Miguel Castaño Arranz

1 Personal InformationName: Miguel Castaño ArranzDate of Birth: December 31, 1981.Gender: MaleAddress: Tulegatan 2 Floor 3, Left

Sundbyberg 172 78

Stockholm

Cell: +46 (0) 72 940 5744

Secondary phone: +46 (0) 76 278 4624

Email: miguel.castano@ltu.seCitizenship: Spanish.

Homepage: http://se.linkedin.com/pub/miguel-castaño-arranz/1a/144/ba9

Current Position

Researcher at Luleå University of Technology (January 2014 - March 2014).I am participating in a project for detecting damage in track switches with ultrasound measurements. Myresponsibilities are:

- Development of ultrasonic imaging algorithms for defect detection.

- Implementation of a prototype software tool.

Previous Position

Researcher at Luleå University of Technology (LTU) (November 2007 - February 2013).This period includes the commitment of doctoral studies (February 2008 - November 2012).During this period, I participated in several projects under the SCOPE programme, which groups severalmajor companies in the pulp & paper industry in Sweden. In these projects, I acquired large practicalexperience in designing and realizing plant experiments, process modeling, analysis, and control design.

Products

ProMoVis (Process Modeling and Visualization).I am a main designer and developer of ProMoVis, which is a software tool for the design of controlstructures for complex industry processes, focusing in the pulp & paper industry. ProMoVis can importmodels described using Modelica. ProMoVis is owned by OProVAT EF.

2 Educational Degrees

Degrees

- Ph.D. in Engineering in the subject of Automatic Control, Luleå University of Technology, November2012. The Ph.D. diploma including the list of courses is enclosed at the end of the application.

- Licentiate in Engineering in the subject of Automatic Control, Luleå University of Technology (Swe-den), 2010. Robust methods for control structure selection in paper making processes.

- M.Sc. in Industrial Engineering: Electronic Systems and Automatic Control, University of Oviedo(Spain), 2008.

Miguel Castaño Arranz 2

Other courses and programs

- IntelliCIS. Training school in "Intelligent Monitoring, Control and Security of Critical InfrastructureSystems", Varna, October 2011.

- Vrije Universiteit Brussel (Belgium). Summer school in "Measuring, Modelling and Simulation of(Non)linear Dynamic Systems", 2009.

- Luleå University of Technology. Student under the Erasmus programme, 2006-2007.

Academical activities

- Luleå University of Technology. Member of the organization committee of Reglermöte 2008. Maintask: organization of activities for Ph.D. students.

- Luleå University of Technology. Orientation guide for new students (Phösare), 2007.

- University of Oviedo. Member of the organization committee of the course study trip, 2004-2005.

3 Scientific Merits

Planned research activities

In this statement, I declare my research interests for my future activities in the structural analysis ofcomplex industry processes for control structure selection. This field differs from the subject descriptionfor the open position. By including this document, I intend to demonstrate my ability and initiative forformulating research questions.North of Sweden is a large industrialized area with industry processes as large as pulp and paper mills,plants for mineral processing and plants for steel production. These processes present a large topologicalcomplexity, since they are composed by hundreds or even thousands of sensors and actuators as well ascontrol loops which connect them. This controllers have to be placed within this complex structure andtuned appropriately for the running the process in optimal conditions.Optimality is usually defined in terms of production targets, safety requirements and energy consump-tion. The large scale aspect of these processes and the existence of recirculations and control loops derivein unpredictable dynamics which have to be understood to avoid loop interaction and performance degra-dation. Therefore, processes have to be analyzed from a holistic perspective, giving raise to plant-wideoptimization.My main research interests within the analysis of complex industrial processes are now stated.

Visualization of industrial processes

Visualization and communication tools and techniques are needed in any modern industrial plant. Dia-grams and flow sheets are used by control and process engineers for communicating process knowledgeto other workers such as operators or to decision boards. There is a need of developing visualization tech-niques and combine them with mathematical tools in order to create diagrams which allow to understandand communicate the process behavior (see Fig. 1).

Structural analysis of complex processes

My previous research derived in methods for the analysis of complex processes by using weighted graphs,resulting in diagrams like the depicted in Fig. 1. The representation of processes as graphs has severalapplications in the optimization of complex processes: process decomposition, input-output selection,controllability and observability analysis, control structure selection, only to mention some.

Miguel Castaño Arranz 3

Figure 1: Functional analysis of the input-output interconnections in a subprocess belonging to the paperindustry.

It is of my interest to find applications for my previous research in the field of plant-wide optimization.

Structural analysis of complex processes from logged process data

The existing methods for control structure selection require that a process model is available. However,modeling is a time consuming task, and its difficulty increases with the number of process variables.It is therefore of interest to create methods to estimate structural properties of processes and thereforeremoving the need of modeling prior to process analysis.My previous research generated satisfactory results in this new field, opening new opportunities andresearch perspectives.

Software platforms for the integration of methods for process optimization

Nowadays there is a large time gap between research, education and finally industry application. Beingable to integrate the latest research on process optimization on software tools allows the direct applicationof research by industry engineers.My current research includes the development of the software tool ProMovis, which is a platform forintegrating methods for process optimization.

Publication list

Peer-reviewed publications in international journals.

The following publications form part of my Ph.D. thesit:

W. Birk, M. Castaño, A. Johansson, An Application Software for Visualization and Control Structure Se-lection of Interconnected Processes, Control Engineering Practice, Volume 26, May 2014, Pages 188200.I worked together with W. Birk in the development and design of ProMoVis and I am the sole pro-grammer of its computational core. The underlying mathematical framework of ProMoVis for therepresentation of dynamic systems was created by A. Johansson.

Miguel Castaño Arranz 4

This publication has been attached to the application for its relevance. It describes the phase of technol-ogy transfer in which a software tool was created to make research results directly available to industryapplication. This has been done in tight collaboration with industry partners which participated in thedesign of the tool with requirements, testing and feedback.

M. Castaño, W. Birk, On the Selection of Control Configurations for Uncertain Systems Using Gramian-Based Interaction Measures, submitted to Automatica.I have a major contribution in this paper. The contribution of the co-author (W. Birk) was limited to thegeneration of the process models on which the case-study (bark oiler) was conducted and the revisionof the paper.

M. Castaño, W. Birk, New methods for interaction analysis of complex processes using weightedgraphs, Journal of Process Control, Volume 22, Issue 1, January 2012, Pages 280-295, ISSN 0959-1524.I have a major contribution in this paper, including conducting the research and creating the illustrativeexamples.

Proceeding in international conferences (Full papers).

M. Castaño, W. Birk, Bounds on a gramian-based interaction measure for robust control structureselection, IEEE ICCA 2011, December 2011, Santiago de Chile.

M. Castaño, W. Birk, B. Halvarsson, Empirical approach to robust gramian-based analysis of processinteractions in control structure selection, 50th IEEE Conference on Decision and Control and EuropeanControl Conference, December 2011, Orlando.

W. Birk, A. Johansson, M. Castaño, S. Rönnbäck, T. Nordin, N.-O. Ekholm, Interactive modeling andvisualization of complex processes in pulp and paper making, Control Systems 2010, Stockholm.

B. Halvarsson, M. Castaño, W. Birk, Uncertainty Bounds for Gramian Based Interaction Measures ,WSEAS International Conference on Systems 2010, Corfu.

M. Castaño, W. Birk, New methods for structural and functional analysis of complex processes, IEEEMulti-conference on Systems and Control 2009, St Petersburg.

M. Castaño, W. Birk, A new approach to the dynamic RGA analysis of uncertain systems, IEEE Multi-conference on Systems and Control 2008, San Antonio.

Theses

Practical Tools for the Configuration of Control Structures. Ph.D. Thesis, Luleå University of Technol-ogy, 2012.

Robust methods for control structure selection in paper making processes, Licentiate Thesis, LuleåUniversity of Technology, 2010.

Sensitivity of Variable Pairing in Multivariable Process Control to Model Uncertainties, Master’s Thesis,Luleå University of Technology, 2007.

Approved Research Grants

PrOSPr (2012). PrOSPr is a continuation of the project MeSTA (2007-2011). The objective of MeSTAwas to develop robust and reliable methods for structural analysis and optimization of complex industryprocesses so that these methods become reliable and sufficiently robust to become packaged in tools. Thesoftware application ProMoVis was a product which resulted form MeSTA, and the goal of PrOSPr is the

Miguel Castaño Arranz 5

open source distribution of ProMoVis. These projects group several major companies in the pulp & paperindustry in Sweden, as well as consultancies. I was a co-author for the funding application for PrOSPrwith Wolfgang Birk.

Network/Research Collaboration

I worked for 5 years in different projects under the SCOPE programme, which is administrated by ProcessIT. In these projects, I also participated leading several work packages. The projects are:

- MeSTA (2007-2011). The objective is to develop robust and reliable methods for structural analysisand optimization of complex industry processes so that these methods become reliable and sufficientlyrobust to be automated and packaged into tools. The project groups several major companies in thepulp & paper industry in Sweden, as well as consultancies. I worked in this project as researcher,developer, and programmer.

- PrOSPr (2012). The objective is to release the software tool ProMoVis under an open source project. Iworked writing the project application, and is currently working as developer, programmer and tester.

- EQoRef (2012-2013). Energy and quality oriented modeling and control of refiners.

My network of contacts is strongly influenced by his work under the SCOPE programme, with additionof other industry and academic contacts derived from personal and professional relationships.

Contacts in consultancy for process industry

Optimation AB This consultancy participated l in the previously mentioned projects, and I am currentlycollaborating with them in different funding raising activities. Optimation AB is also a a co-owner ofOProVAT together with me and other personal entities.

Eurocon AB This consultancy participated in the research project MeSTA, in the analysis of differentindustrial processes and the development of the software tool ProMoVis.

Process industry contacts

SCA Obbola AB I had a strong collaboration with SCA Obbola AB during his Ph.D. studies and workedat their plant with two of their processes: the bark boiler and the stock preparation plant.

BillerudKorsnäs Karlsborg AB I had a strong collaboration with SCA Obbola AB during his Ph.D.studies and worked at their plant with two of their processes: the bark boiler and the secondary heatingsystem.

ArcelorMittal ArcelorMittal is a leading integrated steel and mining company. I keep a strong contactwith many employees, including process engineers, technical staff, operators, administrators or projectleaders. I am currently working with project leaders on the seeking of projects under the FP programme.

Academic contacts

Federico Santa María Technical University, Chile. The department of electronics is a world leadingdepartment in the design of control structures for multivariable processes. The head of the department,Mario Salgado was the faculty opponent of my Ph.D. thesis.

Miguel Castaño Arranz 6

Uppsala University I participated in several publications with Björn Halvarsson, who currently holds adoctoral degree from Uppsala University. In addition, Prof. Bengt Karlsson formed part of the evaluationtribunal for my Ph.D. thesis.

Research Awards

- Norrbottens forskningsrÃeds award on the honor of Curt BostrÃum (2013). This award is yearlygranted to two thesis in Norrbotten in the field of technology.

Peer-review/charing assignments

- Arabian Journal for Science and Engineering (AJSE). One review in 2013.

- ACC 2012 Reviewer.

- ICCA 2011 Co-chair of session Robust Control and Systems I.

- ICECS 2011 Reviewer

- CDC’2009. Reviewer.

- MSC’2009. Chair of session Complex and Chaotic Systems.

- MSC’2008. Reviewer. Co-chair of session Modeling and Identification.

4 Educational Qualifications

Teacher Portfolio

Teaching philosophy

Teaching is a success when the student acquires the targeted knowledge, but also when the teacher changesthe vision of teaching. This can only be achieved through a deep reflection on teaching experiences.The complexity of the teaching task forces a good teacher to gain experience and reflect upon it being ableto derive new teaching strategies as well as discard or modify unfruitful ones.Even the most experienced teacher has to avoid the risk of routines. Routines in teaching lead to a focuson the subject instead of on the link between the student and the subject. Society evolves rapidly and thisderives in an evolution in the academical environment as well as in the taught disciplines. The teacher hasto keep up to date with progress and adapt his teaching with adequate and motivating strategies, sincemotivation is a key for the students to address learning in a deep approach.

How can I develop as a teacher?

The key for developing is to systematically reflect on teaching experiences and actuate in concordancewith the obtained conclusions. In the case of courses, a good tool for reflection is the feedback received inthe course evaluation, which collects the opinion of the students. These course evaluations are usually themost effective way of identifying the weaknesses and strengths of your teaching or your course material.Reflecting on teaching experiences is a need but it is not sufficient for the development of the modernteacher. The current evolution of communication techniques provides excellent channels for the distri-bution of teaching material as well as quick and efficient interfaces to provide support to the studentsand enhance their mutual collaboration. The modern teacher has to master this new technologies and beable to provide with i.e. online lectures for distance teaching, multimedia tools for creating tutorials andstudent feedback or virtual rooms for the interaction with the students.

Miguel Castaño Arranz 7

Reflections on surface and deep approaches to learning.

"Student learning research originated in Sweden, with Marton and Säljö’s ([]) studies of surface and deep approachesto learning. They gave students a text to read and told them they will be asked questions afterwards. Studentsresponded in two different ways. The first group learned in anticipation of the questions, concentrating anxiouslyon the facts and details that might be asked. They skated along the surface of the text, ... using a surface approachto learning. What these students remembered was a list of disjointed facts; they did not comprehend the point thatthe author was making. The second group on the other hand set to understand the meaning of what the author wastrying to say. They went below the surface of the text to interpret that meaning, using a deep approach.Taking a surface or deep approach to learning is mainly a preference from the student. However, theteacher and the course material can influence the students towards one of the approaches or the other.One of the objectives as a teacher should be to try that most of the students use the deep approach tolearn.

Own experiences with surface learning. After knowing about he research from Marton and Säljö’s Iwas able to analyze cases in which I was not maintain any of the contents of the course after taking it, andidentify some of them as cases in which I took a surface learning approach. I consider that my naturalapproach to learning is a deep approach, however after a personal reflection I concluded that in somecases the teacher or the course structure influenced me in such a way that I selected a surface approach.Surface learning during my university studies.An interesting case was a course in which out talkative and amusing teacher succesfuly attracted theattention of the students to his lectures. Most of us got good grades in the course. An evaluation of thecourse performance would probably have brought up a very high ranking, since the students were veryhappy with the course and the grades were good. However, I find myself as unable to recall any detailsof the course. For years I wondered why this happened, until I read the work from Marton and Säljö’s.The main problem of the course was the exams. A large part of the course evaluation was formed bymultiple choice tests, with the peculiarity that most of the questions were repeated, or were very similarto the ones in exams of previous years. The way I studied for the exams, was to start by taking all thequestions of previous exams and face them one by one. If I didn’t know the answer, I would look in thebook and only read the paragraph in which the answer is found. In this way, I built my knowledge in afast way by learning only the parts which had a large probability of appearing in the exam, being ableto discard a larger part of the book. The good grade in the exams was guaranteed with minimal effort.However the knowledge was stored in my mind as a set of unconnected facts, without having a clear ideaof the full picture. Those facts faded away from my memory and nothing remains.Surface learning during my school studies.I often wondered why after my school studies, my knowledge in history, geography and other humansciences was so bad. A reflection based on the theories from Marton and Säljö’s brought up an explanation.I have to admit that the contents of the programmes were quite complete. However, the teachers encour-aged us to memorize texts. Some of them facilitated it even by forcing us to reedit the textbooks addingpen annotations with their suggestions for a simplified text. This was done by crossing words or parts ofsentences and adding some other words to the text. All the exams also encouraged the surface approachsince the student was just asked to repeat full sections of the text book. Even if the teachers were sayingthat it is better to use our own words in the exam instead of directly repeating the text book, this usuallylead to a degradation of the grades, since the teacher would later argue that some important details havebeen distorted. It is obvious that the approach taken by the students in this case would be to just memorizeall the texts word by word during the previous days to the exam without the need of even understandingthe text. It didn’t matter if you don’t understand words such as abdication or democracy, as long as youhave placed them properly in the text.This means that all the knowledge is these topics is lost as soon as you make the exam and stop doing theeffort of memorizing those texts. Years and years of education in social sciences have been mostly wastedfor me. The only contents I keep from those programmes are those in which I found special interest and

Miguel Castaño Arranz 8

for which I also consulted additional sources myself. Some time after my school studies I started to feel asignificant ignorance in historical and geographical knowledge which I’m still trying to compensate.Finally I have to say that I never had this problem with any technical disciplines. The need of understand-ing and applying theoretical concepts made these subjects interesting but also easy to remember. A deepapproach for learning is a natural approach to this sciences almost regardless from the influence of theteachers.

Teaching contributions. Taught courses at Luleå University of Technology

I participated in the following courses at Luleå University of Technology:

- Modeling and control. 2011-2012. Problem solving lectures. This is the course in which I had a morerelevant participation. The student response related to the problem solving sessions (section 8 in theevaluation) is included below.

- Multivariable and robust control systems. 2010-2011. Labs and project assistant and examiner.

- Modeling and control. 2009-2010. Labs and project assistant and examiner.

- Nonlinear and optimal control systems. 2009-2010. Labs and project assistant and examiner.

- Automatic control. 2009-2010. Labs and project assistant and examiner.

The credentials for this courses are enclosed at the end of this section. From these course experiences,the the one which produced a larger impact in me is the lab assistance in the course R0002E during theacademic year 2009-2010. My reflections on this experience are included below.

Reflections on R0002E

This was a very basic course, however it requires preliminary knowledge, mainly in mathematics (i.e.calculus and differential equations) and physics (i.e. cinematics and electricity). Students from severaldifferent programmes and with different backgrounds participate in the course (mechanical engineering,chemical engineering, electrical engineering, electronics . . . ). Even if most of the students are or were en-roled in courses including the preliminary contents it is likely that their knowledge in these preliminariesis not very solid yet, since they are in the first years of their education.This meant that students often came with loads of questions on very basic principles of physics and math-ematics, involving for myself a large time consumption in the demotivating task of reviewing with manystudents such basic topics as Newton’s or Hooke’s laws or Taylor expansions. The worst consequencewas the slow and hard progress of the students, and their loss of time waiting long queues in front of myoffice.The success rate of the students at the final lab Project was almost total. However the path to the goal wastedious for me, the students and the lecturer.One can think of blaming the educational system or the programmes structure for the poor backgroundof the students, or blame the students for their laziness in not reviewing the preliminaries by themselves.Nevertheless, there is little which could be done by me if keeping only these postures.A reflection brought up that leaving freedom to the students for forming the groups means that they willtend to team up with their friends and colleges from their programs, and it is then likely that all the peoplein the same group will have similar background and knowledge. By forcing/encouraging the students toteam up in groups with people from different programmes, the students would have been more likely tolearn from each other since their backgrounds would be more likely to be complementary, and thereforethe overwhelming amount of questions might have been avoided.

[1] F. Marton and R. Säljö, On Qualitative Differences in Learning: 1–Outcome and Process, British Journalof Educational Psychology, 1976, volume 46, pages 4-11.

Miguel Castaño Arranz 9

Experience of supervision

Co-supervision of Master’s thesis: Pablo Fernandez de Dios. Implementation of a Visualization Toolin MatLab for Structural Analysis of Complex Processes. 2011. Principal supervisor: Wolfgang Birk.For the credential, see the attached Ph.D. diploma under the title Supervision of Master’s Thesis students.

Teacher training

Course: Pedagogy in higher education. 7.5 ECTS. for credentials, see the Pd.D. diploma.

Development work in education

I developed the course General topics in applied control. The development included the study guide and thelab material. The study guide is after the credentials of participation in courses and the study guide.

Other educational activities

I wrote the user documentation for the software tool ProMoVis, and participating in creating and instruct-ing several workshops for industry members about ProMoVis.

Wolfgang Birk, Modellbygge och reglering -

20.12.2011 EvaSys evaluation Page 1

Wolfgang Birk

Modellbygge och reglering (R0002E)Response rate = 48.1 %

Survey Results

Legend

Question text Right poleLeft pole n=No. of responsesav.=Meandev.=Std. Dev.ab.=Abstention

25%

1

0%

2

50%

3

0%

4

25%

5

Relative Frequencies of answers Std. Dev. Mean

Scale Histogram

1. SjälvbedömningSjälvbedömningSjälvbedömningSjälvbedömning / Self-assessment

Hur stor del av beräknad studietid (helfartskurs 40 h/vecka, halvfartskurs 20 h/vecka, kvartsfartskurs 10 h/vecka) har lagts påHur stor del av beräknad studietid (helfartskurs 40 h/vecka, halvfartskurs 20 h/vecka, kvartsfartskurs 10 h/vecka) har lagts påHur stor del av beräknad studietid (helfartskurs 40 h/vecka, halvfartskurs 20 h/vecka, kvartsfartskurs 10 h/vecka) har lagts påHur stor del av beräknad studietid (helfartskurs 40 h/vecka, halvfartskurs 20 h/vecka, kvartsfartskurs 10 h/vecka) har lagts pådenna kurs, schemalagd tid plus hemarbetstid? / denna kurs, schemalagd tid plus hemarbetstid? / denna kurs, schemalagd tid plus hemarbetstid? / denna kurs, schemalagd tid plus hemarbetstid? / A full-time course is estimated as 40 hours of study; part-time courses are either20 or 10 hours per week. What percentage of this time did you spend on this course, count the time spent both in class and onself-study?

1.1)

n=22< 25% 4.5%

26-50% 22.7%

51-75% 27.3%

76 -100% 13.6%

> 100% 31.8%

Jag är nöjd med mina insatser under kursen. / Jag är nöjd med mina insatser under kursen. / Jag är nöjd med mina insatser under kursen. / Jag är nöjd med mina insatser under kursen. / Iam satisfied with my efforts during the course.

1.2)Instämmer helt/Instämmer helt/Instämmer helt/Instämmer helt/ Strongly agree

Instämmer ej/Instämmer ej/Instämmer ej/Instämmer ej/ Strongly disagree

n=25av.=3.2dev.=1.3

8%

1

24%

2

28%

3

24%

4

12%

5

4%

6

Jag har deltagit i kursens allaJag har deltagit i kursens allaJag har deltagit i kursens allaJag har deltagit i kursens allaundervisningsmoment.undervisningsmoment.undervisningsmoment.undervisningsmoment. / / / / I have participated in allteaching instances in the course.

1.3)Instämmer helt/Instämmer helt/Instämmer helt/Instämmer helt/ Strongly agree

Instämmer ej/Instämmer ej/Instämmer ej/Instämmer ej/ Strongly disagree

n=24av.=5dev.=1.3

0%

1

12.5%

2

0%

3

4.2%

4

41.7%

5

41.7%

6

Jag har förberett mig inför allaJag har förberett mig inför allaJag har förberett mig inför allaJag har förberett mig inför allaundervisningsmoment.undervisningsmoment.undervisningsmoment.undervisningsmoment. / / / / I have been wellprepared for all teaching instances.

1.4)Instämmer helt/Instämmer helt/Instämmer helt/Instämmer helt/ Strongly agree

Instämmer ej/Instämmer ej/Instämmer ej/Instämmer ej/ Strongly disagree

n=25av.=2.9dev.=1.4

24%

1

8%

2

40%

3

12%

4

16%

5

0%

6

2. Kursens mål & innehållKursens mål & innehållKursens mål & innehållKursens mål & innehåll / Course aims and content

Kursens mål har varit tydliga.Kursens mål har varit tydliga.Kursens mål har varit tydliga.Kursens mål har varit tydliga. / The aims of thecourse are clear.

2.1)Instämmer helt/Instämmer helt/Instämmer helt/Instämmer helt/ Strongly agree

Instämmer ej/Instämmer ej/Instämmer ej/Instämmer ej/ Strongly disagree

n=25av.=3.6dev.=1.3

4%

1

20%

2

20%

3

24%

4

28%

5

4%

6

Kursens innehåll har bidragit till att uppnåKursens innehåll har bidragit till att uppnåKursens innehåll har bidragit till att uppnåKursens innehåll har bidragit till att uppnåkursplanens mål.kursplanens mål.kursplanens mål.kursplanens mål. / The contents of the coursehelp to achieve/meet the course’s aims.

2.2)Instämmer helt/Instämmer helt/Instämmer helt/Instämmer helt/ Strongly agree

Instämmer ej/Instämmer ej/Instämmer ej/Instämmer ej/ Strongly disagree

n=24av.=3.7dev.=1.1ab.=1

0%

1

20.8%

2

12.5%

3

45.8%

4

16.7%

5

4.2%

6

Kursplaneringen/studiehandledningen har gettKursplaneringen/studiehandledningen har gettKursplaneringen/studiehandledningen har gettKursplaneringen/studiehandledningen har gettgod vägledning.god vägledning.god vägledning.god vägledning. / The course planning andsupervision are structured and easy to follow.

2.3)Instämmer helt/Instämmer helt/Instämmer helt/Instämmer helt/ Strongly agree

Instämmer ej/Instämmer ej/Instämmer ej/Instämmer ej/ Strongly disagree

n=25av.=3.8dev.=1.3

0%

1

20%

2

28%

3

16%

4

28%

5

8%

6

Wolfgang Birk, Modellbygge och reglering -

20.12.2011 EvaSys evaluation Page 3

Handledningen vid laborationerna var till hjälp attlösa uppgifterna.

7.2)Instämmer helt/ Strongly agree

Instämmer ej/ Strongly disagree

n=20av.=1.9dev.=1

40%

1

40%

2

10%

3

10%

4

0%

5

8. Övningarna

Övningstillfällen har underlättat inlärningen avkursens teoretiska innehåll.

8.1)Instämmer helt/ Strongly agree

Instämmer ej/ Strongly disagree

n=18av.=3.2dev.=1.1

16.7%

1

0%

2

33.3%

3

50%

4

0%

5

Det fanns tillräcklig med tid under övningarna föratt ställa frågor och diskutera uppgifter

8.2)Instämmer helt/ Strongly agree

Instämmer ej/ Strongly disagree

n=18av.=2.9dev.=1.1

11.1%

1

27.8%

2

27.8%

3

27.8%

4

5.6%

5

Lärarnas insats var ett bra stöd för att lära sigtillämpa det teoretiska innehåll i kursen.

8.3)Instämmer helt/ Strongly agree

Instämmer ej/ Strongly disagree

n=17av.=3.4dev.=0.9

5.9%

1

5.9%

2

35.3%

3

47.1%

4

5.9%

5

Wolfgang Birk, Modellbygge och reglering -

20.12.2011 EvaSys evaluation Page 13

8. Övningarna

På vilket sätt kan övningstillfällen förbättras?8.4)

Wolfgang Birk, Modellbygge och reglering -

20.12.2011 EvaSys evaluation Page 14

General topics in applied control

STUDY GUIDE

Author: Miguel Castaño Arranz

1.- Introduction

Control engineering is a rapidly evolving discipline. There is a large number of traditional control

strategies which are still being improved as well as a number of emerging ones. It is hard for the control

engineer to choose which control strategy is suitable for a specific application, as well as to choose a

topic in control engineering for developing his/her skills.

This course is aimed for Ph.D. students in control who want to explore a new control strategy and have

an understanding of which other control strategies exist and how can they contribute to their

professional development.

2.- Intended Learned Outcomes

After taking this course you should be able to:

Apply acquired theoretical knowledge on a control topic of your choice.

Communicate control theory concepts and applications of the selected control topic.

Criticize and compare different control strategies using the work of other students as source.

Reflect on which control topics you can learn in the future for your professional development.

Present your work formally correct in both written (technical report) and oral form

(presentation).

3.- Preliminary knowledge.

Those Ph.D. students who have taken the course Automatic Control R0002E or similar

are considered to have enough knowledge to face any of the advanced topics.

Ph.D. students without this background which want to learn basic topics in automatic

control should have the following skills:

- Basic knowledge of Matlab. Knowledge of Simulink is desired but not compulsory.

- Knowledge of differential equations and the Laplace transform.

- Basic algebra notions.

- It is desired to have basic knowledge of physical laws, including balances of

masses/energy.

4.- Course activities

The goals of the course activities are:

o Acquire knowledge in a control topic on your choice.

o Design and implement a controller on a benchmark process using the acquired

knowledge.

o Evaluate the implemented controller and the selected strategy.

o Disseminate your results in a presentation form, a formal report, and tutorial sessions

with other students if needed.

o Compare the control strategy that you selected with other strategies.

These activities are arranged as follows:

1.- Choose a control topic from a proposed list or propose a topic of your interest. A proposed list is:

Advanced topics:

Robust Control

Fuzzy Control

Model Predictive Control (MPC)

Adaptive Control

Control with Neural Networks

Decentralized control and interaction measures

Internal model control (IMC) and Smith predictor

Achievable performance of multivariable control structures

Modeling and control of time-varying systems

Sliding mode control

Optimal Control

Flatness Based Control

LMI control

Basic topics:

PID Control (Basic)

Lead/Lag Compensation

2.- Create a plan. After a short period of reading information on that topic, the Ph.D. student will make a

plan for implementing a control strategy on a benchmark project. The plan will be review to see that the

difficulty of the work is in accordance with the scope of the course. A short report of not more than one

page has to be delivered.

3.- Design and implement a control solution on the benchmark process. The selected benchmark

process will be a quadruple-tank system. A full detailed model in MATLAB/SIMULINK will be given to the

students to implement their work. Depending on the availability of a real quadruple-tank process, the

students might be requested to make a demonstration in real life.

4.- Present your results. A poster presentation will be made at the end of the course in which the

student will make an introduction to the used control theory and will show the results when applied to

the benchmark process. The presentation has to be structured as described in section 5.

5.- Compare different control strategies. This is divided in the following two sub-activities:

- Fill a survey on the material presented by other students during the poster presentation.

- Choose the work of other student and compare it with your work, focusing in a qualitative (and

quantitative is possible) comparison of your selected strategy with the strategy of the other student.

6.- Report your conclusions. This is done by delivering your implementation with a report which must

include the contents specified in section 5.

5.- Course evaluation and reporting.

To pass the course, you have to:

Successfully implement the selected advanced control technique on the quadruple tank

system.

It is not sufficient to design a stable controller. The controller has to achieve a satisfactory

performance (track steps on the reference, reject process disturbances, …).

It is possible to pass the course with a poor performance controller, but only if this is a

consequence of the chosen control strategy and not of a poor design. In this case, the

limitations on the achievable performance which are imposed by the selected control strategy

have to understood and described by the student both in the oral presentation and the report.

Disseminate your results.

- Give a poster presentation with the structure described in the table below.

- Deliver your poster and implementation after the presentation.

- Give the needed support to the student who will be comparing his control strategy

with yours.

The dissemination of results is considered as passed when the previous activities have been

done and the supported student shows that he has understood your work by reporting a

comparison with his own work. If the supported student fails to successfully report this

comparison, the examiner will evaluate if the dissemination tasks area passed. In this case, the

surveys filled by the other students will be considered as the main tool to judge the

dissemination of results.

Fill a survey on the different control strategies. The survey will be handed to the students at the

poster presentation and has to be filled with the results reported by the other students.

Report your conclusions. Deliver a report with the structure described in Activity 5. The report

as to be graded as passed by the course supervisor.

To show success the student has to deliver a portfolio with the following files:

Report of activities (Excel file).

Work plan.

Implemented control strategy (Matlab file).

Poster.

Survey.

The structure of these documents and the delivering deadline are summarized in the following table:

DOCUMENT DEADLINE STRUCTURE OF THE DOCUMENT

Draft for the work plan At Meeting 2 1 page describing the control strategy to be designed and implemented on the quadruple tank.

Work plan 1 day after meeting 2 Same as above.

First Simulation Meeting 3 Deliver the files needed to run your simulation on the quadruple tank.

Poster Meeting 4 The poster has to follow a given structure: 1.- Description of the used theory. 2.- Details on the control design and implementation. 3.- Simulation results. 4.- Critical evaluation of the implemented control strategy.

Survey Meeting 4 The survey will be distributed and filled at the poster presentation.

Final Simulation 1 week after meeting 4 Provide in a rar the needed files to run the simulation. Include also a README.txt file describing technical details in how to run/tune the simulation.

Report 1 week after meeting 4 The report given structure: 1.- Description of the used theory supported by references. 2.- Details on the control design and implementation. 3.- Simulation results. 4.- Critical evaluation of the implemented control strategy. 5.- Qualitative(and quantitative if possible) comparison of your method with the method chosen by other student.

6.- Timeline

This section describes the meetings which will take place during the course. It is important to

check which assignments you have to have prepared for the meeting.

Tasks which have to be ready before the meeting

Activities at the meeting Outcome

Meeting 1. Introduction.

Day 1.

None * The supervisor presents brief information on several control topics. * Students choose control topic trying to avoid repeated topics if possible

* Worksheet summarizing the individual choices.

Meeting 2. Review of work plan.

Day 15

* Work plan in written form. (Maximum 1 page)

* Students present their plan. * The plans receive feedback from other students as well as from the course supervisor. * The course supervisor accepts the plans after possible modifications.

* Reviewed work plan.

Meeting 3. Follow up meeting.

Day 60

* It is desired to have a simulation of the implemented controller at this stage.

* Students present the current status of their work. * The plans receive feedback from other students as well as from the course supervisor.

* The produced results in the projects are reviewed and the projects are steered if needed.

Meeting 4. Dissemination of results (poster session).

Day 70

* Poster with the structure described in Section IV. * Final implemented controller.

* Present your results in a poster session. * Fill a survey which helps you to criticize and compare the control strategies chosen by the other students. * Receive feedback from industry personal who will participate in the meeting.

* Survey. * Selection of the work of other student to compare with your own.

7.- Plagiarism

Detected plagiarism will be reported and will involve the failing of the course. The following will be

considered plagiarism:

1. The reuse of unreferenced material.

2. The unreferenced reuse of control designs/algorithms. The implementation and design of the

control strategy has to be original.

Allowed material.

Only the following toolboxes can be used Matlab toolboxes can be used:

- Control Systems Toolbox

- Symbolic Math Toolbox

Using other toolboxes or functions in your final work is only allowed under the explicit permission of the

teacher. The delivering of work with forbidden functions/toolboxes might derive in a rejection and a

resubmission but will not be considered as plagiarism.

8.- Missing a deadline.

Failing to present the required results at the poster session will involve the failing of the course.

9.- Course contact & support.

Course supervisor:

Name: Miguel Castaño

Mail: miguel.castano@ltu.se

Office hours: Monday 10:00 – 12:00. Thursday 15:00-17:00.

10.- Course literature.

There is no course book. Each student has to select the literature which is relevant to the

selected topic with the support of the teacher. The literature might include but not be restricted

to: academic books, journal publications, conference papers, ...

11.- Course credits.

Passing the course awards 7.5 ECTS. The expected amount of time in the course activities is

summarized in the worksheet distributed in Appendix I. You are encouraged to fill in the Excel

worksheet with the time you spent in the activities. The number of awarded credits will be

reviewed in case of large deviations from the total expected spent working hours.

APPENDIX I. WORKSHEET FOR THE ACTIVITY REPORT.

The following worksheet will be used by the students to report the spent time in course activities:

APPENDIX II. REFLECTIONS ON THE COURSE STRUCTURE.

This appendix is not part of the study guide. It collects a set of reflections which motivate the contents of

this study guide.

Most of the Ph.D. students in control engineering have to read a new topic in control. If the

student reads the topic with no supervision, there is a large probability that the student will just

read a book and take a surface approach to leaning (see Pages 22-24 in [1] ). One of the main

goals of this course is ensuring a deep learning approach by planning appropriate activities (see

Pages 24-25 in [1]). These activities are the design and implementation of a control solution, and

the dissemination of results.

The amount of credits to be awarded by this course is a bit uncertain. The original plan is to give

7.5 ECTS. Nevertheless, the students are asked to report the spent time by filling the Excel

worksheet included in Appendix I. This will help to adjust the course credits in case of large

deviations.

An important objective of the course is the dissemination of results. An appropriate

dissemination will ensure that the students receive a broad picture of which other control

techniques exists and how can they contribute to their professional development.

It is of importance that the students are able to compare the different selected control topics.

For this purpose, the design and implementation work will be done on the same benchmark

process. The selected benchmark process is the quadruple-tank system due to its well

demonstrated pedagogic advantages (see [2]).

It is of importance that the documents and simulations generated by the students are collected

and kept as source of information for future students. Therefore, all the deliverables have to

follow a predefined structure in order to improve their readability.

The fact that the dissemination of results is an important task imposes hard deadlines, since the

results have to be ready at the date of the presentation. Section 8 mentions explicitly the

consequences of missing a deadline.

The ILOs have been aligned with the course activities and the evaluation as described in Chapter

4 in [1].

As teacher, I want to to be able to ask students for resubmission of their work if they fail to

report properly. My interest is educational, but also maintaining good documentation as an

outcome from the course. To be able to do this, the following ILO was specified:

”Present your work formally correct in both written (technical report) and oral form

(presentation).”

The ILOS have been designed using the procedure described in [1] (Pages 83-85). A review of the

ILOs as indicated in Page 85 was performed in a final review of the study guide. This review led

to the inclusion of the ILO: “Reflect on which control topics you can learn in the future for your

professional development.”

From my previous experiences as teacher I concluded that it is good both for the students and

the teacher to have a clear and explicit definition of what is considered plagiarism in the study

guide. For this purpose section 7 was added.

[1] John Biggs and Catherine Tang, Teaching for quality learning at University. Third Edition. Mc Graw

Hill.

[2] Karl Henrik Johansson, The quadruple-tank process. A multivariable laboratory process with an

adjustable zero. IEEE transactions on control systems technology. Vol 8, No 3, May 200.

Miguel Castaño Arranz 28

5 Additional Assignments

Board member at OProVAT EF (since June 2012).OProVAT has as goal the Open source distribution of Process Visualization and Analysis Tools. OProVATresulted from a research project in which the software tool ProMoVis was created. OProVAT and ProMoVisare maintained by funded projects with strong industry collaboration.My tasks in OProVAT include:

- Write funding applications which comprise partners in Swedish industry and academia.

- Develop the software tool ProMoVis, being up to the date the sole programmer of its mathematicalpart, which is programmed using MATLAB.

- Actively research in the field of control structure selection.

- Organize seminars, workshops and tutorials regarding control structure selection and ProMoVis.

- Identify industry needs and synthesize new development tracks.

6 References

Reference #1

Name: Wolfgang BirkTitle: Associate ProfessorCompany: Luleå University of TechnologyPrimary e-mail: wolfgang.birk@ltu.sePrimary phone number: +46 725 39 09 09

Secondary phone number: +46 920 49 19 65

Wolfgang was my supervisor during my Ph.D. studies. He was the main manager in the projects in whichI participated. We collaborated in the conceptual design of the software tool ProMoVis. Together withother partners, we founded the company OProVAT EF, which is the copyright owner of ProMoVis.

Reference #2

Name: Björn HalvarssonTitle: Ph.D., Research EngineerCompany: Ericsson ABPrimary e-mail: bjorn.halvarsson@ericsson.comPrimary phone number: +46 107 17 45 05

Secondary phone number: +46 722 44 15 05

Björn Halvarsson graduated as Ph.D. from Uppsala University in 2010. We first met in a conference in2008 and a series of technical discussions brought important research results for my doctoral thesis. Wehave 2 conference publications together.

An Application Software For Visualization and Control Configuration Selection ofInterconnected ProcessesI

Wolfgang Birka,∗, Miguel Castanoa, Andreas Johanssona

aControl Engineering Group, Department of Computer Science, Electrical and Space Engineering, Lulea University of Technology,SE-971 87 Lulea, SWEDEN

Abstract

This paper presents a new application software for control configuration selection of interconnected industrial processes,called ProMoVis. Moreover, ProMoVis is able to visualize process models and process layout at the physical leveltogether with the control system dynamics. The software consists of a builder part where the visual representationof the interconnected process is created and an analyzer part where the process is analyzed using different controlconfiguration selection tools.

The conceptual idea of the software is presented and the subsequent design and implementation of ProMoVis isdiscussed. The implemented analysis methods are briefly described including their usage and implementation aspects.The use of ProMoVis is demonstrated by an application study on the stock preparation process at SCA Obbola AB,Sweden. The results of this study are compared with the currently used control strategy.

The study indicates that ProMoVis introduces a systematic and comprehensive way to perform control configu-ration selection. ProMoVis has been released under the Apache Open Source license.

Keywords: Visualization, signal flow graphs, interaction measures, control structure, control configuration,multivariable control, process control, interconnected systems, pulp and paper industry

1. Introduction

Continuity is an important aspect of industrial process plants. It means that the industrial plant has a certainlevel of availability for production and evolves with maintenance and optimization efforts. Nowadays, availabilityof production plants need to be very high and the production quality needs to be well aligned with customers’requirements, (El-Halwagi, 2006). In turn, the requirements on performance of processes, their control and maintenanceare high, and any changes in hardware should lead to adaptations in the control systems more or less right away.

However, these industrial process plants are interconnected systems where hundreds or even thousands of variablesare connected through dynamic systems, resulting in a so-called topological complexity, (Jiang et al., 2007). Theseconnections can be physical connections between components, plant-wide access of information by the control system,or control actions by the control system on a plant-wide scale. Examples of physical interconnections are materialflows and reflows, like discarded material which is returned to a previous process step and thus gives rise to largerecycle loops.

A consequence of this topological complexity is that adding control loops to a process in an ad-hoc manner mayresult in a system with obscure causality and unforeseen dynamics. Understanding of such systems becomes a challengewhich makes the control configuration task very difficult. Remember, control configuration selection (CCS) addressesthe problem of finding a low complexity structure for a controller for an industrial process that has the potential torender a control system with desirable performance. It does not involve the parametrization of the controller.

The first methods date back more than four decades, initiated by the work published in (Bristol, 1966) and(Rijnsdorp, 1965) where small scale multivariable problems were addressed. Since then, the host of methods has

IThe work has been conducted within the MeSTA project that is hosted at ProcessIT Innovations at Lulea University of Technologyand run within the branch framework SCOPE. Funding provided by VINNOVA, Hjalmar Lundbohm Research Centre and the participantsof the SCOPE consortium, is hereby gratefully acknowledged. The authors also want to thank the reviewers and associate editor for theirconstructive comments that helped to improve the article.

∗Corresponding author: wolfgang.birk@ltu.se, +46 725 390909

Preprint submitted to Elsevier January 13, 2014

increased largely and can now be used to determine feasible control configurations for problems of larger scale. Thishas also led to the introduction of the control structure selection problem which contains the I/O selection problemand the control configuration selection problem as sub-problems. A good overview of the topic and available methodsis given in (van de Wal and de Jager, 2001), (Skogestad and Postlethwaite, 2005) and (Khaki-Sedigh and Moaveni,2009). It is also important to mention that these methods are not viable on a plant-wide scale, where the total numberof inputs and outputs exceeds a few dozen.

Despite the vast host of proposed methods for CCS, there are no up-to-date toolboxes available for industrial useof the methods. To the knowledge of the authors, the only toolbox reported in the literature is by Nistazakis andKarcanias (2004), but it does not seem to be widely available.

As indicated in (Rohrer, 2000), visualization is important both from a collaborative perspective as well as toprovide a comprehensive understanding of processes. Within the areas of construction, manufacturing, or productionmanagement, visualization is recognized as an important tool, see (Bouchlaghem et al., 2005; Browning and Ramasesh,2007), but when it comes to the design and maintenance of control systems in process industries, the use of visual-ization is still very limited. Available software that can be used for visualization focuses mainly on simulation of theprocess dynamics, such as ChemCAD, MATLAB/Simulink, LabView, Extend, or Dymola, the latter based on thegeneric modeling language Modelica. However, there is a lack of user-friendly toolboxes or software aiming at controlconfiguration selection.

The aim of this paper is to propose a new application software, called ProMoVis, that combines a graphical repre-sentation of a process plant and control system with analysis of the dynamic interconnections for control configurationselection. The underlying mathematical framework is the directed graph which is a highly abstract way of representingtopological complexity in various applications.

Based on this mathematical framework a set of selected control configuration methods is implemented and canbe used to analyze interconnected processes. Thereby, even mathematically complex methods become available forindustrial use. Obviously, analyses performed by ProMoVis have the same limitations as the implemented controlconfiguration methods, which means that the user has to select at most a few dozen variables for an individualanalysis. These variables do not need to belong to the same part of the process plant, may be selected on a plant-widescale, and may include variables in the control system, like e.g. estimated variables. It should be noted that ProMoVisis not limited to the selected set of methods, and other analysis methods for interconnected systems can be added.The software is currently in use at several industry partners of the SCOPE consortium within ProcessIT Innovations,(ProcessIT Innovations, 2012), and is made available by the open-source project ProMoVis at Sourceforge, (OProVATEF, 2012).

The paper is arranged as follows. First, the interface for modeling and visualization is discussed and somenecessary notation is introduced. Thereafter, the implemented CCS methods are shortly summarized including theirusage, properties, and limitations. Then the stock preparation process of SCA Obbola AB is introduced as a casestudy. It is shown how the stock preparation process can be represented in ProMoVis and how the CCS task isperformed. Finally, the results from the CCS are compared with the currently implemented control strategy and arediscussed. The paper is concluded with some final remarks.

2. Application software ProMoVis

Selection of a control configuration for processes with many interconnections is facilitated by a systematic ap-proach, which is based on process knowledge in terms of dynamic models of the interconnected process. To theknowledge of the authors there is no software available which can visualize process variables including their dynamicinterconnections and control configuration analysis results in a comprehensive way. For this end, we now propose thesoftware ProMoVis, (Process Modeling and Visualization).

From a practical perspective, selection and assessment of a control strategy would require the following actionsby a practitioner:

1. Derive a dynamic model for the process

2. Select a set of manipulated and controlled variables (I/O selection)

3. Determine a controller configuration

4. Design of the individual controllers according to the configuration

5. Implementation of the controllers

6. Assessment of the control performance

2

(a) (b)

Figure 1: (a) Sketch of the quadruple tank process. (b) SFG for the quadruple tank process which contains informative labeling of thesignals. Red edges indicate the model interconnections, red nodes indicate the actuators (pumps), yellow nodes represent the disturbances(leakage flows), white nodes represent the internal states (level in the upper tanks) and green nodes represent the measurement signals(levels in the lower tanks).

For all actions, besides action three, there exist software tools that support the control engineer. For modelingof processes and control design, toolboxes in MATLAB, (The Mathworks, Inc., 2012), or multivariate analysis andmodeling tools from MKS Umetrics AB, Sweden (2012), are available. For the selection of I/O sets with manipulatedand controlled variables the tools from MKS Umetrics AB, Sweden (2012) can be used from a multivariate perspective,whereas the methods proposed in (Skogestad, 2000), address the problem from a feedback control perspective. For theimplementation of controllers in the control system there are tools proposed that support the automatic generation ofcontrol system code (Estevez et al., 2007) and (Vyatkin, 2012). Additionally, control systems provide standard blocksfor certain types of controllers, like for example the PID. Further, many industrial control systems possess online toolsto monitor the performance of control loops as part of the control system. The remaining gap is action three, whereProMoVis aims at providing support for CCS.

2.1. Software concept

In this section the required mathematical framework and notation is introduced and based on that the softwareconcept is explained.

The signal flow graph (SFG) was proposed by Mason (1953) to represent interconnected dynamic linear systems,where the nodes represent the signals and the edges elementary linear dynamic systems, and will be used as themathematical framework for the application software. Thus, the modeling task in ProMoVis reduces to the effort ofcollecting and combining information on the process plant and its control system. We will now state the algebraicform of the signal flow graph as given in (Johansson, 2010).

Let xi, i = 1, ..., p represent all exogenous signals, i.e. those variables that are not affected by any other variablesin the interconnected system and let zi, i = 1, ..., n be all other variables of interest. The models are assumed to beformulated as

zi = Φi1z1 + ... + Φinzn + Γi1x1 + ... + Γipxp (1)

for i = 1, ..., n where Φij and Γij are linear dynamic systems that may represent process model interconnectionsas well as controllers. The set of exogenous signals may include e.g. external disturbances and manipulated variablesbut also set points. When a control loop is closed using a manipulated variable xi and a variable zj , then xi willbecome an element in z and the associated set point variable will be introduced in x. Now, let us associate each signalxi and zi with a node, each Φij 6= 0 with an edge from zj to zi, and each Γij 6= 0 with and edge from xj to zi. Thenthe SFG is obtained as a graphical representation of the model interconnections. Moreover, by collecting the signalsxi and zi into vectors x and z and defining the multivariable, dynamic systems Φ and Γ whose i, jth element are Φij

and Γij respectively, the signal flow graph representation may now be formulated as (Johansson, 2010)

z = Φz + Γx (2)

In the example in Fig. 1 a process sketch (a) and a signal flow graph (b) of a quadruple tank (Johansson, 2000b)are depicted. While the process sketch provides information on the construction and the variables in the process, the

3

signal flow graph provides information on the dynamic interconnections. There, the exogenous inputs are the nodesd1, d2, u1, and u2 and constitute x = [d1, d2, u1, u2]T , while the nodes h1 to h4 are the measurement signals and theinternal states, which make up the vector z = [h1, h2, h3, h4]T . Therefore, Φ13, corresponding to the arrow from nodeh3 to h1, is a linear system modeling how the level in Tank 3 affects the level in Tank 1. Similarly, Γ43 is a model forhow Pump 1 affects the level in Tank 4, and so on.

In the SFG framework variables are the interface between dynamic models, and some of them constitute theinterface between process and control system. In ProMoVis, the process layout at the physical level is represented byinterconnected entities referred to as components. These components do not contribute to the dynamics of the plant,but provide important information on the geographical location of the process variables and how they relate to theprocess physics.

In Fig. 2, this concept is captured and depicted for the quadruple tank example. Naturally, one could think ofthree layers: components, process models, and controllers. In each of these layers, the process variables are visibleand represent the interface between the layers. This concept is very much in line with the industrial understanding ofa plant where process variables and their properties are the central element. Performance requirements for processesand product qualities are always related to variables that are measured online, estimated, or derived from laboratoryassessments. Therefore, components and process variables are the natural point to start modeling and visualizinga process, which is the component layer, similar to Fig. 1a. The process model layer then represents the dynamicinterconnections in the process, which is the same SFG as already shown in Fig. 1b. The controller layer representsthe dynamic interconnections in the control system, in this case an SFG of two SISO controllers for the quadrupletank with their associated set point variables r h1 and r h2.

A visualization can become very complex when all elements are visible at the same time, which might be ofinterest during composition or building, but unadvisable during analysis and decision making. In the latter case it isof interest to select certain information that should be visible, which can be achieved by the use of layers and theirvisibility. Such a complete representation of a plant in ProMoVis will be denoted a scenario.

2.2. Objects in ProMoVis

In ProMoVis a process plant including its control system is modeled using generic objects that are connected andarranged in different layers. There are four classes of objects: Variables, process models, controllers, and components.Process models, components, and controllers are collected in separated layers, which enable a differentiation of theview based on the class of the objects.

2.2.1. Variables

The variables represent the signals (nodes) in the SFG and can be divided into categories based on their character.For each category a color code is used in the interface in order to increase clarity for the user. Here, the default colorsetting is used but the user can reconfigure it.

Measured variables (green) represent the sensor input from the process into the control system.

User reference variables (blue) represent set points for controllers and can be interpreted as a manual setting by anoperator. As such, they are the interface between the operator and the control system.

Manipulated variables (red) represent the interface from control system to process. Usually, actuator signals aremanipulated variables.

Disturbance variables (yellow) represent exogenous disturbance signals, which may be induced by another processof the plant.

Estimated variables (orange) represent the result of a computation based on manipulated, controlled, or referencevariables.

Internal or state variables (white) represent all variables which do not belong to any of the previous categories.These represent internal variables of the process or the control system, which are of importance for the controlengineer.

Intermediate variables (white) are added automatically when two objects of the control system are connected withno interface variable. They are needed for the implementation of the SFG framework. They are considered asinternal variables but have no user defined properties.

4

Figure 2: Different layers in the modeling and visualization concept. Components (top), Process models (middle), Controllers (bottom).Manipulated variables (red), Measured or controlled variables (green), Reference variables (blue).

5

Table 1: Applicability of variable properties depending on the variable category

PropertyVariable type

Man

ipu

late

d

Mea

sure

d

Ref

eren

ce

Dis

turb

an

ce

Est

imate

d

Inte

rnal

Inte

rmed

iate

Range X X X XLimit X X X X XVariance X X X XSensor noise XOperating point X X X X XUser set value XDelay X X X X X

These categories are of importance as they determine how variables can be interconnected and how they interactwith the information in the layers. It is important to note that controlled variables are either measured or estimatedvariables. In the sequel, the term controlled variables is used when the variables can be either estimated or measured.

Variables have different process related properties that can be set by the user, see Table 1. Some of these propertiesform part of all the dynamic models which connect a specific variable. These properties are:

• Limit (Saturation), which determines the allowed operating range of a variable.

• Delay, which allows the user to define input or output delays.

The value of the delay is integrated into the process models during the analysis. The remaining properties allowthe user to specify process operating conditions which can be used for the scaling of the process variables during theanalysis.

2.2.2. Process models

The process models correspond to the edges of the SFG and are the interconnections between variables representingthe dynamic behavior of the plant. Generally, process models can be defined on a single-input-single-output basis, butmulti-input-multi-output models are supported as well. In both cases, a process model can be defined as a transferfunction or state space system in continuous or discrete time. When a process model is defined it is represented by ared edge, as shown in middle layer of Fig. 2.

In order to simplify adding process models, some model structures which are used within system identification ofprocess models are pre-defined, like for example

Γij(s) =K

Ts + 1e−Ls or Φij(s) =

K

Ts + 1e−Ls

where the user only has to provide the parameters K, T and L in order to define the dynamics.Currently, only linear time invariant models are supported. Clearly, a dependency on the operating points of the

different variables arises, but most available CCS methods are only applicable on linear models.It has to be noted that ProMoVis is an offline tool and does not derive the process models and their parameters.

This has to be done in a previous step by the user.

2.2.3. Controllers

In most cases, controllers do not differ from process models in their implementation. Single-input single-outputcontrollers can be represented by two edges, from reference and controlled variable to manipulated variable, seebottom layer in Fig. 2. Alternatively, Single-input single-output controllers can also be defined as blocks with twoinputs (reference variable, controlled variable) and one output (manipulated variable). Either way, the resulting edgesor blocks are then automatically generated. The reason is to simplify for users to create and connect controllersproperly and thereby to avoid incorrect connections. Similar to process models, some controller types are pre-defined,such as PID controllers and filters. The user can choose between the block or edge representation. Multivariablecontrollers can be defined with multiple input and output ports.

6

Figure 3: Software architecture of ProMoVis.

2.2.4. Components

The process layout of a plant at the physical level can usually be decomposed into smaller building blocks whichare components. These components can have a graphical representation which can be used to create a visualizationof the plant.

In ProMoVis, components have no functionality other than providing an understanding of the layout and con-struction of the plant with a rather coarse level of detail and realism. An effective representation of components canbe created by using symbols according to industry standards (see for example SSG Standard Solutions Group AB,2007a,b), or bitmap images of drawings or sketches.

For the design of symbols a simplistic script language is implemented that enables the user to create new sets ofsymbols and libraries. At the moment, there are sets of symbols available for the pulp and paper industry and miningindustry. The script language is mainly composed of drawing commands for lines, polygons, ovals, coloring, and text.ProMoVis will interpret the commands and then draw the component symbols accordingly.

2.3. Software implementation

Building a representation of an interconnected process does not require any intense computations. Additionally,the focus is on interactivity and a graphical user interface which is versatile and easy to use on any computer platform.CCS methods depend on many mathematical operations that have to be performed on the SFG.

Therefore it was decided to implement the modeling and visualization in Java and the computational engine inMATLAB. A schematic of the software architecture is shown in Fig. 3. There, it can be seen that the Java GUI isconfigured using configuration files. The information flow between the Java GUI and the computational engine islimited to the transfer of the model data, the analysis commands and the reporting of the result data back to the JavaGUI.

After startup, ProMoVis enters the building mode, where the user can create new scenarios or load existingscenarios from stored files. The user can then switch between building mode and analysis mode using menu commands.In the analysis mode, the user makes a selection of the analysis that should be performed and selects the parts ofthe scenario which should be considered. As soon as the analysis is called, the current model data is transferredto the computational engine where it is buffered until the user leaves the analysis mode. Additionally, the analysiscoordination is executing the necessary analysis functions. Thereafter the result arbitration will combine the resultsfrom the analysis functions and report them back to the result display in the Java GUI.

The interface between the Java GUI and the computational engine is well defined and enables the porting of theMATLAB code onto other platforms without significant changes to the Java GUI. For industrial use, it is possible tocombine the computational engine with the Java GUI into a stand-alone software.

7

Table 2: Available options for each method implemented in ProMoVis.

RGA DRGA NI PM HIIA Σ2 SET FETr FETc FDPTr FDPTc

Consideration of time delays X X X

Frequency options X X X

Scaling optionssaturation, range X X X X X X X Xinput scaling X X X X Xoutput scaling X X X X X

Filter options X X X

Plot type X

3. Analysis methods for the selection of control configurations

The goal is to select a set of Interaction Measures (IMs) which is sufficient to solve the CCS problem for mostof the cases. It is the belief of the authors that this includes traditional IMs like relative gains for the selection ofinput-output pairings, Niederlinski Index for testing the stabilizability of the resulting decentralized configurations, aswell as more modern gramian-based IMs which are used for the design of sparse control configurations.

We define now the selected CCS methods and discuss their implementation. A typical procedure for CCS usingIMs is described for the user of ProMoVis.

3.1. Implementation of the analysis tools

The implemented tools depend on the availability of accurate process models, which have to be derived prior tothe analysis.

When the method to be used is selected, the user is required to choose an input/output set for which the analysisis performed. In general, the inputs are restricted to be manipulated variables, however future consideration ofhierarchies will require including controller references in order to select higher level structures like the outer loops ofcascades. Depending on the selected method, a different set of options is available, with predefined default values.These options are grouped in the following subsets:

• Consideration of time delays. For those methods which are sensitive to time delays, the user can decide ifthese are considered in the computation. If so, the order of the Pade approximation has to be given for the caseof continuous-time systems.

• Frequency options. For those methods which result in an array of diagrams in the frequency domain, it isallowed to select the frequency unit, as well as the set of frequencies considered for analysis.

• Scaling options. Usual methods for scaling signals involve dividing each variable by its maximum expected orallowed change (Skogestad and Postlethwaite, 2005). For those methods which are sensitive to the scaling of theprocess variables, it is allowed to choose to scale the process variables by using the values entered in either theSaturation or the Range fields of the process variables. As an alternative, it is allowed to manually introduceinput and/or output scaling matrices depending on the method.

• Filter selection. For the gramian-based IMs, it is possible to restrict the analysis to a range of frequencies ofinterest, e.g. around the crossover frequency, which is where most of the control action is usually present. Thisis done by filtering the input-output channels such that frequencies outside the selected range are attenuated(Birk and Medvedev, 2003). In ProMoVis, such filters can be declared in the calculation options.

• Plot type. This option is exclusive of the Dynamic Relative Gain Array (DRGA), which results in a complexarray represented in the frequency domain. The user can choose to represent its magnitude, phase, real part orimaginary part.

The options which are available for each of the subsequently defined methods are summarized in Table 2.For the analysis methods described here, the transfer function matrix G(s) from the selected subset u of the

exogenous inputs into the selected subset y of the process outputs is required and will be derived now from (2).Provided that (2) is well-posed (see (Johansson, 2010) for details) we may infer that the variables z are related to theexogenous inputs x as

z = (I − Φ)−1Γx (3)

8

Now, let B be a matrix selecting the variables u from x, i.e. u = Bx. Then ΓBT will contain those columns fromΓ that correspond to u. Similarly, let C be a matrix selecting the variables y from z, i.e. y = Cz. Then, for thecontinuous-time case transfer function matrix from u to y is

G(s) = C(I − Φ(s))−1Γ(s)BT (4)

In ProMoVis, the calculation (I−Φ(s))−1Γ(s) = G0(s) is done only once in order to reduce computational effort.Selecting different sets of inputs and outputs, i.e. multiplication by different C and BT is then accomplished by pickingout the appropriate rows and columns from G0(s).

After this computation, the selected method is applied to G(s) and the result is appropriately displayed.

3.2. Analysis tools based on relative gains

The most popular tool based on relative gains is the RGA, introduced by Bristol (1966) to design decentralizedcontrol configurations based on steady-state gain information. Later, several authors addressed some of the limitationsof the RGA, usually by introducing variants of this IM. This includes different extensions of the RGA to considerprocess dynamics, like evaluating the RGA at different frequencies by Witcher and McAvoy (1977), which was namedDynamic RGA (DRGA).

In the default set of CCS methods in ProMoVis, the RGA and DRGA methods have been implemented for thedesign of decentralized control configurations as well as the Niederlinski Index for discarding unstable configurations.Other advanced techniques based on relative gains are candidates for future versions of ProMoVis, like the Block RGAintroduced by Manousiouthakis et al. (1986) for the design of block diagonal control structures and the partial relativegains introduced by Haggblom (1997) for the selection of sparse control configurations.

3.2.1. Relative Gain Array (RGA)

The RGA of a continuous process described by (2) and with input-output transfer function G as in (4) is:

RGA(G) = G(0)⊗G(0)−T (5)

where ⊗ denotes element by element multiplication, and G(0)−T is the transpose of the inverse of the steady-stategain matrix. The normalization used in this calculation implies that the sum of all the elements in the same row orcolumn of the RGA add up to 1.

Each of the values of the RGA is the steady-state gain of the corresponding input-output channel when all theother loops are open divided by the steady-state gain when the rest of the process is in closed loop under tightcontrol. Based on this definition, the following rules have been formulated for the selection of a decentralized controlconfiguration:

• The preferred pairings are those with RGA values close to 1 (Skogestad and Morari, 1992).

• The selection of positive values for the decentralized pairing is a necessary condition for closed-loop integrity,provided that all elementary subsystems are linear time invariant, finite dimensional, stable, and strictly proper(Campo and Morari, 1994). Integrity is a desirable property of the decentralized control system, which meansthat the closed-loop system should remain stable as each of the SISO controllers is brought in and out of service(Bristol, 1966). This is not applicable to time delayed systems due to their infinite dimensional aspect.

• Large values should not be selected since they are related to ill-conditioned behavior of the plant (Chen et al.,1994). Values exceeding 1 by more than a few tenths are very sensitive to model uncertainty and the nominalvalue can be easily perturbed to a large value, as indicated in the studies on 2× 2 systems by Castano and Birk(2008).

Note that these properties imply that the RGA might not indicate any appropriate decentralized control configuration,requiring other tools to design configurations. Moreover, the RGA is insensitive to input and output scaling and totime delays.

In addition, the RGA has certain limitations which need to be considered. Several of these limitations have beenresolved by different authors, and some of these solutions have been implemented in ProMoVis, like the applicationto non-square plants with the use of the pseudo-inverse (Chang and Yu, 1990), or the computation of the RGA forsystems with pure integrators (Arkun and Downs, 1990; McAvoy, 1998).

An important limitation is that the RGA is originally evaluated only at steady state, and therefore is not reflectingthe dynamic properties of the process.

9

3.2.2. Dynamic RGA (DRGA)

The DRGA of a continuous process described by (2) and with input-output transfer function G as in (4) is:

DRGA(ω) = G(jω)⊗G(jω)−T (6)

The DRGA is an array of complex numbers and has a more obscure interpretation than that of the RGA. Usually,it is preferred to use its magnitude as indicator due to the gain interpretation, however only the sums of the rows orcolumns of the resulting complex array (or its real part) add up to 1. Moreover, by evaluating the magnitude alone,the sign of the DRGA is lost as an indicator, which is often used to rule out certain input-output pairings.

A shortcoming of the DRGA is that perfect control for all frequencies is assumed in its computation. Thisassumption is only valid for a very low frequency range. Other dynamic versions of the RGA have been defined toovercome this situation, like the Effective RGA (ERGA) introduced by Xiong et al. (2005). Nevertheless, the DRGAversion implemented here has been selected for its simplicity and widespread use.

3.2.3. Niederlinski Index (NI)

For a system under decentralized control, and assuming that the process is described by (2) and with input-output transfer function G as in (4) which has been reordered so that the controller is a diagonal matrix, the NiederlinskiIndex (NI) can be computed as (Niederlinski, 1971):

NI = det(G(0))/n∏

i=1

Gii(0) (7)

This indicator is traditionally used to test the stabilizability and/or integrity of a decentralized configuration.Under the assumptions of stability of all the elementary subsystems represented by rational functions Gij(s), a

value of NI < 0 is a sufficient condition for the instability of the closed loop system when all the SISO loops areunder integral action. This condition is widely used for discarding unstable decentralized control structures prior tothe design of the multi-loop controller.

Integrity can be verified by testing the stabilizability of the systems which result from opening each of the SISOloops. This is done by computing the value of NI for any of the principal sub-matrices Gii(0) resulting from removingthe ith row and column from G(0). The system will not possess integrity if NI < 0 for any of the principal sub-matrices.

More restrictive conditions for stability and/or integrity exist like the tighter conditions derived by Chiu andArkun (1991) for 2× 2 plants. However, these tests are more complicated, and the reader can refer to the surveys inChapter 10 by Skogestad and Postlethwaite (2005) or Chapter 2 by Khaki-Sedigh and Moaveni (2009).

3.3. Gramian-based IMs

For the design of control configurations other than decentralized, the modern gramian-based IMs can be used.The gramian-based IMs are Index Arrays (IAs) in which a gramian-based operator is applied to each of the

single-input single-output subsystems in order to quantify its significance. The use of different operators results indifferent IMs. The Hankel Interaction Index Array (HIIA) introduced by Wittenmark and Salgado (2002) uses theHankel norm. The Participation Matrix (PM) introduced by Salgado and Conley (2004) uses the trace of the productof controllability and observability gramians, and the Σ2 introduced by Birk and Medvedev (2003) uses the H2 norm.

For a continuous process described by (2) and with input-output transfer function G as in (4), the IAs arecalculated as:

[IA]ij =[Gij(s)]p

m,n∑

i,j=1

[Gij(s)]p

(8)

where [·]p denotes the corresponding operator of the used IA.As a result of the normalization, all the elements of any of these IAs add up to one. The selection of the control

configuration is made by selecting a subset of the most important input-output subsystems, which will form a reducedmodel on which control will be based. Choosing a configuration with a total contribution of the selected input-outputchannels larger than 0.7 is likely to result in satisfactory performance (Salgado and Conley, 2004).

An advantage of the gramian-based IMs over the RGA is their ability to be used for designing sparse controlconfigurations. A disadvantage is that the quantification of the significance of the input-output subsystems dependson the scales used to represent the inputs and outputs.

10

At the moment there is no clear procedure for interpreting the gramian-based IMs in the presence of time delays.For the case of the operators used by the HIIA and the PM, the quantified significance of an input-output channelincreases as the channel delay increases (Castano and Birk, 2012). This might result in inadequate configurations,since channels exhibiting large time delays but low gain and bandwidth might end up forming part of the reducedmodel. This was revealed in Halvarsson (2008), where simulation work indicated that the presence of a time delayitself is not sufficient for saying that a particular input-output channel should be used in the controller. Due to thisproperty of the PM and HIIA, it was decided that the user can select if the time delays will be neglected or not in thecomputation. No decision needs to be taken in the case of using Σ2 due to its insensitivity to time delays.

3.4. Methods for structural analysis using weighted graphs

ProMoVis is able to visualize analysis results together with the process, e.g. as an overlayed weighted directedgraph that shows the significance of the connections as the thickness of the edges. For this purpose, the analysismethods described by Castano and Birk (2012) have been implemented: SET and FET.

These methods use the squaredH2 norm as operator for quantifying the significance of the process interconnectionsin terms of signal energy transfer.

3.4.1. Structural graphs.

The method Structural Energy Transfer (SET) is applied to obtain a weighted structural graph describing theimportance of the direct process interconnections.

Structural graphs have been extensively used for the design of control structures, and the work in Nistazakis andKarcanias (2004) describes its importance for deriving properties such as decomposability (Sezer and Siljak, 1986) andstructural controllability and observability (Lin, 1974).

The novelty of the method SET is adding weights to these structural graphs. This gives an enhanced visualunderstanding of the process and allows application of advanced methods for CCS which consider weighted graphssuch as described by Johansson (2000a).

3.4.2. Functional graphs.

In Functional Energy Transfer (FET) a normalized weighted directed graph is derived for the input-output chan-nels, which quantifies their significance. Two different normalizations are used such that either the weights of theedges entering an output node or the weights of the edges leaving an input node add up to 1. These normalizationsare denoted as FETr and FETc, and for a process described by (2) and with input-output transfer function G as in(4) they are calculated as:

[FETr]ij =||Gij ||22n∑

l=1

||Gil||22; [FETc]ij =

||Gij ||22m∑

k=1

||Gkj ||22(9)

For each output, the relative effect of the selected process inputs is described by FETr. For each input, the relativeeffect on the selected outputs is described by FETc.

It should be noted that FETr is insensitive to output scaling, and FETc is insensitive to input scaling.Several case studies indicated the usefulness of FETr in CCS. A controlled variable should be associated with the

minimum number of actuators which result in a value of the sum of their contributions (edge widths) larger than adesigned threshold. Previous work indicates that a value larger than 0.7 should be achieved in order to expect a wellbehaving closed loop system (Castano and Birk, 2012).

These measures can also be assessed in the frequency domain resulting in a function of frequency instead of a scalarnumber for each edge. This is done by normalizing the squared magnitude of each of the input-output interconnectionsso either all the edges entering a node add up to one or all edges leaving a node add up to one. These operations resultin the methods named FDPTr and FDPTc. For a process described by (2) and with input-output transfer functionG as in (4), FDPT is calculated as:

[FDPTr(ω)]ij =|G(jω)ij |2n∑

l=1

|G(jω)il|2; [FDPTc(ω)]ij =

|G(jω)ij |2m∑

k=1

|G(jω)kj |2(10)

11

3.5. Tools for the reconfiguration of control structures

A control deficiency could be the consequence of a tuning deficiency or of a structural deficiency in the controller.In the latter case, a redesign of the control structure should be done, preferably by adding or removing a minimalamount of controller interconnections.

The development of tools which identify if there is a structural deficiency in the controller and suggest a redesigningof the controller configuration has only recently received attention. A method was proposed by Birk (2007), that makesuse of the factorization of the closed loop sensitivity function matrix and has been implemented in ProMoVis. Thismethod quantifies the performance loss due to neglected interconnections in the process and considers the currentlyused controller. The method was further analyzed and assessed in a comparative study in (Birk and Dudarenko, 2012).

An appropriate output scaling is required for the application of the method, which is limited to control systemswith decentralized or block diagonal control structures.

In ProMoVis, this method can be used if a 1-DOF controller is used and the parameters of the controller aredeclared.

3.6. Typical procedure for CCS using IMs

The following procedure can be used for selecting control configurations based on the IMs.

Step 1. Seek a decentralized control structure using methods based on relative gains. If a decentralized structure isindicated by the use of the RGA as described below, then the DRGA will help to determine if the structure isstill feasible at other frequencies different to steady-state. The value of the DRGA at the crossover frequency isof special interest, since it is usually the range of frequencies at which control is more active.

Step 2. Check the stabilizability of candidate decentralized configurations. Decentralized structures with negativevalues of NI must be discarded for being unstable under integral action in all the SISO loops. Several othertests for stability and/or integrity of the decentralized control structure using NI and the RGA can be used.The reader can refer to Chapters 10 and 2 respectively in the books by Skogestad and Postlethwaite (2005) andKhaki-Sedigh and Moaveni (2009) for surveys on these tests.

Step 3. Design a sparse control configuration if needed. It is recommended to contrast the indications obtained usingrelative gains with other CCS methods. One reason is that the RGA might indicate severe loop interaction ifa decentralized structure is to be used. Another reason is that there might be severe loop interaction whichis not captured by the RGA, i.e. in triangular plants. These cases present severe difficulties for decentralizedconfigurations, and the gramian-based IMs can then be used to design a sparse control configuration. As analternative to the gramian-based IMs, the method FETr can be used, which provides a visual and intuitiveanalysis as well as being insensitive to the scaling of controlled variables.

4. Case study: A stock preparation process.

The stock preparation process in SCA Obbola AB, Sweden is described below and will be used as illustrativeexample for the typical work flow with ProMoVis. At the moment of the described work, the plant was operating witha decentralized controller under stable conditions but exhibiting significant perturbations in the controlled variables.The case study will therefore be considered a success if an analysis with ProMoVis indicates the same decentralizedconfiguration as feasible, and gives insight in potential modifications on the control structure relating to the deficiencies.

Prior to the use of ProMoVis, process information has to be acquired in the form of mathematical models and/orprocess flow charts. First the process model is implemented in ProMoVis by creating a visual representation of theflow charts and declaring the mathematical process models. Then the control structure can be selected using theimplemented methods.

4.1. Description of the stock preparation process.

The stock preparation process is present in many paper plants for the refining of pulp and chemical treatment. Inconventional refining, the pulp is pumped through the gap between two grooved discs. A moving disc can be rotatedand displaced in the axial direction, and the friction of the fibres with the discs and with each other creates the refiningeffects. Refining creates major changes in pulp properties as described by Annergren and Hagen (2004). Its goal is toimprove web strength, but also results in decreasing the dewatering capacity of the paper web and thus needs to betightly controlled for optimum results.

12

Figure 4: Schematics of the stock preparation process at SCA Obbola. Pipes with indicated flow directions are wide solid lines, descriptionsor component names are in italics and variables are in capital letters. Symbols are in accordance with the SSG standard.

A schematic of the process is depicted in Fig. 4. First the pulp is pumped from a storage tank and the flowbifurcates towards two parallel refiners. Note that a fraction of the pulp is recirculated again for balancing themechanical load in the refiners. The pulp is then diluted to the required concentration for the chemical treatment,being finally discharged to a storage tank, in which starch is added, and from where the pulp is pumped to subsequenttanks to continue with the chemical treatment. The structural complexity of the process requires a deep analysis ofthe process interconnections in order to design a control configuration. The set of considered sensors and actuators issummarized in Table 3.

The refiners have internal controllers to track a set point for the specific energy that is used to affect the pulp.Safety, quality, and production depend on well-maintained set points for the considered flows and the pressure atthe entrance of the refiners. In the current control of the process, four independent single-input single-output PIDcontrollers are used to maintain the flows at the desired operating points. The centrifugal pump is then used asactuator in another control loop to keep the pressure before the refiners at the operating point. The dilution water isdelivered to each of the branches with the use of cascade structures, which have as outer loops the desired concentrationfor the pulp, and as inner loops the needed flow of pulp to achieve such concentration. In both branches, the pressureat which the pulp is discharged to the storage tank is controlled by a valve with a PID controller.

4.2. Implementation of the stock preparation process model in ProMoVis

The visual representation resulting from implementing the stock preparation process in ProMoVis is depictedin Fig. 5. First, a visualization of the process layout at the physical level was created by connecting componentsrepresenting elements such as pipes, valves, pumps, and refiners. Secondly, the corresponding process variables weredeclared.

In order to collect significant process data for the modeling task, the process was excited during normal operationby perturbing the actuators with additive white noise. In a first modeling step, a model structure was created byselecting a subset of controlled variables and actuators to be considered for control, and identifying which actuatorsgenerate an observable impact on certain controlled variables. System identification techniques were used to modelthe input-output channels reflected by the identified model structure, and the resulting transfer functions of the

13

Table 3: Considered sensors and actuators in the refining section.Actuators

Tag Name Description

PA Pump Actuator Pumps the flow through the refiners

VA1 Valve Actuator 1 Valve after refiner 1

VA2 Valve Actuator 2 Valve after refiner 2

VA3 Valve Actuator 3 Valve at the recirculation from refiner 2

VA4 Valve Actuator 4 Valve at the recirculation from refiner 1

VA5 Valve Actuator 5 Valve at the dilution for the pulp form refiner 1

VA6 Valve Actuator 6 Valve at the dilution for the pulp form refiner 2

VA7 Valve Actuator 7 Valve before discharge to the storage tank(branch from refiner 1)

VA8 Valve Actuator 8 Valve before discharge to the storage tank(branch from refiner 2)

Sensors

Tag Name Description

PI1 Pressure Indicator 1 Pressure before the flow bifurcation

PI2 Pressure Indicator 2 Pressure at the entrance of refiner 1

PI3 Pressure Indicator 3 Pressure at the output of refiner 1

PI4 Pressure Indicator 4 Pressure at the entrance of refiner 2

PI5 Pressure Indicator 5 Pressure at the output of refiner 2

PI6 Pressure Indicator 6 Discharge pressure before the storage tank(branch from refiner 1)

PI7 Pressure Indicator 7 Discharge pressure before the storage tank(branch from refiner 2)

FI1 Flow Indicator 1 Pulp flow through refiner 1

FI2 Flow Indicator 2 Pulp flow through refiner 2

FI3 Flow Indicator 3 Pulp flow recirculated from refiner 2

FI4 Flow Indicator 4 Pulp flow recirculated from refiner 1

FI5 Flow Indicator 5 Dilution water for pulp from refiner 1

FI6 Flow Indicator 6 Dilution water for pulp from refiner 2

CI1 Concentration Indi-cator 1

Concentration before the flow bifurcation

CI2 Concentration Indi-cator 2

Concentration before the flow bifurcation

Estimated Variables

Tag Name Description

CE1 Concentration Esti-mation 1

Average of two redundant concentration sensorsbefore flow bifurcation.

CE2 Concentration Esti-mation 2

Concentration of pulp to be diluted after refiner1

CE3 Concentration Esti-mation 3

Concentration of pulp to be diluted after refiner2

FE1 Flow Estimation 1 Flow of pulp from refiner 1 which is not recir-culated

FE2 Flow Estimation 2 Flow of pulp from refiner 2 which is not recir-culated

interconnections in the model of the stock preparation prcess are summarized in Table 4. Each of the obtainedtransfer functions was declared in ProMoVis, resulting in red interactive edges represented in Fig. 5, which can beused to access and edit the parameters of the associated process model.

Finally, the controllers representing the current control of the process were defined in order to visualize andmaintain the information on the control system.

Notice that controlled variables can be either measured or estimated. The distinction is used in order to makethe user aware of the fact that estimated variables are the result of a calculation in the control system, represented byobservers. Therefore, measured variables may only be connected to other process variables, while estimated variablesmay be connected to variables in the control system as well.

CE1 is the average of two redundant concentration sensors. FE1 and FE2 are the flows of pulp before adding thedilution water, and they are computed as the difference between the flow through the refiners and the recirculationflow. CE2 and CE3 are the concentrations of pulp before the dilution; they are the controlled variables of the outerloops in the cascades to control the addition of dilution water, and they are estimated as being the concentration ofpulp before the refiners with a transport delay which depends on the flow of pulp before adding the dilution water.

Note that reference variables can be part of a control loop referring to an operating point for a controlled variable,or the manual setting of an actuator. As an example, in the pressure control loops actuating VA7 and VA8 in Fig. 5,the user can switch from manual to automatic mode. The position of the switches determine different operationalmodes for the analysis.

4.3. Analysis of the stock preparation process with ProMoVis

A control configuration for the stock preparation process will now be selected using ProMoVis.The existing controllers of the process for the pressures PI6 and PI7 were causing large oscillations during the

experiment. For this reason, the valves VA7 and VA8 were manually placed at a certain opening during most of theexperiments, which means that the collected data was not informative enough to create models which include thesevariables. Therefore further experiments need to be conducted in order to include those variables in the CCS problem.

14

Figure 5: ProMoVis screenshot. Refining section of the stock preparation process at SCA Obbola.

15

Table 4: Transfer functions of the interconnections in the model for the stock preparation process.

Input-Output Transfer function

PA - PI1 2.8961/(5.8279s2 + 2.965s + 1)

VA1 - PI1 −0.54313/(2.9712s + 1)

VA2 - PI1 −0.87994/(0.92472s + 1)

VA1 - FI1 1.5359/(42.331s + 1)

VA2 - FI1 0.40553/(20.046s + 1)

VA1 - FI2 0.35223/(18.8525s + 1)

VA2 - FI2 1.8979/(29.6601s + 1)

VA3 - FI3 0.24843/(4.5058s + 1)

VA4 - FI3 −0.019761/(2.1097s + 1)

VA3 - FI4 −0.042479/(4.7492s + 1)

VA4 - FI4 −0.20199/(2.018s + 1)

VA5 - FI5 4.3294/(2.9661s + 1)

VA6 - FI6 0.35864/(15.9248s + 1)

As for the dilution water, it is trivial from the structure of the determined model, that the pairings VA5-FI5 andVA6-FI6 have to be selected.

Our problem is now reduced to find a control configuration for the sensor/actuator set in Table 5.

Table 5: Input-output set to be considered for analysis.

Actuators PA VA1 VA2 VA3 VA4

Sensors PI1 FI1 FI2 FI3 FI4

Applying the RGA to the selected input-output set yields the numbers in Table 6. According to the pairing rules,the diagonal pairing of inputs and outputs is preferred for decentralized control.

Table 6: Result of the RGA analysis of the stock preparation process.

Output

InputPA VA1 VA2 VA3 VA4

PI1 1 0 0 0 0FI1 0 1.05 -0.05 0 0FI2 0 -0.05 1.05 0 0FI3 0 0 0 1.02 -0.02FI4 0 0 0 -0.02 1.02

Using ProMoVis, the value of NI for the selected pairing was calculated to be NI = 0.935.This steady state analysis of the process indicates that a decentralized control structure is likely to lead to

acceptable performance.To complement this information, one or several of the gramian-based IMs could be used. The result of applying

the PM to the stock preparation process is given in Table 7.There, the sum of the diagonal elements in the PM is 0.950, which means that a decentralized configuration

will consider 95% of the system dynamics. However, the user should at this stage be aware of the fact that thegramian-based IMs are sensitive to input-output scaling. The manipulated variables are expressed in % of actuatoropening. However, the measured variables are represented in different scales, since there are both pressure and flowmeasurements, and also the recirculation flows (FI3 and FI4) are only a fraction of the primary flows (FI1 and FI2).This is probably the reason why the input-output channels related to the pressure measurement have a rather highsignificance, whilst the ones associated with the recirculated flows have a rather low one. A possible remedy is to scalethe process variables using their observed range of variation.

Another possible approach is to use the recently introduced method FETr, which is independent of output scaling.The result of applying FETr is depicted in Fig. 6. The most significant edges entering a measured variable identifythe actuators which can deliver the highest energy contribution on the measured variable. An optional threshold of0.1 on the significance of the edges determines their visibility and thereby the graphical representation is simplified.

By inspecting Fig. 6, and pairing each of the measured variables with the actuator connected with the mostsignificant edge, it is clear that the best decentralized control structure suggested by FETr, is the same as the one

16

Table 7: Result of the analysis using the Participation Matrix of the stock preparation process.

Output

InputPA VA1 VA2 VA3 VA5

PI1 0.724 0.011 0.029 0 0FI1 0 0.088 0.006 0 0FI2 0 0.005 0.134 0 0FI3 0 0 0 0.002 1.5 · 10−5

FI4 0 0 0 6.7 · 10−5 0.002

Figure 6: Combined ProMoVis result displays. Analysis of the stock preparation process with the method FETr. Either a graph, or theconnectivity matrix related to the graph can be chosen as displayed result. The layers including the components, the process models, andthe controllers with their corresponding references are selected as not displayed.

suggested by the RGA, and coincides with the one currently used in the process. Nevertheless, it is suspected thatthere exist a potential for improving the control performance by considering the dynamic connection from VA2 to PI1in the control system, since this will increase the sum of contributions on PI1 from 0.7153 to 0.9737.

To obtain a deeper insight in the effects on PI1, the tool FDPTr is applied using ProMoVis, and the resultis depicted in Fig. 7. This tool results in a frequency domain description of the relative power contribution of theactuators on a given measured variable. At each frequency, the sums of all the contributions on a measured variableadd up to one. It can be observed that the contribution from VA2 has a significant impact at frequencies around themaximum crossover frequency of the considered channels, causing interaction between the control loops which mayresult in oscillations. This conclusion is supported by the following facts: (i) the centrifugal pump has rotor dynamicswhich are slower than the dynamics of the valve; (ii) by the observations of the plant operators and engineers, whichconfirm the existence of the mentioned oscillations.

A potential of improving the existing control configuration has therefore been identified. The suggestion is toconsider the actuator-sensor connection from VA2 to PI1 in the control configuration, e.g. by adding a decouplingpre-compensator which compensates for the affect of VA2 on PI1.

10−2

100

102

0

0.5

1

PI1

PA

rad/sec10

−210

010

20

0.5

1 VA1

rad/sec10

−210

010

20

0.5

1 VA2

rad/sec

Figure 7: ProMoVis result display of FDPTr. The tool FDPTr describes the contributions on PI1 from the actuators in the frequencydomain. The largest crossover frequency of all the considered actuator-sensor channels is marked by a dashed line.

17

5. Conclusions

In this paper we have presented a new application software for control configuration selection in interconnectedsystems. The software is designed to support control engineers in the selection of a control configuration for a process.At the same time, information on the process and the control system can be maintained together which simplifiesthe effort to keep process control systems updated with the processes. This is achieved by combining (i) a graphicalrepresentation of the process layout at the physical level, (ii) a directed graph that represents the process dynamics andcontrollers and, (iii) control configuration analysis tools, into one and the same user interface. Additionally, analysisresults can be depicted in the same view as the visualization of the process and control system.

The software concept has been described and many implementation aspects of the visualization and the CCS toolsare discussed. The architecture of ProMoVis has been chosen to facilitate the implementation of additional analysismethods. ProMoVis features traditional tools for CCS as well as methodologies for the selection of sparse controlstructures and the visualization of them.

Moreover, a case study is presented where the refining stage in a stock preparation process is visualized andthe CCS task is performed using ProMoVis. It is demonstrated that ProMoVis confirms the control configurationcurrently in use, which achieves a sufficient level of performance and that it indicates modifications that could lead tofurther improvement of the process performance.

It can be concluded that ProMoVis is useful in industry in its current version and can be further enhanced byimplementing new research results, which makes these new results quickly available for industrial evaluation.

Finally, it should be mentioned that ProMoVis is now available under the Apache Open Source license to keepthe software updated with additional and new CCS tools.

Annergren, G., Hagen, N., 2004. The Ljungberg Textbook. Chapter 34 Industrial Beating/Refining. Department ofPulp and Paper Chemistry and Technology, KTH.

Arkun, Y., Downs, J., 1990. A general method to calculate input-output gains and the relative gain array for integratingprocesses. Computers chem. Engng 14 (10), 1101–1110.

Birk, W., 2007. Towards incremental control structure improvement. In: Proc of 46th IEEE Conference on Decisionand Control, New Orleans, December 2007.

Birk, W., Dudarenko, N., 2012. Evaluation of two methods for reconfiguration of multivariable controllers in processcontrol systems. In: Proc of 2012 IEEE Multiconference on Systems and Control, Dubrovnik, Croatia. pp. 627 –633.

Birk, W., Medvedev, A., 2003. A note on gramian-based interaction measures. In: Proc. of European Control Confer-ence ECC, Cambridge, UK. pp. 2660–5.

Bouchlaghem, D., Shang, H., Whyte, J., Ganah, A., 2005. Visualisation in architecture, engineering and construction(AEC). Automation in Construction 14, 287–295.

Bristol, E., January 1966. On a new measure of interaction for multivariable process control. IEEE Transactions onAutomatic Control 11, 133–134.

Browning, T. R., Ramasesh, R. V., 2007. A survey of activity network-based process models for managing productdevelopment projects. Production and Operations Management 16 (2), 217–240.

Campo, P., Morari, M., 1994. Achievable closed-loop properties of systems under decentralized control: conditionsinvolving the steady-state gain. IEEE Transactions on Automatic Control 39 (5), 932–943.

Castano, M., Birk, W., 2008. A new approach to the dynamic RGA analysis of uncertain systems. In: Proc of 2008IEEE Multi-conference on Systems and Control, San Antonio, USA, September 3-5, 2008. pp. 365 – 370.

Castano, M., Birk, W., 2012. New methods for interaction analysis of complex processes using weighted graphs. Journalof Process Control 22 (1), 280 – 295.

Chang, J.-W., Yu, C.-C., 1990. The relative gain for non-square multivariable systems. Chemical Engineering Science45 (5), 1309 – 1323.

18

Chen, J., Freudenberg, J. S., Nett, C. N., 1994. The role of the condition number and the relative gain array inrobustness analysis. Automatica 30 (6), 1029–1035.

Chiu, M.-S., Arkun, Y., 1991. A new result on relative gain array, niederlinski index and decentralized stabilitycondition: 2× 2 plant cases. Automatica 27 (2), 419 – 421.

El-Halwagi, M. M., 2006. Process Integration. Academic Press.

Estevez, E., Marcos, M., Orive, D., 2007. Automatic generation of PLC automation projects from component basedmodels. International Journal of Advanced Manufacturing Technology 35, 527–540.

Haggblom, K. E., 1997. Control structure analysis by partial relative gains. In: Proc. of the 36th Conference onDecision & Control, San Diego, USA. pp. 2623–2624.

Halvarsson, B., 2008. Comparison of some gramian based interaction measures. In: IEEE International Conference onComputer-Aided Control Systems (CACSD), 2008. pp. 138–143.

Jiang, Z.-Q., Zhou, W.-X., Xu, B., Yuan, W.-K., 2007. Process flow diagram of an ammonia plant as a complexnetwork. AIChE 53 (2), 423–428.

Johansson, A., 2010. Commuting operations on signal-flow graphs for visualization of interconnected systems. In:Proceedings of the 4th IFAC Symposium on System, Structure and Control in Ancona, Italy, Sept. 2010.

Johansson, K.-H., 2000a. Control structure design in process control systems. IFAC World Congress.

Johansson, K.-H., 2000b. The Quadruple-Tank Process : A Multivariable Laboratory Process with an AdjustableZero. IEEE Transactions on Control Systems Technology 8 (3), 456–465.

Khaki-Sedigh, A., Moaveni, B., 2009. Control Configuration Selection for Multivariable Plants. Lecture Notes inControl and Information Sciences. Springer.

Lin, C.-T., 1974. Structural controllability. IEEE Transactions on Automatic Control 19 (3), 201–208.

Manousiouthakis, V., Savage, R., Arkun, Y., 1986. Synthesis of decentralized process control structures using theconcept of block relative gain. AIChE Journal 32 (6), 991–1003.

Mason, S. J., 1953. Feedback theory - Some properties of signal flow graphs. Proceedings of the IRE, 1144–1156.

McAvoy, T., 1998. A methodology for screening level control structures in plantwide control systems. Computers andChemical Engineering 22 (11), 1543 – 1552.

MKS Umetrics AB, Sweden, 2012. MODDE and SIMCA product families. www.umetrics.com.

Niederlinski, A., 1971. A heuristic approach to the design of linear multivariable interacting control systems. Auto-matica 7, 691–701.

Nistazakis, M., Karcanias, N., 2004. Interaction analysis toolbox: Control structure selection based on process controlmethodologies. White Paper.

OProVAT EF, 2012. ProMoVis Open Source Project. sourceforge.net/projects/promovis/.

ProcessIT Innovations, 2012. Collaboration Centre ProcessIT Innovations in Northern Sweden.www.processitinnovations.se.

Rijnsdorp, J., 1965. Interaction on two-variable control systems for distillation columns. Automatica 1 (1), 15–28.

Rohrer, M. W., 2000. Seeing is believing: The importance of visualization in manufacturing simulation. In: Joines, J.,Barton, R., Kang, K., Fishwick, P. A. (Eds.), Proceedings of the 2000 Winter Simulation Conference.

Salgado, M. E., Conley, A., 2004. MIMO interaction measure and controller structure selection. Internation Journalof Control 77 (4), 367–383.

19

Sezer, M. E., Siljak, D. D., 1986. Nested ε-decompositions and clustering of complex systems. Automatica 22 (3),321–331.

Skogestad, S., 2000. Plantwide control: the search for the self-optimizingcontrol structure. Journal of Process Control10 (5), 487–507.

Skogestad, S., Morari, M., 1992. Variable selection for decentralized control. Modeling, Identification and Control13 (2), 113–125.

Skogestad, S., Postlethwaite, I., 2005. Multivariable Feedback Control - Analysis and Design. John Wiley & Sons.

SSG Standard Solutions Group AB, May 2007a. SSG5269E: Symbols for simplified process flow diagrams. Standardsdocument available at www.ssg.se.

SSG Standard Solutions Group AB, May 2007b. SSG5270E: Graphic symbols for process flow diagrams. Standardsdocument available at www.ssg.se.

The Mathworks, Inc., 2012. MATLAB/Simulink Product Family. www.mathworks.com.

van de Wal, M., de Jager, B., 2001. A review of methods for input/output selection. Automatica 37, 487–510.

Vyatkin, V., 2012. IEC 61499 Function Blocks for Embedded and Distributed Control Systems Design. ISA.

Witcher, M., McAvoy, T., 1977. Interacting control systems: Steady-state and dynamic measurement of interaction.ISA Transactions 16 (3), 35–41.

Wittenmark, B., Salgado, M. E., 2002. Hankel-norm based interaction measure for input-output pairing. In: Proc. ofthe 2002 IFAC World Congress, Barcelona.

Xiong, Q., Cai, W.-J., He, M.-J., 2005. A practical loop pairing criterion for multivariable processes. Journal of ProcessControl 15 (7), 741 – 747.

20