JARA 2i A MODELICA LIBRARY FOR INTERACTIVE SIMULATION OF PHYSICAL-CHEMICAL PROCESSES

Transcription

1 JARA 2i A MODELICA LIBRARY FOR INTERACTIVE SIMULATION OF PHYSICAL-CHEMICAL PROCESSES Carla Martin, Alfonso Urquia and Sebastian Dormido Dept. Informática y Automática, E.T.S. Ingeniería Informática, U.N.E.D. Juan del Rosal 16, Madrid, Spain {carla, aurquia, sdormido}@dia.uned.es KEYWORDS Interactive simulation, object-oriented, continuous simulation, education, automatic control. ABSTRACT Virtual laboratories, supporting interactive dynamic simulations, can be easily implemented by combining three software tools: Ejs, Matlab/Simulink and Dymola (with Modelica language). This approach has been successfully applied to set up virtual labs, with the intent to be used in education of the control of chemical processes. To achieve this goal, the following tasks have been completed: (1) the object-oriented design and programming of JARA: a Modelica library of dynamic hybrid models of some fundamental physical-chemical principles; (2) the description of the JARA physical models in a way suitable for interactive simulation; and finally (3) the implementation of the virtual labs. The virtual-lab view has been developed using Ejs, the physical models using JARA library and the controllers using Simulink block diagrams. The design of JARA 2i is discussed in this manuscript. In addition, its application to the modeling of virtual labs is illustrated by means of a case study (the control of an industrial boiler). INTRODUCTION Virtual laboratories supporting interactive simulations are effective pedagogical resources. During the interactive simulation run, the students can change the value of the model inputs, parameters and state variables, perceiving instantly how these changes affect to the model dynamic. The students understanding is enhanced by the interaction possibilities and this promotes their motivation to study the subject. A methodology well suited for virtual-labs implementation is discussed in (Martin et al. 2004). It is based on combining the use of Ejs, Matlab/Simulink and Modelica/Dymola. This approach takes advantage of the best features of each tool. Ejs capability for building interactive user-interfaces composed of graphical elements, whose properties are linked to the model variables (Esquembre 2004; Matlab/Simulink s capability for modeling automatic control systems and for model analysis ( Modelica s capability for physical modeling ( and finally Dymola s capability for simulating hybrid-dae models (Dynasim 2002). Ejs is a freeware, open source, Java-based tool intended to program web-based virtual labs. It is easy to install and use ( Ejs guides the user during the definition of the virtual lab, and it automatically performs all the tasks required to generate the virtual-lab code: a Java application or an applet. Ejs is based on a simplification of the model-viewcontrol paradigm. Ejs structures the virtual-lab definition in the following three parts: 1. Introduction: html pages including educational content related with the virtual-lab topic. 2. Model: dynamic model whose interactive simulation is the virtual-lab basis. 3. View: user-to-model interface. It is intended to provide a visual representation of the model dynamic behavior and to facilitate the user s interactive actions on the model. Ejs includes a set of ready-to-use visual elements, that allows easy creation of the virtual-lab view. The graphical properties of the Ejs view elements can be linked to the model variables, producing a bi-directional flow of information between the view and the model. Any change of a model variable value is automatically displayed by the view. Reciprocally, any user interaction with the view automatically modifies the value of the corresponding model variable. Ejs gives the user a procedure to define the model, and provides set of built-in ODE solvers to simulate it. Ejs virtual labs can run stand-alone or in conjunction with Matlab/Simulink. In this last case, the model can be partially or completely described using Matlab code and Simulink block diagrams ( Ejs synchronizes the numerical solution of the complete model. The model part described using Matlab/Simulink is solved using the Matlab workspace and the Simulink s numerical algorithms. On the other hand, Modelica language and Simulink block diagrams can be jointly used for model definition.

2 Dymola 5.0 interface to Matlab 5.3 / Simulink 3.0 can be found in Simulink s library browser: DymolaBlock block (Dynasim 2002). This block is an interface to the C-code generated by Dymola for the Modelica model. This C-code contains the numerical algorithms required to simulate the Modelica model, which is formulated in terms of differential-algebraic equations (DAE) and discrete-time equations. DymolaBlock block can be connected to the Simulink blocks and also to other DymolaBlock blocks. Contributions of this Paper This software combination approach has been successfully used to program a set of virtual labs for chemical process control. To achieve this goal, the following three tasks have been completed: 1. Object-oriented design of JARA (Urquia and Dormido 2003; Urquia 2000). Briefly, JARA is a set of libraries of dynamic hybrid models of some fundamental physical-chemical principles. The main application of JARA is the modeling of physical-chemical processes in the context of automatic control. 2. The description of JARA physical models in a way suitable for interactive simulation using Ejs. The modeling methodology proposed in (Martin et al. 2004) has been applied. The JARA library version, that is written in Modelica language and intended for interactive simulation using Ejs, is JARA 2i (release July 2004). 3. Implementation of the Ejs virtual labs. The physical models of the chemical plants are composed using JARA 2i, and they are embedded within Simulink s DymolaBlock blocks. The controllers are described using Simulink s block diagrams. The design and implementation of JARA 2i is discussed in this manuscript, and its application to the development of Ejs virtual-labs is illustrated by means of a case study: the control of a boiler. DESIGN OF JARA 2i The methodology applied for designing JARA was proposed in (Urquia 2000; Urquia and Dormido 2003). The fundamentals of this methodology and its application to the JARA design are discussed next. JARA s Design Methodology JARA s design methodology is based on the concepts of the object-oriented modeling, fully exploiting the principles of abstraction and information encapsulation. It distinguishes between the role of the library designers/programmers and the role of the library users. One of the main goals of this methodology is to guarantee that the users are able to use the model libraries without having to know their internal details. In particular, the users should not be confronted with numerical problems. A key idea is the following: the identification and solution of the numerical difficulties associated to the library use should be carries out (as far as possible) by the library designers. To achieve these objectives, non-declarative techniques are recommended in the library design phase: causal analysis, explicit selection of tearing variables, decoupling via dummy dynamics, etc. Another methodological key concept is the library design rules. They cover the numerical aspects of the design and use of the libraries, defining: (1) the complete set of requirements that each library s model has to fulfill; (2) how each library s model can be used and what models are interchangeable; and (3) how the libraries can be enlarged with new classes to be used in conjunction with the existing ones. The two pieces of information that the library designers have to provide to the library users are: (1) the hypothesis which each model of the library is based on; and (2) the library design rules. The JARA design methodology is composed of the following five steps (Urquia and Dormido 2003): 1. Phenomena and hypotheses. The phenomena to be modeled and the general modeling hypotheses are defined. Once all the pairs of phenomenon to be modeled modeling hypothesis, have been defined, then they can be grouped into libraries. 2. Phenomena interaction: causal-connector classes. A causal-connector class is defined by means of the following four pieces of information: (i) the variables composing the connector, classified into across and through; (ii) the allowed computational causalities of the connector class; (iii) the way of breaking the computational causality loops; and (iv) a graphic icon. All the classes of causal connectors should be gathered in a library. 3. Connection rules of the causal connectors: nature and number of the allowed connections among causal connectors. The library design rules are stated during the second and third steps of the design. The design rules are composed of the causal connector definition and their connection rules. 4. Definition of the interface classes. These interfaces should be only composed of an arbitrary number of causal connectors. The interface classes should be gathered in a library. 5. Description of the phenomena. The rules imposed by the interface s causal-connectors strongly condition the phenomena description. Techniques for facing these conditionings are discussed in (Urquia 2000; Urquia and Dormido 2003). Fundamental Modeling Hypotheses of JARA The usual way of enunciating the mass, energy and momentum balances is by means of the definition of a control volume (Bird, Stewart and Lightfoot 1975; Incropera and DeWitt 1996; Himmelblau and Bischoff 1992). The properties of the medium inside the control volume are considered time-dependent, but independent of the spatial coordinates. The only exception to this rule is the pressure inside the liquids. The control

3 volumes exchange mass and energy with their environment through certain control planes. The JARA control volumes and the control planes are considered macroscopic and fixed in the space. All the interactions among control volumes, and all the interactions of a control volume with itself (i.e., chemical reactions inside the control volume), are considered transport phenomena in JARA. This system decomposition into control volumes and transport phenomena suggests that: 1. The control-volume models should contain the equations describing the properties of the medium inside the control volume, as a function of the mass and energy transport through its control planes. 2. The transport-phenomena models should contain equations describing the flow of mass, energy and momentum through the control planes, as a function of the medium properties at these control planes. Three types of control volumes have been modeled in JARA: (1) control volumes containing an ideal, homogeneous liquid mixture, composed of an arbitrary number of components; (2) control volume containing an ideal mixture of an arbitrary number of semi-perfect gases; and (3) control volume containing a homogeneous solid. The control volumes containing liquid or gaseous mixtures are considered open systems (i.e., they can exchange mass and heat with their environment), and chemical reactions can take place inside them. In both cases, the volume of the control volume is considered a time-dependent property. The control volume containing a solid is considered a closed system (i.e., it only exchanges energy, not mass, with its environment). The only modeled characteristic in solids is the heat conduction (for modeling the walls of reactors, pipes, etc.). Two kinds of control planes are distinguished in JARA: mass-flow and heat-flow control planes. An arbitrary number of flows can flow through each control plane. The hypotheses made about the properties of the medium inside the control volume determine the nature and number of the control planes: 1. A solid control volume contains only one heatflow control plane: (i) the solid properties are spatially homogeneous, so that all control planes are equivalent; and (ii) the solid control volume is a closed system, so it has no mass-flow control planes. 2. A gaseous control volume contains one heat-flow control plane and one mass-flow control plane: all the gaseous mixture properties are spatially homogeneous. 3. A liquid control volume has two mass-flow control planes and one heat-flow control plane: (i) as the liquid properties related with the heat-flow are spatially homogeneous, all the heat flow controlplanes are equivalent; (ii) as the liquid pressure depends on the position, the simplest and most general control-plane selection is placing a control plane at the control-volume bottom and the other at the control-volume top. Any arbitrarily complex configuration can be modeled by decomposition into this kind of control volume (Urquia 2000). The JARA models of transport phenomena can be divided into two main groups: (1) mass transport due to the pressure or concentration gradients, the gravitational acceleration, chemical reactions, liquid-vapor phase changes, etc.; and (2) heat transport due to the temperature gradient. The interface variables are grouped into connectors according to this criterion, so that they describe the transport of mass and heat independently. A hypothesis related to the stirred-mixture approximation is to assume that the fluid going out from a control volume has the same properties than the fluid contained in it. In JARA, this approximation is applied to the calculus of: (1) the temperature and the composition of the fluid leaving a control volume by convection; and (2) the temperature of each mixture component leaving the control volume by diffusion. All the JARA models of transport phenomena make the flow direction reversible during the simulation run. As a consequence, the properties of the flow established between two control planes are calculated from the appropriate control plane at any time. An important property associated to the transport phenomena is the transport delay. There are different ways of modeling delays in one-dimensional geometry systems (EPRI 1984). The way used in JARA is the energy balance method (Incropera and DeWitt 1996). It consists in dividing the flow path into multiple control volumes. Adjacent control-volumes are connected by the transport-phenomena models describing the heat and mass transport between them. As the number of control volumes increases, the solution gets closer to a transport delay (EPRI 1984). Modeling for Interactive Simulation The modeling methodology proposed in (Martin et al. 2004) has been applied to the JARA 2i programming. This methodology states how a Modelica model can be formulated to suit interactive simulation using Ejs. The obtained Modelica model fulfills the following two requirements: 1. In order to embed the Modelica model within a Simulink block, the computational causality of the Modelica model interface needs to be explicitly set. 2. As a result of the user interaction, the Modelica model needs to support instantaneous (discontinuous) changes in the state, and in the value of its input variables and parameters. In addition, several choices of the state variables can to be simultaneously supported, in order to provide alternative ways of describing the state changes. The reader is referred to (Martin et al. 2004) for a complete discussion.

4 a) b) c) d) e) f) g) h) Figure 1: a) JARA Packages; b) JARA.cuts Package; c) JARA.interf Package; d) JARA.heat Package; e) JARA.liq Package; f) JARA.gas Package; g) JARA.phase Package; h) JARA.chReac Package JARA Architecture The JARA libraries ( packages in Modelica terminology) have been organized in order to facilitate their use and maintenance. The package hierarchy is shown in Fig. 1a. JARA is composed of seven packages (see Figs. 1.b - 1.h). The causal connectors are defined in the JARA.cuts package (see Fig. 1b), and the model interfaces in JARA.interf package (see Fig. 1c). The JARA.gas package (see Fig. 1f) contains models of control volumes containing gaseous mixtures, of gas transport (i.e., gas-flow by convection and diffusion, valves, pumps, etc.) and boundary conditions (i.e., gasflow and pressure sources). Similarly, JARA.liq package (see Fig. 1e) contains the equivalent models for liquid mixtures. The mixtures of liquids and gases are considered ideal and they can be composed of an arbitrary number of components. The models related with the heat transport are collected in the JARA.heat package (see Fig. 1d): models of control volumes containing solids, thermal resistances and boundary conditions (heat and temperature sources). The JARA.phase package (see Fig. 1g) contains models of vapor-liquid phase-change: boiling and condensation. The models of chemical reactions are in JARA.chReac package (see Fig. 1h). The modeling details and the library design rules can be found in (Urquia 2000; Urquia and Dormido 2003). CASE STUDY: BOILER VIRTUAL-LAB The mathematical model of an industrial boiler is found in (Ramirez 1989). The input of liquid water is placed at the boiler bottom, and the vapor output valve is placed at the top. The water contained in the boiler is continually heated. The diagram of the boiler model, composed using JARA 2i, is shown in Fig. 2a. Two control volumes are considered: (1) a control volume containing the liquid water stored in the boiler; and (2) a gaseous control volume containing the vapor. The vapor volume is equal to the difference between the boilerrecipient inner volume and the water volume.

5 Gaseous control volume Valve Pressure source Boiling process Heat flow source Recipient Liquid control volume Liquid source Control module Control module a) b) Figure 2: a) Boiler Model Composed Using JARA 2i; b) View of the Boiler Virtual-Lab Implemented using Ejs The boiling is a transport phenomena represented by a model connecting both control volumes. The heat-flow into de boiler, the pressure at the valve output and the water pump are modeled using JARA source models. The Ejs view of the boiler virtual-lab is shown in Fig. 2b. The plant has been modeled using JARA 2i. The user can interactively choose between two control strategies: manual and decentralized PID. The control system has been modeled using Simulink: a PID is used to control the water level and another PID is used to control the vapor flow. The manipulated variables are the pump water-flow and the heater heat-flow respectively. The parameters of these PID controllers can be changed interactively. In addition, the value of the model state-variables (mass and temperature of the water and the vapor), parameters (inner volume of boiler), and input variables (temperature of the input water, valve opening and output pressure) can be changed interactively during the simulation run. CONCLUSIONS The design and implementation of JARA 2i has been discussed, and its use to develop Ejs virtual-labs has been illustrated by means of a case study. The use of Modelica language has reduced considerably the modeling effort and it has permitted better reuse of the models. Ejs visual elements have allowed easy creation of the virtual-lab view. ACKNOWLEDGEMENTS Part of this work has been supported by the Spanish CICYT, under DPI & DPI grants. REFERENCES Bird, R.B; W.E. Stewart and E.N. Lightfoot Transport Phenomena. John Wiley & Sons. Dynasim Dymola. User's Manual. Version 5.0a. Dynasim AB. Lund, Sweden. Electric Power Research Institute (EPRI) Modular Modeling System. Theory Manual. MMS-02 Release. Esquembre, F "Easy Java Simulations: a Software Tool to Create Scientific Simulations in Java" Computer Physics Communications, 156, pp Himmelblau, D.M. and K.B. Bischoff Process Analysis and Simulation. John Wiley & Sons. Incropera, F.P. and D.P. DeWitt Fundamentals of Heat and Mass Transfer. John Wiley & Sons. Martin, C. et al Interactive Simulation of Object- Oriented Hybrid Models, by Combined use of Ejs, Matlab/Simulink and Modelica/Dymola, In Proc. of the 18 th European Simulation Multiconference, pp Ramirez, W.F Computational Methods for Process Simulation. Butterworths Publishers, pp Urquia, A and S. Dormido Object-Oriented Design of Reusable Model Libraries of Hybrid Dynamic Systems. Mathematical and Computer Modelling of Dynamical Systems, Vol. 9, No. 1, pp Urquia, A Modelado Orientado a Objetos y Simulación de Sistemas Híbridos en el Ámbito del Control de Procesos Químicos. Ph.D. Thesis. Dept. Informática y Automática, UNED, Madrid, Spain. AUTHOR BIOGRAPHIES CARLA MARTIN received her M.S. degree in Electrical Engineering in 2001 from the University Complutense of Madrid (UCM), Spain. She is currently completing her doctorate studies at UNED, Madrid. ALFONSO URQUIA received his M.S. degree in Physics in 1992 from UCM and his Ph.D. in 2000 from UNED. His work experience includes six years as an R&D Engineer at AT&T and Lucent Technologies - Bell Labs. Since 2002, he has been working as an Associate Professor at UNED, Madrid. SEBASTIAN DORMIDO received his Physics degree from UCM (1968) and his Ph.D. from Country Vasc University (1971). In 1981, he was appointed Full Professor of Control Engineering at the UNED Faculty of Sciences. Since 2002, he is President of CEA-IFAC.