Heinz A. Preisig Process Systems Engineering Methodological Approach to Process Operations & Design Modelling Control Synthesis 1 The Subject and its Components why modelling experiments process identification STMF modelling concepts project map model construction mixed continuous & event dynamic model water management process design container transport state discretisation controller design supervisor design DEDS research 2 even-dynamic model fault analysis MODELS Central role of models 3 Models the Central Object Models are used for just about everything in chemical engineering design control kinetics separations mixing, flow patterns etc. Different models for different systems and different models for the same system but for a different purpose model simplification methods such as model reduction, time-scaling, linearisation Model components may be re-used for different applications Model components for the physical structure of units or plant sections only or process units with particular reaction systems model libraries 4 5 MODELLER PROJECT Map 6 Modeller Project: Overview transfer Concepts enforcing database editors black-box models database kinetics activity Phys properties encapsulation model Thermo state trans Species & reactions Concept-enforcing model structure editor Subsystem selection algebraic manipulation ie linearisation rapid construction, modification, validation and maintenance of consistent process models Library of documented consistent process model Assumption handling Time scale selection problem instantiation Instantiated simulation model solver Instantiated design| identification model solver solver applications Instantiated optimal control model solver solver solver 7 Model reduction solver solver solver reduction of model development time and overall effort by 75 to 90% return was map for MATCH project 8 MODELLING METHODOLOGY 9 Basic components of networking approach Components of the mathematical description Physical and species topology example … Goals Research: Develop structured modelling approach Implementation of Model Design Tool supporting the construction and maintenance of process models Key issues: Model consistency also under simplifying assumptions Support of instantiating specific (mathematical) problems i.e. simulation, design and identification, and (optimal) control problems 10 Why? Common experiences: Time spent on modelling often greater than time needed for finding solution Higher complexity of models Many different ways to model the same process Our experience tool with only implements rudimentary systematic speedup is impressive. Estimated factor for simulations 10-100. Main reason: • • • • makes you think about time scales assumptions made aides in model instantiation, an often tricky business automatic code generation including splicing, thus no transcript errors transfer of information on model structure into the solver allows for all kind of conveniences, for example structured data analysis. • no low-level modelling errors 11 What is this all about Model construction Automatic but not constraint Maybe several different models each describing the process from a different point of view and with different fidelity Component software (separation of problem definition, analysis and solving) Efficiency and correctness, thus also trust Handling complexity Generating means to gain insight 12 The Modelling Process theory primary model A secondary model process assumption s A A A A instantiation experiment solution method verification 13 solved model Staged Approach 1: 14 Physical Topology Physical view Network of primitive systems and connections 2: Species Topology Colour with species add reactions 3: Equation Topology Transfer laws Kinetics Geometry Physical properties Equations of state Additional variables 4: Information Processing Control 5: (Simulation) Problem Definition Instantiate consistently Apply assumptions Modelling Basics plant primary abstraction time & length scale assumptions explode simplification & abstraction 15 Approach: Network of communicating control volumes 16 Modelling Concepts Process Dynamics Control volumes on which conservation principles are applied conservation of component mass and energy, momentum etc. Transfer between control volumes communications between control volumes Static constitutive equations Transposition of extensive quantity Generalized reaction concept Transformations between different state representations link between fundamental quantities and measured quantities or quantities used in transfer or reactions Properties the grey box of property approximations. 17 Example : Equations 18 Model equations for a system s in its environment e 19 Complete system: Stack all systems in the network up 20 Essentials State information (variables) used to describe transfer and reactions (transpositions) are mapped from the basic state variables (= conserved quantities). F + flow of ext quantities accum primary ex quantities state R reaction rates transfer of extensive quantity state variable transformations transposition of extensive quantity Flow matrix F is a function of the structure (from graphical input) Transposition (reaction) matrix represents the ratios of the species involved (stoichiometry) Equation structure is analysed on-line 21 secondary state Basic framework of MODELLER 22 Step 1 : Structure process using control-volume concept network of capacities and connections physical topology Step 2 : Define species distribution using species and potential reactions colouring of the physical topology species topology Step 3 : Define nature of network, the detailed mechanisms • transfer laws • kinetics • state variable transformations • properties (species, reactions and transfers) • geometry • assumptions: fast reaction, transfer and capacity equation topology Example of Physical Topology A A,C B CH B R C A+BD+E S E hardly any D,E P B,C,D,E 23 CC Example of Species Topology A A B B A C B A CH CH R R C C E CC P A C E CC P B A R A B CH R C E P E CH R C CC P D B C CC CC P CH 24 CH R E B E CC P E S Completion of the Model 25 Step 4 : Adding control control topology Step 5 : Model simplification derived secondary models Step 6 : Instantiation of problem mathematical problem to be solved Step 7 : Translation into target language specific to solver plus solver parameter instantiation mathematical | numerical problem to be solved Stage 4 : Add Controller CV CH B A level R temp CT C E CC 26 P Typical assumptions Make late order of magnitude assumptions: constant volume {unknown | fast | large} flows fast reactions fast process hydraulics Three key assumptions: fast process compared to flows and reactions negligible capacity effect, a singular perturbation problem fast flows with no constraints on magnitude first assumptions equilibrium fast reaction reaction equilibrium for fast parts (discussion see ACC 2002) 27 Steady state assumptions The state of system s is solely a function of the state of the environment reduces this part of the network to a connection, that is, the state of this system can be eliminated, if this is algebraically solvable. 28 Fast flows & Equilibrium We augment the description with the an equilibrium assumption for two systems, that is equations of the type: which must be solvable for the fundamental state vector x. This introduces an index problem, which can be resolved by first splitting the flow term into two separating the unknown flows for which an equilibrium assumption is made: Next these unknown flow term is eliminated by multiplying the whole equation with the null matrix of the respective flow matrix: 29 Current Status Modelling methodology that works for essentially any physical-chemical-biological process. Implementation of this methodology for component mass and energy First serious application was a great success. Model building time was cut by one to two order of magnitude in time Program has been transferred to industry together with student 30 Achievements Construct consistent algebraic models Eliminate transcript errors Increase turn around Implement high-level intuitive interface Minimal number of primitives Maximal flexibility and coverage Document everything transparently Allow for applying late time-scale assumptions Resolve all index problems All to manipulate everything except hard facts, which must be a minimum and as universal as possible. Support any level of detail and complexity Allow for inheritance | reuse of models components 31 Things to be done Extension of recursive structures approximations of distributed systems with networks of lumped systems Implement more physical concepts impose basic thermodynamic structures on defined transformations Separation of equation topology definition and problem instantiation Implement additional model manipulation tools such as time-scaling and linearisation Different target languages (currently MatLab), second one has just been added. Applications, applications, applications…. 32 return PhD: Mathieu Westerweele Collaboration: Protomation BV 33 SUSTAINABLE DESIGN Water management in households 34 Design: Water Management System for Households Is a distributed waste water system more economical more sustainable Can we get new products from human waste How about flexibility acceptance and cultural issues what can be said about the costs and their estimation horizon effectiveness problems that can be avoided or are generated Interest in NEW and SUSTAINABLE processes 35 The considered system Collaboration with Ralf Oterphol, Hamburg 36 A flow sheet 37 and a Simulink Model WASTEWATER TRANSPORT: WATERUSE: WASTEWATER TREATMENT: RESULTS: RESOURCES: Resources BinAmount -KFin (g/d) 1 Biowaste Bout Bin B1 B2 Sustainability indicators Blackwater transport FinAmount-K- Toilet Uin (g/d) 2 Onsite treatment B1 BW2 UinAmount -K- BW1 Fin BW1 B1 Blackwater treatment 1 BW3 R7 BW4 BW6 BW5 BW4 BW5 BS3 BS4 H2 R2 BS1 Y W4 C1 Disinfection d (g/d) D1 D2 D3 D4 D5 H2 H3 H4 H5 c (g/d) C1 C2 C3 C4 C5 C6 R3 HD2 Kitchen HD1 D2 C2 HD2 C6 Rin R1 R2 R3 R4 R5 Y W6 Y Wout Mass balance Bout BGout BWout BG2out BSout CSO Y Wout Eout GWout Rout Oout Greywater transport K3 Y W3 K3 GW1 W4 O2 PH4 GW2 R8 PH2 C3 PH3 Preliminairy GWT (TS removal) Primairy GWT (BOD reduction) GW2 GW3 GW3 GWout Washing D4 H4 W2 Water balance W3 Outdoor Fin (g/d)1 Y W5 Y W6 BSout Blacksludge Blacksludge transport treatment 1 Evaporation C4 FinAmount Yellow water treatment BS5 K2 D3 R4 Rainwatersystem Yellowwater transport BS5 H3 HD1 Personal Hygiene h (g/d) BG2out BS3 Y W3 BS2 Water demand 38 BW5 BS4 BS2 CSOout GW1 D1 Blackwater treatment 4 BW4 BW3 Uin Blackwater treatment 3 BGout BW2 BS1 Blackwater treatment 2 B BS YW Rout BW GW CSO BG Bin (g/d) D5 H5 R5 C5 O2 O3 Rainwater R6 R5 R7 Rout Reuse Potential Achievements Design tool A simple tank of less than a 250 l fed with sieved rainwater is sufficient to supply toilet flushing water for two people all year around in the Netherlands. Saves in the order of 30% drinking water at very little costs. Socio-cultural issues are important. The Real Challenge: Can I find new products being derived from (human) waste ? 39 Return PhD: Annelies Vleuten-Balkema Collaboration: Ralf Otterpohl, TU Hamburg-Harburg part of Sustainable Technology Program TU-Eindhoven 40 MODEL-BASED CONTROLLER DESIGN Modelling is the key Model reduction based on network analysis 41 Storing and moving fresh products (apples, mangoes,…) Project with ATO, the agriculture research organisation of the Netherlands 42 Abstraction of the storage and transport problem 43 Key : time scale assumptions leads to controller 44 Time Scale Analysis: Key to Many Problems Think about relative dynamics of interaction and processes involved. Am I interested in the fast behaviour or the slow behaviour Do I need one in order to get the other one These thinking leads very often to very significant simplifications of the models and consequently the application in which it is used (design, control, operations, identification, ….) 45 Return PhD: Gerwald Verdijck Collaboration with ATO (Dutch Agrotechnical Research Institute, Wageningen) 46 DEDS RESEARCH Why important for process industry What are they Some results on control 47 Discrete-Event Dynamic Systems Natural behaviour overflows, switching devices, bursting pipes, unit break down, measurement problems, … time-scale assumptions: fast flows, reactions or small capacities (singularly perturbed systems) Supervisory control continuous plants: start-up and shut-down, change-over level Fault detection What can I achieve with simple boundary detection? Observer temp Can I reconstruct the continuous trajectory? Issues practical: safety, availability of models for the continuous plant theoretical: discrete-event dynamic models are not deterministic can thus not be inverted for the design of the controller. 48 Event-Based Control recipe disturbance command event signal supervisor controlled plant state-event detector state state 49 events time example A toy problem in a toy plant, a demo 50 Recursive Control-Invariant Sets I have control available to keep process in the set of subdomains 51 A possible approach to controller design 52 The toy works 53 Achievements Compute the automaton given input event space, continuous process model and event detector linear plants: very simple nonlinear plants: mostly simple dimensional explosion problem: resolved, not an issue anymore key: insight in modelling, state-space approach using also observers. Some ideas on controller synthesis Automaton tailored for fault detection == observation of not measured discrete input in a nonlinear model, thus can also model process internal faults as being seen triggered from the outside. 54 Return PhD thesis : Philips, Yun-Xia Xi Collaboration: National University of Singapore 55 IDENTIFICATION STMF principle 56 Identification: STMF Kalman filtering Spline filters == multi-wavelets (also used for observers) Model mismatch of particular interest plant input spline filter output spline filter “derivatives” “parameter estimator” 57 Return 58 Heinz A Preisig Education Activities Pictures 59 Education Sulzer, Winterthur (CH): Chemical Lab Assistant; wet analytical chemistry, material sciences, distillation, cristallisation Polytechnic Institute (HTL): Chemistry, chemical engineering ETH Zuerich: Chemical Engineering @ microbiology, signal theory & ODEs & process dynamics, construction & corrosion ETH Zuerich PhD with David Rippin: Identification using Spline-Type Modulating Functions (being multi-wavelets) @ system theory (Kalman), stochastic systems & signal theory, advanced statistics, linear systems (Mansour), music, history of industrialisation … 60 Texas A&M University, Assistant professor; collaboration with C D Holland, R White University of New South Wales, Sydney; Senior Lecturer TU-Eindhoven, chair on Systems & Control, physics, chemistry & elec eng NTNU, chair on Process Systems Engineering Processes 61 Distillation (Sulzer, CD Holland Texas A&M) Crystallisation (Sulzer) Membrane process (DuPont, UniLever) Spark erosion machining (Producer in Switzerland) Electron accelerator (TU Eindhoven) Dryer (Dutch starch producer) Potatoes, mangoes, apples storage and transport (ATO Netherlands) Bio Reactor (Dutch reactor construction company) Wastewater plants (Several Dutch organisations) Life support system design (Texas A&M and NASA) Catalytic bed modelling (Ford Germany) Glass oven modelling and control (Several Dutch and German companies) Optimizing & plant-wide control (Shell chemicals, Bayer, …) Sugar product distribution network (CSR Australia) Moving catalytic bet reactor (CSIRO Australia) Control laboratories, computer networks (UNSW, TUE) Simulator design (ETHZ, TUE, Protomation) etc return 62 VIEWS Live experience: some pictures 63 Other “Views” A mix of mountains, hills and sea 64 Herisau Panorama 65 Cogee 66 Oberdorf 67 Oberdorf 68 Winter outside the Village 69 Wurzelmannli our Troll 70 Chlausete 71 Wineglass Bay 72