This document summarizes Vida Meidanshahi's PhD thesis on advances in design, optimization, and control of semicontinuous processes. The research objectives were to reduce operating costs and expand the economical production range of semicontinuous distillation. A novel semicontinuous without middle vessel configuration was proposed that reduced total direct costs by 42% and operating costs by 45%. Mixed integer dynamic optimization was used to design the process, reducing total annualized costs by 43%. Finally, model predictive control was developed and reduced cycle time by 10% and operating costs by 11% compared to conventional control.

1. Advances in Design, Optimization and
Control of Semicontinuous Processes
By:
Vida Meidanshahi
September 23rd, 2016
Supervisor: Dr. Thomas Adams II
Committee Members: Dr. Prashant Mhaskar
Dr. Timothy Davidson
Dr. Li Xi (Internal Examiner)
Chair: Dr. Gary Bone

2. What is a semicontinuous distillation?!
2
Conventional continuous Semicontinuous
B: Benzene
T: Toluene
X: o-Xylene

3. How does a semicontinuous system
work?!
3
BTX
T
B
X
V1
V2
MV
Light component
Intermediate component
Heavy component

4. 4
BTX
T
B
X
V1
V2
MV
How does a semicontinuous system
work?!
Light component
Intermediate component
Heavy component

5. Why semicontinuous distillation?
5
Total Annualized Cost TAC = + Annual Operating Cost
Total Direct Cost
Payback Period
Desirable separation technique for low to intermediate processes such as:
 Bio-fuel
 Pharmaceutical
Proven by 16 years of
research in the area!

6. Research objectives
 Reduce the operating cost of the process.
 Reduce the TAC  Expand the economical region.
6
Lower
product cost

7. Research objectives
7
Achieve the objectives by:
1) Investigating a new design.
2) Optimizing the process.
3) Applying advanced control system.

8. Design
Semicontinuous without middle vessel
8

9. Semicontinuous without middle vessel
(SwoMV) configuration
9

10. SwoMV configuration
10
Producing Mode

11. Economic analysis results
11Feed: 0.1/0.8/0.1 of BTX.
0 20 40 60 80 100
600
1000
1400
1800
2200
TotalDirectCost($1000)
Toluene Production Rate (Mmol/yr)
Benzene Production Rate (Mmol/yr)
Conv. Continuous
CSC
SwoMV
Side Stream
0 20 40 60 80 100
0
50
100
150
200
OperatingCost($1000/yr)
Toluene Production Rate (Mmol/yr)
Benzene Production Rate (Mmol/yr)
Conv. Continuous
CSC
SwoMV
Side Stream
42% less total
direct cost than SC
45% less operating
cost than CSC
0 20 40 60 80 100
200
300
400
500
600
700
800
900
Benzene Production Rate (Mmol/yr)
TotalAnnualisedCost($1000/yr)
Toluene Production Rate (Mmol/yr)
Conv. Continuous
CSC
SwoMV
Side Stream
expanded the
economical
production range by
44.5% over SC
43% less total
annualized cost
than CSC

12. Contributions
 Proposed a novel process intensification technique.
 It can purify a ternary mixture in one distillation column.
 No tank  Lower total direct costs (e.g. -42%).
 No charging and discharging modes.
 Lower TAC for the whole range of feed compositions (e.g. -42%).
 Lower operating costs for a range of feed compositions (e.g. -45%).
 Expand the economical production range of semicontinuous (e.g. -44.5%).
 Facilitates retrofit of available distillation columns for ternary purification.
 Lowers the production costs of speciality chemicals potentially!!
12Space-Constrained Purification of Dimethyl Ether through Process Intensification using Semicontinuous Dividing Wall Columns.
Sarah E. Ballinger, Thomas A. Adams, 2016

13. Integrated Design
&
Control
Mixed integer dynamic optimization
13

14. Motivation
• No formal design procedure was available for semicontinuous process.
• Higher operating cost of the semicontinuous process can be due to non-optimal
designs.
14
Heuristicdesignrules
Trial-and-error procedure
Sequential approach
Time consuming method
Does not guarantee
optimality/sub-optimality
Proposeddesign
methodology
Systematic method
Simultaneous method
Much faster near-optimal
designs.
Guarantee local optimality

15. Simulation environment
Modeled in gPROMS V4:
1) PML  Convenient model setup
2) Imbedded MIDO optimization
toolbox
3) Faster simulation time 
Reasonable PSO computational
time
4) Easy link with Matlab for control
studies
15
Aspen Plus Dynamics is not suitable for mixed integer dynamic
optimization of semicontinuous system.
1) Can’t change structural parameters.
2) Computational time is high.
Integrated
design and
control
Optimization of
semicontinuous
system
MIDO

16. MIDO methodology
Decision Variables:
1) Number of trays
2) Location of feed and side trays
3) Middle vessel hold up
4) Tuning parameters of PI controllers
4 Integer variables:
− CTs
− Middle vessel height
9 Continuous variables:
− Gains and reset times of controllers.
− Cycle time.
16

17. Optimization problem
Objective function:
Subject to:
 MESH equations
 Distillate purity
 Bottoms purity
 Middle vessel purity at the end of cycle
 Condenser and reboiler liquid levels
17
𝑀𝑖𝑛:
𝑇𝑜𝑡𝑎𝑙 𝐷𝑖𝑟𝑒𝑐𝑡 𝐶𝑜𝑠𝑡
𝑃𝑎𝑦𝑏𝑎𝑐𝑘 𝑃𝑒𝑟𝑖𝑜𝑑
+ 𝐴𝑛𝑛𝑢𝑎𝑙 𝑂𝑝𝑒𝑟𝑎𝑡𝑖𝑛𝑔 𝐶𝑜𝑠𝑡 𝐴𝑛𝑛𝑢𝑎𝑙 𝑝𝑟𝑜𝑐𝑒𝑠𝑠𝑖𝑛𝑔 𝑟𝑎𝑡𝑒 𝑜𝑓 𝑓𝑒𝑒𝑑
Total Annualized Cost (TAC)

18. MIDO results
MIDO solved by:
1) Gradient-based outer approximation method (OA)
2) Gradient-free particle swarm optimization (PSO)
18
Almost 35%
improvement
in TAC
Almost 33%
improvement
in OC

19. Extended to a superstructure case
19

20. Contributions
 System is simulated in gPROMS for the first time.
 For the first time, a methodology is presented to simultaneously
design the structural and operational parameters.
 A methodology is presented to design a semicontinuous system for
any arbitrary non-azeotrpoic mixture.
 The methodology can find locally-optimum designs with less
computational effort.
 The obtained design has lower operating cost and TAC (e.g. -33%).
 Performance PSO and OA methods are compared for a highly
nonlinear process and superiority of each one is explained.
 The computational time of OA method is 5-20 CPU-min where it is
900-1500 CPU-min for PSO method.
20

21. Advance Control System
Model predictive control
21

22. Motivation & objective
Objective: Minimize the operating cost of the
process while maintaining the product purities. 22
Advance
control
Consider
dynamics of
the process
Change the
operational
policy
Improve the
economics
Linear PI controllers have been used
in this highly nonlinear process.

23. Model predictive control (MPC)
23
Direct
configuration
Cascaded MPC with PI
configuration

24. Subspace model identification
24
Inputs:
- Distillate composition controller setpoint
- Bottom composition controller setpoint
- Side stream flow controller setpoint
- Reflux drum level controller setpoint
- Sump liquid level controller setpoint
Quality variables:
- Middle vessel purity
- Average distillate purity
- Average bottom purity
- Operating cost
Outputs:
- Reflux drum liquid level
- Sump liquid level
- Vapour velocities at top and bottom of the column
𝒙(𝑘 + 1) = 𝑨 𝒙(𝑘) + 𝑩 𝒖(𝑘) (7)
𝑦 𝑘 = 𝑪 𝒙 𝑘 + 𝑫 𝒖 𝑘 (8)

25. MPC objective function
min
𝒖
𝒒 𝑘𝑓 − 𝒒 𝑑
′
𝑴 𝒒 𝑘𝑓 − 𝒒 𝑑 + ∆𝒗 𝑘 ′
𝑷 ∆𝒗(𝑘)
𝒒 = 𝑪 𝑨 𝑘 𝑓−𝑘
𝑥 𝑘 + 𝑨 𝑘 𝑓−𝑘−1
𝑩 𝑨 𝑘 𝑓−𝑘−2
𝑩 … 𝑩 𝒗 𝑘
∆𝒗 𝑘 = [𝒖 𝑘 , 𝒖 𝑘 + 1 , … , 𝒖 𝑘𝑓 − 1 ]′
− [𝒖 𝑘 − 1 , 𝒖 𝑘 , … , 𝒖 𝑘𝑓 − 2 ]′
𝒗 𝑚𝑖𝑛 𝑘 ≤ 𝒗(𝑘) ≤ 𝒗 𝑚𝑎𝑥 (𝑘)
0.1 ≤ 𝑅𝑒𝑓𝑙𝑢𝑥 𝑑𝑟𝑢𝑚 𝑎𝑛𝑑 𝑠𝑢𝑚𝑝 𝑙𝑖𝑞𝑢𝑖𝑑 𝑙𝑒𝑣𝑒𝑙 𝑓𝑟𝑎𝑐𝑡𝑖𝑜𝑛𝑠 ≤ 0.9
𝑉 𝑚𝑖𝑛
𝑖
≤ 𝑉 𝑖
≤ 𝑉𝑚 𝑎𝑥
𝑖
, 𝑖 = 𝑡𝑟𝑎𝑦 𝑛𝑢𝑚𝑏𝑒𝑟 5 𝑎𝑛𝑑 20
Subject to:
25

26. MPC results
26
Inputs:
Qualities:

27. Contributions
• For the first time an advance control system is studied for the
semicontinuous process.
• A linear state-space model is identified by subspace identification
method.
• Two different configurations for implementing MPC are investigated.
• The cascaded MPC with PI configuration reduced the cycle time by
10% and also reduced the operating cost by 11%.
• MPC showed a better stabilizing effect on the process.
• It is shown that the MPC can maintain its performance under the
changes in the feed stock composition.
27

28. Conclusion
 A novel SwoMV configuration is proposed.
 A methodology for designing a semicontinuous system based on mixed
integer dynamic optimization is presented.
 Advance MPC is implemented on the system.
28

29. Future work
 Implementing the proposed MIDO and MPC on the
SwoMV configuration.
 Including the charging and discharging modes of
semicontinuous system in the MIDO.
 Including the charging and discharging modes in the
model identification and the MPC.
29Acknowledgment: Ontario Trillium Scholarship

30. 30