Rapport fra Adchem06 i Brasil Exchange rate: 1 BRL = 3.1 NOK (was 2.3 one year ago!) Saturday 01 Apr: Short-course on plantwide control in Porto Allegre. About 30 participants Organizer: pedro + leandro Sunday 02 Apr: Tour to Itaimbezinho canyon. 100 BRL MONDAY 03 APR 2006 ----------------------------------- Plenary lecture by Sebastian Engell 1, the gap servo- and regulatory problem Control is considered a subordinate task because of restrictive view 2. control structure selection skogestad (2000) was probably the first to consider this systematicallu "self-optimizing control. Engell find the term misleading" He looks at J - loss seems better If his first term=0 -> no reg. control necessary ? Must be wrong! Could have had improvement! 3. rto Marlin and Hrymax (CPC 1996): Rto should provide controller design . Wants to make RTO faster - How can you use QP/LP setpoint optimization? Seems strange but interesting. NEED TO CHECK! Zanin, gouvea, Odloak (2000, 2002). New method tested on RTO 4. Proposed: From RTO to dynamic optimization Direct finite horizon optimizing control Example. Simulated moving bed (SMB) chromatography: ring of columns - Coventional hierarchical worked OK for "linear isotherm" (Carsten Klatt) - But may get too complex, so led to direct dynamic on-line optimization (Toumi, Engell; CES 2004) 5. Application to reactive SMB process (Hashimoto) 6. New Reduction of complexity - ala NCO tracking ----------------------------------- Semiplenary by Frank Allgower. Nonlinearity measures G best linear approx; N = G + Delta Can make 6 delta's/measures (as for robust control) Simple nonlienarity measure: inf G ||Delta|| ----------------------------------- After this I went to the MPC session, but tthere was not too much of interest. Jeg gikk ikke til keynote av Ton Backx men den var meget dårlig ifølge Bjarne Jørgen on PRICO LNG process. Went well. Questions Fra inder: On meas combination (not much to save - should have mentionede Vidar method( from Margret Bauer (Pretoria/Nina student): On colling (active constraint!) ----------------------------------- TUESDAY 04 APR 2006 ----------------------------------- Claire Tomlin Pleanary on Systems biology Continuous "kinetic" model with 37 parameters gave good results (adjoint method) ----------------------------------- B. Srinivasan semiplenary og batch control (rather obvious and fundamental, but I agree with what he says) Two types of outputs: run-time and run-end Two types of adaptations: run-time and run-to-run " feedback" - correction for present batch "feedforward" - correction from previous batch ("run to run" to adapt feedforward using "iterative learning control") Stability. (a) run-time/trajectory "stability" is not something you necessarily want in a batch reactor! sensitivity issue Analogy: plug flow reactor with temperature increasing with z (towards end) (b) But Of course feedback control loops (e.g. Treactor) mist be stable ----------------------------------- Semiplenary Joe Qin on performance monotoring ----------------------------------- jens: wilson eos -> wilson eq "need to use 4 of the DOFs for stabilization". No, no DOFs lost!! Setpoints DOFs We probably ned to use exact local method Qu1; on stabilization Qu2: on split of L and V ----------------------------------- Hidic ideal HIDiC has no reboiler/condenser Claims introducew reflux and boilup for controllability resons ----------------------------------- Wednesday Bilde fra middag onsdag: Bjarne Foss, Joe Qin, Tunde Ogunnaike, Jorge Trierweiler, Moses Tade, Frank Allgower, Ridvan Berber, Sigurd, Ton Backx, Heinz Preisig Disco: Margret Bauer, Moses Tade + bl.a. Nina, Andrzej Banazuk, UT "social gap" From what he explains the interaction between control and design engineers in the aeroe.ngine industry is very poor. "Passive solutions are always preferred" Flutter control; Transmission zeros are difficult to understand for non-control engineers Get rid of RHP-zero by chemhing sensor "detrimental symmetry" Jeg tror ikke på hans forklaringer men det ser ut til å virke. Need to go from control algorithm to control architecture to make insight. Use physics of natural feedback!! Tom Edgar: Equilibrium model better than rate-based for experimental azeotropoc column! Cheng and Forbes Claimed "multi-objective": Maxomize production + save energy. NO. One cost 19 MVs and 21 CVs. How did you get them? Snakket med: Harvey Arcellaono-Garcia, Berlin (Wozny) Summary 209 submitted, 22 rejected, 12 keynote, 95 oral, 67 poster authors: 47 brazil, 28 usa, 21 Candaa, Argentina Primary athor: 185 academic, 15 industry, 9 giverment Adchem 2009 in Istanbul, Turkey. Koc Univ. 6-8 juli 2009 ---------------------------------- Petrobras CENPES visit 07 April 2006. Arrive 1000. Mario Campos + Herbert 7th biggest petroleum company in the world. By law: Need to invest money with universities, 0.75% by law to research in universities A lot of the developements they described are done in coorporation with various universities RTO: In-house using seq. modular process simulator PETROX (distillation) from about 1990 Are comapring with equation-oriented from Aspentech, which is looking better In the future: will use ALSOC - EMSOC (english) (Secchi) Staring to look at dynamic RTO (PhD) MPC: Has own linear MPC (more than 30 applications) + nonlinear/multiomdeol MPC PID control. New: Campina Grande Univ. Closed-loop relay tuning. Based on Schei paper for 1997 But: Most common today is open-loop step test Performance assessment. Harris index etc. Active constraibs change: Tuning of MPC difficult - chooses most conservative tunings Have impemented in 1997 MPC on Braschem ethylene plant, but new update will be Aspen because Petrobras did not have time to do the update! Benefit MPC: Estimated at USD 35e6 /yr Intelligent system for startup and optimization using G2 from Gensyn - have 4 aqpplications running - but G2 is too general. Will change away from G2 Intelligent system of knowledge management in petrleum. Have tried to develop intelligent diagnosis system - but difficult... too many false alarms CHECK: Koktotovic on hydrogen fuel cell instabilities ----------------------------------- Dinner Friday night 06apr2006 with Petrobras at meat restaurant: Giovanni Nunes, Erick Coelho, Mario Campos + Herbert .... + 3 profs (Sigurd, Heinz & Bjarne) Afterwards: Scenarium (?) club i Lapa ----------------------------------- Saturday: Rio tour with Oscar Rotava + eldest son Pedro (wife Maria, yongest daughter Helena) Address: Rua Almirante Alexandrino 3183, 29241-262 Rio de Janeiro, Brasil email: rotava@petrobras.com.br 2100 (delayed copuple of hours): Fly til Maceio sammen med Antonio Møtt av Julio Inacio (Klor-alkali). Hotell: Melio Maceio Neste dag kl. 930: Beach-runde med Julio og Antonio ----------------------------------- VISIT Braschem 10APR 2006 0800 Pick-up 0830 safety 0900 Information plant John Brooman, Julio Inacio + Antonio John: 2800 people in Braschem - largest petrochemical in South America Plants: Maceio-Alagoas (Chlor-alkali 1975, EDC 1979; VCM-PVC; 1989), Camacari-bahia (olefins, PVC) , Sao Paulo (small old PVC emulsion), Triunfo (olefin;1982) Head office: Sao Paulo Maceio: Salt mines 8km away caustic soda: 460 kt/y Chlorine: 400 (little exported, goes to EDC) EDC: 600 (1/3 exported, rest to VCM-PVC) HCl: 42 Sodium hypo: 20 PVC: 240 kt/y Julio on Chlor-Alkali plants Technology: Eltech/Swenson 170 MW (same as Maceio) 500km ethylene pipleline from Bahia plant! 155 kt/yt, 8in=20cm This plant was bought by Braschem (1994 - 1996 - 2002 100% Braschem) John on VCM plant Technology: Badger VCM: 200 kt/y (240 kt since 2005 with new furnace) Raw material: EDC (Cl source) + make own from ethylene by oxo-process * Cracking of EDC to VCM (63% conversion). quench. steam generators. HCL column. 65 trays, HCl top (max VCM 100 ppm), VCM+EDC btm (max 50 ppm HCl due to pressure problem next column). Two temp controllers. Possible bottleneck in top cooling. Tray flooding * VCM column. Top product VCM (99.7%, max on "dry lights"), Btm product EDC (recycle; 0.1% VCM) * EDC column ("lights"). Top product "dry lights" (side products), Btmp product EDC * Final EDC distillation 3 columns * Oxychloronation: C2H4 + 2 HCl + 1/2 O2 (air) = EDC + H2O ? (fluidized bed with steam production). 99.7% conversion of HCl. TC with steam pressure. Then to "hot" quench. Abrober. Solv. dist. MAIN operational problems: Coke + corrosion in wet end. COLUMN control: HCl two temperatures, rest mostly has no temp. control Found a loss: 0.1 t/h EDC in vacuum column btm (20% of the flow). 0.1 t/h* $300 /t * 8000 = 240 000 $/y Strange system: EDC-column w/ heavies in btm (T1302) + Vaccuum column John claims you need EDC loss in heavies to improve flows ??? But it must vary a lot since the EDC column T1302 seems to have no control. John on PVC plant Batch process. "not a serious process". 3 different products. 16 reactors at 60m3. 6 with reflux condenser, rest with only jacket coolers. Charge: Water + VCM + surfactant + initator. (4-10 hours batches; fast and slow resins) Small distillation column, Plant tour with John After lunch: From Campina Grande: Romildo Brito + Shiva Prasad prasad@deq.ufcg.edu.br brito@deq.ufcg.edu.br NB. Sende informasjon om Peter Singstad og optimal PVC batch control send information dept and electrochem <--- sendte dem annual report 21. april 2006 Tuesday: Seminar on plantwide control from 0930-1130 with about 15 present Then took plane back to Norway! %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% ---------------------------------- ----------- Research ideas etc. - Define optimization and control. Control layer="setpoint tracking" - Make a list of my important self-opt. papers ("pensum"): 1. Review paper 2. Simple self-opt. paper 3. Plantwide paper 4. Linear paper and explain a little about each paper! - Jens: He is looking at case where we do not have avtive constraint control and thus need to "back off" in order to guarantee feasibility. Back-off = not optimal nominally - Do not use variables that may saturate in lower layer - or must at least use "input restting" to avod, e.g. bypass heat exchanger used for temperature control When can no control be used? L/D V maybe, not l/D B as Braschem doesw on the EDC column... ---end-----