
• Repeat till vapour flows are equal
• Change recoveries of A and C

16
Simplified model of the Petlyuk Column
Work presented at Congreso Mediterraneo de
Ingenieria Quimica, 1996
17
Determination of mixtures that take major profit
of the Petlyuk Column

• Case study with pro-II simulations
• Studied separations
• different quantities of B in feed (33, 33,
  -33)
• different Easy Separation Index (lt1, 1, gt1)
• Savings compared to the best train of columns
• more B in feed, more savings (23, 20 , 14)
• more savings when ESI is close to 1 (34)

18
Dynamic behaviour

• SPEEDUP model
• Neural Network simulation
• MATLAB model
• linearised model transfer functions
• Model approximations
• constant relative volatility throughout the
  column, equimolar overflow, no heat losses
  equilibrium in each plate, constant pressure,
  liquid and vapour flow dynamics, tray
  hydraulics...

19
Dynamic features

• Interaction
• Speed, magnitude and shape of response stiff

20
Neural Network simulation - MPC?
Work presented at III Congresso de Redes
Neuronais, 1997

• The used NN
• three layer
• feedforward with autoregressive neurones
  connected to the output
• Sampling frequency from lowest time constant of
  all outputs C in feed to B in sidestream, 6 min
• Training of the NN
• PRBS signal applied to all inputs (until 3
  manipulated variables and 3 disturbances)

21
NN forecasting example
902 patterns 20000 epochs 3, 6, 1 neurons Sigm.,
linear shift param. 1 autoregressive param. 1
22
(No Transcript)
23
Control problem

• Control product compositions
• 3 composition specifications (holes in some
  operation regions)
• inventory control
• Control to minimise energy consumption
• Robustness?
• Linearity far from nominal steady state?
• Disturbances rejection and set point changes
  achievement?

24
Descentralised control
Work presented at CHISA 98

• Skogestad acceptable control seems feasible (no
  energy control, linear model)
• Study of descentralised control with MATLAB
  models
• Tyreus method
• Design and test inventory control
• 7 control valves - 5 steady state DOF 2
  inventory loops
• Design composition control
• Design optimisation control (energy minimisation)

25
Diagonal control for the Petlyuk Column

• Control of A, B, and C purity
• For each inventory control (D-B, L-B, D-B)
• Transfer function
• MRI, CN, Intersivity Index
• For the decided control structure D,B L, S, V
• Chose one pairing
• For the decided pairing L-A, S-B, V-C
• BLT tuning procedure
• controller gains 0.74, -2.33, 0.65
• controller reset times 14.16 for all loops

26
(L-A, S-B, V-C) Controlled system MATLAB
simulation
Set point change in A purity example
No instability problem was found, better tunning
can be achieved
27
MIMO feedback control

• Controllability analysis in frequency domain
• bandwidth
• RGA, CN, singular values
• stability (Nyquist plots)
• poles and zeros
• MIMO robustness

28
Self-optimising control
Work to be presented at PRES, 1999

• Published works from NTNU
• Problem once the minimum is located, control is
  required to keep the operating point at the
  minimum when disturbances are loaded
• Solution Improve robustness with feedback
  control to careful selected outputs
• Require measurable output variable which when
  kept constant keeps minimum energy consumption
  (self-optimising control)

29
Studied controlled variables for indirect energy
minimisation

• For each candidate, sensitivity to disturbances
  in feed composition and liquid fraction is
  computed
• heavy key fraction in vapour leaving top of
  prefractionator
• middle component recovery in prefractionator
• main column flow balance
• Temperature profile symmetry
• others
• The best?

30
Conclusions

• A design method
• Mixture characterisation for Petlyuk Column
• Dynamic features
• NN are able to simulate the Petlyuk Column
• Diagonal control works in our simplified model
• Self-optimising control fits the Petlyuk Column

31
Future work

• Better characterisation of mixtures fitting
  different complex distillation columns
• Other designs to compare with. Energy integration
• Robustness for different nominal steady-states
• HYSYS dynamic rigorous simulations
• Design and control together
• NN simulation into Model Predictive Control