Course: Advanced process control (TKP10, KP8115)
Main topic: PLANTWIDE PROCESS CONTROL
For overview of the course: see at the end

News:

08 Nov. 2012

I have updated the file pensum.txt.
The reference material on MPC is from the book by Seborg (2011).
You find all the pensum papers on the course home page.

The final exercise will be ready soon.
Unfortunately, MPC will not be included.


News: 

29 Oct. 2012
The exercises are required (mandatory), with a grade E or better.
In terms of the final grade in the module the exercises only count positive (20% exercises, 80% exam); otherwise the exam counts 100%.
Grading exercises: Excellent: A. Very good: B, Good: C, OK: D/E, Resubmit: F


26 Oct. 2012

hello all
I plan the exam as a written exam on Tuesday, 04 December, 9-11




22 Oct. 2012

Topics next two weeks:
24. Oct. : Multivariable control, decentralized, RGA (plantwide7..)
31. Oct. : Guest lecture by Stig Strand (Statoil) on MPC  (last lecture)

previous weeks:

21 Aug: Course overview
28 Aug: Plantwide control (plantwide1 up to slide 46)
05 Sep: CV selection (plantwide2) 
12 Sep: exact local method (on board) + Distillation example (2 columns)
19 Sep: Finish up CV selection (plantwide 2) 
26 Sep: TPM and concistency (plantwuide3...)
03 Oct: SIMC PID controller tuning (plantwide6...)
10 Oct: Guest lecture by Krister Forsman (Perstorp) on control structures
17 Oct: Regulatory control + Distillation control (plantwide4..)


---
News: 27 Sep. 2012

Last week (19 Sep) I finsihed the two plantwide2-presentations.
This week (26 Sep) I went through plantwide3_tpm-location.
You can now do exercise 4 ("Kida exercise on consistency").
Deadline: Monday 01 October.


---
News: 14 Sep. 2012

Hello all,
Attached you may find the exercise 3 for the Advanced Process Control course. It will be also available at the course web page later today. 
The deadline for delivering the exercise is Monday 24 Sep.
There is going to be an exercise session for this exercise on Thursday
20 Sep. 14:00-15:00.
Regards,
Vladimiros

---
News: 12 Sep. 2012

Todays lecture:

1. Deviation of loss using "exact local method" using board
   (see http://www.nt.ntnu.no/users/skoge/vgprosessregulering/lectures/plantwide2-exact-local-method_lecture-notes.pdf)

2. SIGURD's rules for CV selection:

Rule 1. Always control active constraints! (almost always)

Rule 2. Purity constraint on expensive product always active (no overpurification): 
(a) "Avoid product give away" (e.g., sell water as expensive product) 
(b) Save energy (costs energy to overpurify) 

Rule 3.
Unconstrained optimum: NEVER try to control a variable that reaches max or min at the optimum

For example, never try to control directly the cost J
- Assume we want to minimize J (e.g., J = energy) - and we make the stupid choice os selecting CV = J - Then setting J < Jmin: Gives infeasible operation (cannot meet constraints) - and setting J > Jmin: Forces us to be nonoptimal (which may require strange operation; see Exercise 3 on evaporators)

Then: Went through example with two distillation column in series. See last slides from:
http://www.nt.ntnu.no/users/skoge/vgprosessregulering/lectures/plantwide1-intro+optimal-operation.ppt


---
News: 03 Sep. 2012

Exercise 2. Please send the solution to Vlad by Thursday 05 Sep.
For questions/help for the exercises, ask Vlad.
You do not need to do items 5 and 6 at the end of exercise 2.
These will be covered in the lecture next week.

Exercise 3. coming later.

Lectures: Next lecture is Wednesday 11 sep.,  10-12.

Today (Mon 03 Sep) I covered the following (partly using the board):
I first discussed active constraint control and backoff.

Active constraint on MV: Usually no backoff required
Soft active output constraint: backoff = steady-state measurement error 
Hard active output constraint: backoff = steady-state measurement error + dynamic control error (Shift and squeeze rule applies) 

Then we considered remaining unconstrained CVs

1. Assume that all active constraints are controlled and consider remaining
unconstrained degrees of freedom (denoted u).
What to find "self-optimiszing" CVs c=Hy assocuated with u.
This means that when c is kept constant by using u, then u is close to uopt.

Options: H is selection matrix (mostly 0's with some 1's), or 
         a combination matrix; we consider linear measurement combinations only

2. The ideal self-optimizing CV is the gradient, c = Ju (where Ju = dJ/du),
The problem is that Ju is generally a function of unmeasured states (x) and
disturbances (d), that is, we cannot write Ju= Hy.

3. Simple rules for selecting "self-optimizing" CVs c=Hy

(i) Want small optimal sensitivity F^c = dcopt/dd  (where F^c = H F).
     Reason: So we can use constant setpoint when there are disturbances.
(ii) Want c to be easy and accurate to measure (and control)
     Reason: Obvious
(iii) Maximum gain rule: Want large gain G = dc/du (where G = HG^y)
    Reason: So we are insensitive to changes in c ("noise")
    Note: Can include also disturbances in maximum gain rule by proper scaling, 
    but I did not cover this in the lecture today.

Note:
   F = dyopt/dd = optimal sensitivity (corresponding to uopt(d))
   G^y = dy/du = steady-state gain

We used graphics to try to understand these rules.

3. Nullspace method for case with no measurement noise is to select H such that HF=0 (which is always possible if we have enough measuments, ny >= nu+nd).

Example marathon runner
  u = power
  d = inclination of slope
  y1 = speed [km/h]
  y2 = heart rate [pr. min]

Have one unconstrained degree of freedom (u) and want to dind one CV:
  c = H y = [h1  h2] y = h1*y1 + h2*y2
Optimal runner with d=0 (flat track)
  y1opt = 10 km/h, y2opt = 180 1/s
Optimal runner with d=1 (1 degree hill)
  y1opt = 9.5 km/h, y2opt = 181 1/s
Get optimal sensitivity
  F = [f1; f2] where f1 = dy1opt/dd = (9.5-10)/1 = -0.5 and f2=1/1=1
Nullspace method. select H such that
   HF = h1*f1 + h2*f2 = 0
Without loss of generality we can set h2=1. HF=0 then gives
    h1 = - h2*f2/f1 = 1/0.5 = 2
Conclusion. optimal CV as measurement combination:
     c = h1*y1 + h2*y2 = 2y1 + y2 
where we want to adjust u such that c=0 when there are changes in the slope (d).
Makes sense: Optimal pulse (y2) is lower when speed (y1) is higher. 

Note: If we want to remain optimal if there is one more disturbance (d2=wind), 
then we need to add one more measurement (otherwise we do not have enough degrees 
of freedom in H to make HF=0).

Proof 1 of nullspace method (Alstad): Want dcopt = 0 * dd
Here dcopt = H * dyopt = HF * dd
so we want to select H such that HF=0 (nullspace method)

Proof 2 of nulspace method (Jaschke): Ju = Juu (u-uopt) = Juu G^-1 (c-copt)
    Constant setpoint policy and no meas. error: c=0 (steady state)
    Optimal sensitivity, copt = F yopt = H F d
    Also note that, G=HGy
    So get: Ju = - Juu (HG^y)^-1 H F d
Ideal: Want Ju=0, which we see is achieved if we choose H such that HF=0 (nullspave method). 
The advantage with the second more complicated proof is that we see that the nullspace methodactually is the same as controlling the gradient Ju to zero.


---
News: 29 Aug. 2012

In today's lecture we covered the slides in plantwide1....ppt  up to slide 46
(Step S2b). We will continue on Monday 11-13 with the remaining few 13 slides and
then move on to plantwide2-cv1-selection-soc.ppt

Exercise 1 should be handed in tomorrow (Thursday 30 Aug). 
You can then start working on Exercise 2 which should be handed in next week (Thursday 06 Sep).  

---
News: 26 Aug. 2012

Based on the results of the poll, the lectures in the Process control module will be 
     Wednesdays 10-12 (room K4-205)
Only one student had a conflict at this time. 
However, I have already booked a trip to England on Wednesday 05 Sepember, 
so in that week the lecture will be moved to Monday from 11-13:
   03 september (Monday): Lecture 11-13


---
News: 21 Aug. 2012 (Tuesday)

We had the first lecture today, or actually, is was really a course overview. 
I covered up to slide 12 in plantwide1-intro+optimal-operation.ppt.

Reponsible for exercises: Johannes Jaschke  and 
Vladimiros Minasidis .

We will continue the lectures next week, after we have the results from the doodle poll:
http://doodle.com/zxqy8aiiaq9cy2zu

In the meantime:
1. Start reading the three papers that begin with plantwide (from 2000, 2004 and 2011)
2. Do Exercise 1. Please send the solution to Vlad by Thursday 30 August.
For questions/help for the exercises, ask Vlad.

------

News: 13 Aug. 2012

The first lecture and information meeting will be 

    Tuesday 21 August at 1015-1200 in room K4-205.

There will be an introduction and we will discuss time and place for the rest of the term. 

You can start with Exercise 1 NOW (Problem 1). It gives you a good introduction!
Questions:  Johannes Jäschle  (sits in K4-232)

More information: http://www.nt.ntnu.no/users/skoge/vgprosessregulering/

Note: The initial lectures are combined for the "emnemodul" TKP8 and the "PhD course" KP8115. 
Do not hesitate to ask or send emails.
Welcome to the course!

-Sigurd


-----------------------------------------------------------------------------


Course overview:
PLANTWIDE PROCESS CONTROL

This is a module (half a normal course) offered to the 5th year students at NTNU. 
It is taught every autumn, starting in late August.

Topics:
1. Optimal operation (economics) and degrees of freedom
2. What should we control from an economic point of view? (including self-optimizing control)
3. Inventory control, including location of throughput manipulator
4. Base layer control (regulatory layer; stabilizing control)
5. Tuning of PID controllers
6. Multivariable control: Interactions, decentralized control, practical use of MPC


---------------------------------------------------------------------------

BASIC INFO:

The lectures form the basis for two courses:

1. TKP10  Process Control, Advanced Course. Master specialization module  (3.75 EDU; must be taken together with another 3.75 EDU module to get full course)

2. KP8115 Advanced process control. PhD course (7.5 ECU)

----

1. TKP10  Process Control, Advanced Course (before 2009: TKP8 = TTK18) 

TKP10 PROSESSREGULERING VK 
Prosessregulering, videregående kurs 
Process Control, Advanced Course 
Lecturer: Professor Sigurd Skogestad 
Credits: 3.75 Sp
Time: According to agreement
Examination aids: D Exercises: marks 
Learning outcome : The student should be able to design plantwide control system 
Content: Control structure design for complere chemical plants. 
Selection of controlled variables (self-optimizing control). 
Consistent inventory Control. 
Regulatory control. 
Tuning of PID controllers. 
Multivariable control. 
Decentralized control. 
RGA. Introduction to MPC. Use of dynamic simulators. 
Teaching activities: Lectures, computer simulation. exercises. 
Course material: Copies from scientific papers and books including Chapter 10 in Skoegstad and Postlethwaite, "Multivariable Feedback Control, Wiley, 2010
Exam: Oral


---------------------------------------------------------------------------

2. Ph.D. course KP 8115 - Advanced process control (7.5 ECU)
Lecturer: Sigurd Skogestad 

The Ph.D. course and the 5th year Master specilization ("emnemodul") have
an initial common part, but the Ph.D. course has additional topics and
a larger emphasis on theory.

http://www.nt.ntnu.no/users/skoge/vgprosessregulering


----------
OLD!  NEWS:

10 Oct. 2011

What we have done the last few weeks:
26.09.2011: PID tuning (plantwide6_tuning)
03.10.2011: Stabilizing layer (plantwide4_regulatory+distillation)
10.10.2011: Advanced/supervisory control. MPC (plantwide5, plantwide5b)

Plans:

11.10.2011: 1300: Krister Forsman, Perstorp (with pizza)
11.10.2011: 14-16: Mehdi Panahi on MPC using Unisim (exercise 6)

17.10.2011: No lecture, deadline exercise 6
24.10.2011: Decentralized control lecture (plantwide7), final lecture.


-----------
19 Sep. 2011

Today we went through the remaining part of plantwide2-cv1-selection (active constraints) 
+ we went through plantwide3-tpm-location (see the Aske paper for more details).

You can now start Exercise 3 on consistency of inventory control. 
The deadline is Monday 26 September (I hope this is OK because we got a bit delayed with Exercise 2).

The solution for Exercise 1 (Problems 1 and 2) is now available. 


---

06 Sep. 2011

This week: Please do exercise 1, part 2 (it has been updated). Deadline: Monday 12 Sep)

Next week (Monday 12 Sep. 13-15): Introduction to use of Unisim/Hysys by Mehdi (Exercise 2)


---

05 Sep 2011

We had the second lecture today (Monday 13-16).
I covered up to slide 60 in plantwide2-cv1-selection-soc.ppt (that is, everything except active constraints + some case studies). 
I also derived the "exact local method" on the board. For details see: 
plantwide2-exact-local-method_lecture-notes.pdf

Summary of main topic:
Let y be the available measurements, including the inputs (MVs).  The CVs are 
   c = H y 
and the issue is to find the non-square matrix H.
H could be a selection matrix (if we control individual y's) or a combination matrix. 
What should c=Hy be?
  1. Control active constraints (important to reduce back-off for these)
  2. Control "self-optimnizing" variables with small loss when there are disturbances.
	- maximum gain rule (maximize gain from u to c)
	- optimal combination: 1. Nullspace method HF=0 , 2. Exact local method



---
22 Aug. 2011

We had the first lecture today. I covered up to slide 35 in plantwide1-intro+optimal-operation.ppt (up to the HDA example).
We will continue the lectures on 05 or 06 September.

In the meantime:

1. Start reading the three papers that begin with plantwide (from 2000, 2004 and 2011)

2. Do Exercise 1 (but on Problem 2 you only need to do
part 2-1 now). Please send the solution to Mehdi by Thursday 01 September.
For questions/help for the exercises, ask Johannes (for Problem 1) and Mehdi (for Problem 2).


-----