------------------------------------------------------ Rapport fra DYCOPS'07, 06-08 June 2007, Cancun, Mexico ------------------------------------------------------ The proceedings for the conference are at the folllowing location: http://www.nt.ntnu.no/users/skoge/prost/proceedings/dycops2007-and-cab2007/ Go to DYCOPS_intro.pdf and press Program. Then you get the program and press on the individial contributions. Dette er ikke noen skikkelig rapport, men mest noen notater jeg har samlet på en blanding av norsk og engelsk. Oppsummert var dette en bra konferanse med svært gode muligheter til å møte folk. Passe strørrelse. Det var bra oppmøte av folk innen prosessrefulering - spesielt fra Europa. Det var overraskjende få fra USA tatt i betraktning av at Cancun er kun 2 timer fra USA. DYCOPS IPC chair: Bjarne Foss, NOC Chair: Jesus Alvarez 1 felles dag med bio-konferansen CAB (onsdag). DYCOPS: 185 papers submitted. 162 accepted (84%). 2.6 reviews per paper --- DYCOPS deltagere: det var ca. 200 deltagere (men jeg er litt usikker for vi fikk ikke noen deltagerliste). Norge meget godt representert med minst 12 delegater + diverse partnere og barn. Derimot var det kun 1 fra Sverige og 0 fra Finland og 0 fra Danmark. NTNU Kyb: Professorene Bjarne Foss, Morten Hovd og Ole Morten Aamo NTNU IKP: Sigurd Skogestad (prof.), Elvira Aske (PhD-stip.), Sridhar Narasimhan (postdoc) Cybernetica: Tor Steinar Schei Høgskolen i Østfold/ Prediktor: Helge Mordt SINTEF: Finn Are Michelsen IRIS Petroleum, Bergen: Geir Nævdal, Gerhard Nygaard Høgskolen Telemark: Beathe Furenes (PhD stipendiat) --- Så over til litt notater fra foredragene. *** Jim Rawlings. Plenary on distributed MPC. - Normally in industry: decentralized MPC - Communication-based MPC: min. local objectives (Nash equilibrium) - Sis: this is feedforward/decoupling - Cooperation-based MPC: min. overall objective (Pareto optimal) Han viste noen eksempler men det var uklart for meg hva som kunne forklares fra interaksjon i lineære systemer, og hva som skyldes ulinieæritet (endring i aktive constraints). *** There were several intersting papers on estimation, including the nice overview given in the pleanry by Tor Steinar Schei (der kan Tore Lid nesten få prøveforelesningen gratis... :-). THURSDAY *** Graham Goodwin plenary on identification Identification in closed loop: Use large changres in setpoint to avoid identifying (-1/C)! Talked about delta-domain model (Euler integration) where lapalce variable is replaced by delta. Better insight than z-transform - for example there is a problem that very similar z-models can be co,mpletely different! Folding - more robust idea - don't trust models near samling frequency + don't use high-frequency data for identification! NOTE: If you want to dentify with limited outputs then it can be proved that closed-loop ID is the best. Use test signal between step and PRBS Question: How does ZN-identification and the close related relay identification fit into your ideas? Answer: I do not quite know.... ***Fra vår gruppe (IKP) hadde vi 5 bidrag; 4 presentasjoner og en poster (Preisig). 1. Presentasjonen til Elvira Aske onsdag formiddag om "Throughput maximation by improved bottleneck control" gikk meget bra. Hun fikk mange positive kommentarer i etterkant. 2. Sridhars presentasjon om "Implementation of optimal operating using off-line calculations" gikk også meget bra, men har var litt mer ueheldig med å ha sin presentasjon fredag ettermiddag. Jeg (Sigurd) hadde two talks in same session (nesten etter hverandre). Both had theme "maximum gain rule". 3. First was with Hori on distillation: key point: input scaling Juu-1/2 may be important for an ill-conditioned plants 4. Second was with Michela Mulas on waste water treatment. Went well - at lest many questions - especially on the wastewater treatment process - which is a very good sign! Q4a. Long comment and question from prof. M.A. Brdys from Gdansk. Poland (in wheelchair) (a) What comments did you get from control people at the plant? (b) It seems that the outlet specs should be active constraints? How do you ensure that these are met? My Answer: Good question. Probably you should have a syustem on top (MPC?) that looks after the water quality constraimnts and changes the setpoints for oxygen (2), MLSS, Qr/Qin Q4b. Vinay: Should't also the two CVs for oxygen be selected in a systematic manner? My answer: Yes, you are right. This would give a 4x4 matrix. Q4c. You should try the procedure on the original benchmark problem where you got your cost. Then you could compare. My answer: Yes, that would be interesting FRIDAY *** Tom Bewley, San Diego on Flow control. He is looking at control of bubble flow (with van der Shaft, I think). Feedback using "input" at bottom seems very difficult. he is looking into using some "feedforward" where he sets up a periodic variation that on average gives a desirable flow. (Sigurd idea: What about pressure control? seems faster!) *** Bewley mentioned that people were using moving horizon estimation (or its equivalence) for weather prediction. The standard approach is to iterate on the estimate of the intitial estimate at some time in the past and compute forward. Bewley thinks it would be better, especially for unstable or chapotic systems, to iterate omn the estimate of the present state and then compute back in time to compare with data. Seems reasonable to me! *** Jay Lee. Showed Bayes' theroem and mentioned that he did not want to say more about it since it already had been mentioned so many times in the conference (this is a comment to Heinz who had reviewers on his paper who did not know Bayes' theorem..) *** Vinay Kariwala presented his results on optimal measurement combination c=Hy (extension of the results of Halvorsen et al.). He considered minimizing the norm of M with respect to y. One interestimng results was that minimizing the 2-norm (Frobenius-norm) is "super-optimal" in that it also minimizes the singular value norm. *** Discussion with Erik Ydstie. Why is MPC used and not H-infinity? SiS: MPC is tuned FAR away from achievable bandwidth, so no robustness problem. H-infinity is good for the few applications where you really want fast control (so more applicable to aerospace, mechanical etc.) *** Unstable PID paper (poster by Shamsuzzoha Thursday). smzoha2002@hotmail.com. Looks good, although controller is not necessarily a PID, as the filter is unbounded. In a "real" PID the filter (as+1 / bs+1) should have btheta so the instability is "slower" than the delay). With lamda (=tauc) = 0.2 he gets Kc=0.46, tauI=0.27, taud=0.1 and a=1.58 and b=0.1. So bb so I would not call this a PID controller with a filter (but maybe it could be futher simplified to be one as I note that b=taud=0.1) Let me also note that for a real unstable plant, the form g(s) = k' e-theta s/ (s-a) is usually more physical, in the sense that k' and theta are less coupled than k and tau in the alternative form used by the author. ------- Final plenary Friday 08 June 2007 *** Bertusiak, Exxon (with simulations made by Robert Young) Most money is made from 1) PID, 2) Sequentiual and MPC 3) RTO (Note: RTO goes back to Texaco in the 50's). Today about 200 NLMPC applications running in industry. But most use empirical models (static nonlinear + linear dynamics) Exxon NLMPC. Started cooperation with Maryland and Georegia tech, Arkun et al. proposed sequential linear MPC, but they changed that. Fontaine et al., First application commissioned in 1994. Polymerization processes. Commercialization: Exxon decided to concentrate on applications, not software, so it was licensed to DOT products in 1997. In 2007 it was taken over by Honeywell. Some features of the Exxon NLMPC cost J = j1 + J2 + J3 J1-econmomic cost J2-trajecvtory cost J3-input cost Trajectory: Specify first-order response to desired setpoint with given tauc. Use 1-norm. SQP code: NOVA optimizer. Based on Biegler Discretize DAE using orthogonal collocation Simultaneous solution of prediction, control and optimization, with embedded state estomator (inclujding disturbance meodelling) Anti-windup handled by bound propagation from the distributed control system -used on Dell servers running Windows. Typically run every 3 to 6 minutes. Estimator: Extended Kalman or Moving horizon estimator? Genereally prefer EKF for computaional efficiency. 1990: Operators resisted MPC (no problem today) 2000: Engineers resist RTO Another thing Today: Some DCS vendors are putting in MPC is DCS, so may replace PID ?? ----------------------------------- Så litt fra CAB-konferansen 04-05 June (samme sted, en overlappende dag med DYCOPS) *** ÅPNINGSFOREDRAG George Bastin Den opprinellige foredragsholderen fra Genetech hadde trukket seg så åpningsforedraget ble holdt av George bastin fra Belgia. Han snakket om forenkling av reaksjonsnettverk der man antar at det er pseudo- steady state inni cellen. Da kan han selvsagt eliminere det som skjer inni cellen så har får et "input-output" modell. Det lignet på reaksjonsinvarianter, men var ikke helt det samme. Fluxer v er non-negative solutions to the kernel of the stoich. matrix N, *** SYSTEMS BIOLOGY PLENARY, Peter Wellstead, Hamilton Institute, Ireland. Original area: Nuclear reserarch. Then: Control Professor UMIST from 1990, Now: Systems biology Linneaus (1740). systema natue.. Book (1943): What is life? Erwin Schrödinger (stayed in Ireland 1939-1956) 1932: Book by Walter B. Cannon describing feedback in body 1948: Robert Winer. Vybernetics 1968: Mike Mearovic coined the term "systems biology" 1990-2001: Humane genome project The conditions have finally come together in the last 10 years for systems biology There are a lot interpretations of what systems biology is, e.g., Biology: gene sequencing automatic, Medicine: Homostatis (body feedback), etc. BLIND ALLEY: bioinformatics . Misses dynamics. Topics systems biology: Metabolism ("energy supply") + Cell signalling + Gene regulation Caution: well-mixed principle not very good Power-law models are poplular Virtual heart simulator Control opportunities: General + Oscillations + Networks More speculative: Principles + Junk DNA + Evolvability General control: Modelling and analysis of signalling pathways, sensitivity analysis (BST biochemical systems theory, MCA Metabolic control analysis, Doyle) control theory: understanding feedback, stability, robustness, analysis software: Recommends Systems biology toolbox for Matlab, uses SBML. Written by Henning Schmidt (Chalmers) -Ellings student! Conclusion Systems biology. Is it scientific pornography (as claimed by a dutch minister) or is a NEW oppurtunity for (chemical) control systems and engineering? Of course. his answer is that it is the latter. Upcoming meeting: 2nd intl. workshop on systems biology 17-20 Aug. 2008, Ireland end SYSTEMS BIOLOGY PLENARY ----------- Slutt på rapporten fra et bra DYCOPS-møte