- Email: [email protected]
SOFTWARE FOR SUPERVISION SYSTEM DESIGN IN PROCESS ENGINEERING INDUSTRY
SOFTWARE FOR SUPERVISION SYSTEM DESIGN IN PROCESS ENGINEERING INDUSTRY B. Ould Bouamama ∗ M. Staroswiecki ∗∗ A.K. Samantaray ∗∗∗ ∗
LAGIS, UMR CNRS8146 Cité Scientifique, Polytech’Lille F59655 Villeneuve d’Ascq Cedex, France E-mail: [email protected] Tel. (+33) 320337139, Fax: (+33)320337189 ∗∗ Ecole Polytechnqiue de Lille, Cité Scientifique F59655 Villeneuve d’Ascq Cedex, France. ∗∗∗ Systems, Dynamics and Control Laboratory, Department of Mechanical Engineering, Indian Institute of Technology, 721302 Kharagpur, India.
Abstract: The software "ModBuild" presented in thid paper is developped for process monitoring applications. It automatically creates complex process dynamic models from a simple graphical interface, where system components can be dragged from a component data base and interconnected so as to produce the overall system, following the Piping and Instrumentation Diagram. Once the model has been created, ModBuild checks its consistency and performs its structural analysis in order to automatically determine the diagnosis algorithms which should be implemented, and their fault detectability and isolability performances. The friendly graphical user interface allows to test several sensor configurations in order to optimise the diagnostic possibilities. The obtained model can also be used for the simulation of the process and its diagnosis algorithms in normal and faulty c situations. Copyright °2006 IFAC Keywords: FDI, Software, Bond graph, process engineering
1. INTRODUCTION
The system modelling is an important and difficult step in FDI design. Once the model is built, fault indicators can be designed using analytical redundancy or observer based approaches. However, obtaining the model is a difficult task, since complex systems, considered in process engineering, are characterized by the coupling of several phenomena of different natures.
Supervision systems include a set of tools and methods for the control of industrial processes in normal working conditions as well as in the presence of failures. The main activities addressed by supervision systems are monitoring and fault detection, diagnosis and decision making for fault accommodation or system reconfiguration. Fault Detection and Isolation (FDI) procedures are essential for improving the process safety.
Yet, in supervision tasks, human operators do not consider the running process in terms of its mathematical behavior model, but rather in terms of
646
the functions it achieves. The association of Bond Graphs models (as multidisciplinary tool and unified language) with functional models is highly desirable. Furthermore, Bond Graphs allow to develop generic techniques, which can be applied to any thermofluid process, rather than concentrating on developing specific models for specific situations. Those structural, graphical and causal properties of the bond graph tool are then exploited to design a powerful software for supervision in process engineering. The theory behind this application is developed in ((Medjaher et al., 2005), (Ould-Bouamama et al., 2005), (A.K.Samantaray et al., 2004), (Mukherjee and Samantaray, 2001)). This research work is developed and tested in the real processes in framework of an European project. ("CHEM", 2001-2004).
based on the derived residuals. The same IDE (integrated development environment) can be used to create models for newer database items based on existing capsules (derived or hierarchical submodels) or bond graph models or combination of both. The paper is organized as follows : section 1 presents the inputs and outputs of the software and the main functions. The Graphical user interface and how the generic data base is organized are discussed in the second part. in the third section it is shown how the bond graph tool is used for the behavior model and ARRs generation. Section 4 concludes the paper.
2. INPUT AND OUTPUT DATA OF THE SOFTWARE
There has been a significant research on software modelling tools for process engineering, starting with the description of stationary behaviors (Westerberg and Benjamin, 1985), (Briegler, 1989). Dynamic systems are successfully considered in Mathlab/Simulink libraries (Simulink, 1997), or in recent achievements like the modelica modelling language (Elmqvist et al., 1993), (Wollhaf et al., 1996). However, in the first case, the user must introduce the model equations (which means that the model has already been set), while in the second case, the physical phenomena are not explicitly displayed to the user, and the model can not easily be refined, when its structure changes as the result of more assumptions. For chemical process applications, the Odysseo (Object-oriented Dynamic Simulation Software Environment) toolbox of the Prosim software allows to model chemical processes for the simulation of their dynamic behavior, using fixed input/output causalities (Moyse et al., 1999). None of these software derive symbolic state equations, the numerical models being used just for simulation.
From a functional point of view, the main functions (see fig.1) supported by the software are : (i) Create an energetic and control model of the thermofluid process in iconic format. (ii) Derive the dynamic model in faulty and normal operation. (iii) Derive the Analytical Redundancy Relations (ARRs) in the symbolic format and the monitoring ability analysis based on technical specification. (iv) A sensor placement for structural monitoring ability. The software requires as input the architectural model of a plant to start with. Such models are built using a generic plant-items database and are closely related to the Process & Instrumentation Diagrams (P&ID). The objects for these plantitems are derived from certain existing sub-model classes. Optimal sensor placement New sensor architecture
Diagnosability results
Diagnosability analysis
Tchnical specifications
Process
Generate a dynamic and formal models
This paper presents a model builder software for the design of FDI algorithms for thermofluid processes. The methodology is based on the analysis of bond graph, structural and functional models. The system dynamics model, and the residuals associated with Analytical Redundancy Relations (ARRs) are generated under a symbolic format. Thanks to a developed generic item database which consists of a set of predefined process, controllers and sensor classes, and has been incorporated as capsules in the software, the designer can easily build the dynamic and functional models of most thermofluid processes from the Process and Instrumentation Diagram (P&I D) of the plant, and automatically generate symbolic dynamic models and ARRs . It also provides structural monitoring ability (ability to detect and to isolate a faults) analysis and isolation properties
Generate a formal residuals Formal Residuals
P&ID
• Fonctional analysis • BG • Physics
Sensors Data from sensors
• Structural analysis • Causal and structural propeerties of BG
Online implementation
Numerical residuals
Fig. 1. Main activities of model builder These models are then parsed and connection syntax is validated. User specified nomenclature is applied to all known inputs and also to all measured quantities. The fault signature matrix, isolation ability and monitoring ability are output as a single compact matrix. The input model and all the outputs are exported in XML form to be used by other toolboxes in the supervision platform. The outputs are all shown in form of equations and matrices in the Graphical User
647
Interface (GUI). All the sub-models are created through bond-graph modeling language and are linked by the implementation to create a global model in memory to start equation derivation. When the signature faults is generated, the user can add a new sensors graphically to satisfy the imposed technical specifications. Finally the numerical residuals can be implemented on line for supervision.
3. SOFTWARE PRESENTATION The overall Graphical User Interface (GUI) of the toolbox is shown figure 2. Fig. 3. Generic data base The generic database consists of fixed causality and generic causality capsules. While storage tank is modelled as a fixed causality capsule, which has a flow input and effort output, the valve is modelled as a variable causality generic capsule. Valve sub-model capsule can have either effort or flow as the input and the flow through the valve is the output. Capsules are well modelled subsystems with their partly derived equations. Their equations are linked together in the main model and then reduced to result in compact behavioral equations. The internal model of the tank capsule is shown in figure 4. Fig. 2. Graphical User Interface
3.1 Generic plant Item Data Base The system architecture is a component oriented model, which directly describes the process plant as a network of interconnected plant items (valve, pump, pipe, ...). The Piping and Instrumentation Diagrams (P&ID) is used for the visual description of the system architecture. A plant item is a quantity of matter or space that is, or is intended to be, a part of the process plant, e.g. pump, valve, pipe, ... A process plant is a functional assembly of plant items. The user build the P&ID of the supervised process based on implemented generic plant item data base. The generic plant item data base consists of a set of predefined process classes. The structure of the library (based on functional and bond graph theory (Ould-Bouamama, 2003)) is divided into three main parts :the thermofluid process class ((boiler, valve, pipe, tank,...), the energy sources (pumps, heater, ...), the controller class (P, PI, PID, ...) and sensor class which represents the available instrumentation. The generic plant item data base (fig. 3is displayed under icon model library familiar to the industrial user. Based on the P&ID of the process plant, the architectural model can be built.
Fig. 4. Internal Bond Graph Model of the Tank The elements connected to bonds 1, 3, 4, 5 and 6 are external flow input ports. Elements connected to bonds 7-11 are external effort output ports. Bond number 12 is connected to an effort output port which may be connected to measurement devices. All these external ports have an attribute that specifies them as optional ports. Optional ports may or may not be connected in the main model. For example, only three ports of this capsule are connected to other capsules in the illustrated model. The C- element for storage of
648
liquid appears in bond 2. The parameter for the element is C2. One or more representative icons and some help text can be embedded as resources to the capsule object.
FDI system. This has been implemented using a routine to create the Global System Model. The behavioral model for the system is created using preferred integral causality. The part equations of the capsules are linked together to arrive at the compact set of state equations. If the residuals for the global model has already been derived, they are also integrated to these equations. There is an option to derive the equations for faulty systems. Each capsule may be associated with another hidden capsule that has fault models embedded to it. As an example, in the capsule for storage tank, a leakage flow is added as a negative flow source and stored as the corresponding faulty mode capsule. The non-faulty mode capsule is used to derive the residuals whereas the hidden capsule with embedded faults may be used for the behavioral model. If no corresponding faultymode capsule is found, the non-faulty capsule is used instead. Linking residuals with the behavioral model requires automatic creation of certain relations to obtain results of integrated and differentiated terms appearing in the residuals.
3.2 Connectivity rules The connectivity of capsules is checked for validity in the main model. Any errors found there are reported. All post-processing is suspended till the model is syntactically valid. As an example, if two tanks are connected directly without a transportation process such as valve or pipe between them, then the network process is not valid as discussed in(Bouamama et al., 2004). This invalidity is displayed by the software as shown in figure 5. In thermofluid systems, the connections are vector bonds. The input and output side vector dimensions of each bond is checked for consistency of dimensions. A syntactically valid model can be linked to create a global model. While creating global models, consistent input and output ports of capsules are assimilated and a memory model of the junction structure is created.
By default, the Software assigns integral causalities. To assign differential causalities, all measurements are converted initially to reverse sources and all I and C elements are dualized. After integral causality assignment on partially dualized graph, the dualized elements are again dualized to return to normal graph. The global model in preferred differential causalities and inverted measurements is now ready for residual derivation algorithm and monitoring ability analysis (fig.6).
(a)
(b)
Fig. 5. Valid (a) and invalid architectural model
4. ARRS AND BEHAVIORAL MODEL GENERATION It has been implemented using SYMBOLS 2000 ((Mukherjee and Samantaray, 2001)) interface and symbolic manipulation routines. The model is drawn using sub-models only from the set of generic object database based on thermofluid defined classes previously. They contain the bond graph and equation model for the component and interface constraints. These sub-components when connected to create architectural model for a system create syntactically valid models. The software using the power directions, causal properties and vector dimensions for each connection checks the model validity automatically.
Fig. 6. Monitoring ability analysis interface The behavioral model can be used to create a simulation model in the software. The model can also be exported as S-function and then simulated in the Matlab-Simulink environment. Additionally, the Model builder simulator module provides slider variables whereby faults may be simulated interactively by the user at the run-time (fig.7). Such interactive variation of parameters leads to
However, for generation of residuals we need a global model on which causal properties are analyzed and altered to create a proper computable
649
availability of options to try out several combinations and to make subjective observations from the online plotting of the residuals.
Medjaher, K., A.K. Samantaray, B. Ould Bouamama and M. Staroswiecki (2005). Supervision of an industrial steam generator. part II: On line implementation. Control Engineering Practice (CEP) 14(1), 85—96. Moyse, A., L. Jourda, J. M. Le-Lann and X. Joulia (1999). Process, hydraulic and control devices in odysseo, an object-oriented framework for process dynamic simulation.. Monpellier. Mukherjee, A. and A. K. Samantaray (2001). System modelling through bond graph objects on SYMBOLS 2000. In: Int. Conf. Bond Graph Modeling and Simulation. Vol. 33. SCS. Phoenix, Arizona. pp. 164—170. Ould-Bouamama, B. (2003). Bond graph appraoch as analysis tool in thermofluid model library conception. Journal of Franklin Institute 340(1), 1—23. Ould-Bouamama, B., K. Medjaher, A.K. Samantaray and M. Staroswiecki (2005). Supervision of an industrial steam generator. part 1: Bond graph modelling. Control Engineering Practice (CEP) 14(1), 71—86. Simulink (1997). Dynamic System Simulation for Matlab. Math Works. France. Westerberg, L. T. and B. R. Benjamin (1985). Thoughts on a future equation-oriented flowsheeting system. Computers and chemical Engineering 9(5), 517—526. Wollhaf, C., K. Schultz and S. Engell (1996). BaSiP-batch process simulation with dynamically reconfigured process dynamics. Computer and Chemical Engineering supplement pp. 1281—1286.
Fig. 7. Simulation Interface
5. CONCLUSION The software aims at creating an integrated interface to easily create models for plants and derive fault diagnostics specific results from it in symbolic equation forms. It also aims at providing means for the validation of these results through offline simulations and exporting the results for on-line implementation by other toolboxes when they run together in the supervision platform. All these are implemented in a single integrated development environment and significantly reduce the time and cost of modeling and manual derivation of equations and residual.
REFERENCES A.K.Samantaray, K. Medjaher, B. Ould-Bouamama and M. Staroswiecki (2004). Component based modelling of thermofluid systems for sensor placement and fault detection. SIMULATION : Transactions of SCS 80(Issue 7-8), 381—398. Bouamama, B. Ould, K Medjaher, A.K. Samantaray and G. Dauphin-Tanguy (2004). Model builder using functional and bond graph tools for FDI design. submitted to Control Engineering Practice (CEP) journal 13(7), 875— 891. Briegler, L. T. (1989). Chemical process simulation. Chemical Engineering pp. 50—61. "CHEM", European Project (2001-2004). Advanced Decision Support System (DSS) for Chemical/Petrochemical Manufacturing Processes. g1rd-ct-2001-00466 ed. Elmqvist, H., F. E. Cellier and M. Otter (1993). Object oriented modelling of hybrid systems. In: ESS’93 European Simulation Symposium. The Netherlands.
650
LAGIS, UMR CNRS8146 Cité Scientifique, Polytech’Lille F59655 Villeneuve d’Ascq Cedex, France E-mail: [email protected] Tel. (+33) 320337139, Fax: (+33)320337189 ∗∗ Ecole Polytechnqiue de Lille, Cité Scientifique F59655 Villeneuve d’Ascq Cedex, France. ∗∗∗ Systems, Dynamics and Control Laboratory, Department of Mechanical Engineering, Indian Institute of Technology, 721302 Kharagpur, India.
Abstract: The software "ModBuild" presented in thid paper is developped for process monitoring applications. It automatically creates complex process dynamic models from a simple graphical interface, where system components can be dragged from a component data base and interconnected so as to produce the overall system, following the Piping and Instrumentation Diagram. Once the model has been created, ModBuild checks its consistency and performs its structural analysis in order to automatically determine the diagnosis algorithms which should be implemented, and their fault detectability and isolability performances. The friendly graphical user interface allows to test several sensor configurations in order to optimise the diagnostic possibilities. The obtained model can also be used for the simulation of the process and its diagnosis algorithms in normal and faulty c situations. Copyright °2006 IFAC Keywords: FDI, Software, Bond graph, process engineering
1. INTRODUCTION
The system modelling is an important and difficult step in FDI design. Once the model is built, fault indicators can be designed using analytical redundancy or observer based approaches. However, obtaining the model is a difficult task, since complex systems, considered in process engineering, are characterized by the coupling of several phenomena of different natures.
Supervision systems include a set of tools and methods for the control of industrial processes in normal working conditions as well as in the presence of failures. The main activities addressed by supervision systems are monitoring and fault detection, diagnosis and decision making for fault accommodation or system reconfiguration. Fault Detection and Isolation (FDI) procedures are essential for improving the process safety.
Yet, in supervision tasks, human operators do not consider the running process in terms of its mathematical behavior model, but rather in terms of
646
the functions it achieves. The association of Bond Graphs models (as multidisciplinary tool and unified language) with functional models is highly desirable. Furthermore, Bond Graphs allow to develop generic techniques, which can be applied to any thermofluid process, rather than concentrating on developing specific models for specific situations. Those structural, graphical and causal properties of the bond graph tool are then exploited to design a powerful software for supervision in process engineering. The theory behind this application is developed in ((Medjaher et al., 2005), (Ould-Bouamama et al., 2005), (A.K.Samantaray et al., 2004), (Mukherjee and Samantaray, 2001)). This research work is developed and tested in the real processes in framework of an European project. ("CHEM", 2001-2004).
based on the derived residuals. The same IDE (integrated development environment) can be used to create models for newer database items based on existing capsules (derived or hierarchical submodels) or bond graph models or combination of both. The paper is organized as follows : section 1 presents the inputs and outputs of the software and the main functions. The Graphical user interface and how the generic data base is organized are discussed in the second part. in the third section it is shown how the bond graph tool is used for the behavior model and ARRs generation. Section 4 concludes the paper.
2. INPUT AND OUTPUT DATA OF THE SOFTWARE
There has been a significant research on software modelling tools for process engineering, starting with the description of stationary behaviors (Westerberg and Benjamin, 1985), (Briegler, 1989). Dynamic systems are successfully considered in Mathlab/Simulink libraries (Simulink, 1997), or in recent achievements like the modelica modelling language (Elmqvist et al., 1993), (Wollhaf et al., 1996). However, in the first case, the user must introduce the model equations (which means that the model has already been set), while in the second case, the physical phenomena are not explicitly displayed to the user, and the model can not easily be refined, when its structure changes as the result of more assumptions. For chemical process applications, the Odysseo (Object-oriented Dynamic Simulation Software Environment) toolbox of the Prosim software allows to model chemical processes for the simulation of their dynamic behavior, using fixed input/output causalities (Moyse et al., 1999). None of these software derive symbolic state equations, the numerical models being used just for simulation.
From a functional point of view, the main functions (see fig.1) supported by the software are : (i) Create an energetic and control model of the thermofluid process in iconic format. (ii) Derive the dynamic model in faulty and normal operation. (iii) Derive the Analytical Redundancy Relations (ARRs) in the symbolic format and the monitoring ability analysis based on technical specification. (iv) A sensor placement for structural monitoring ability. The software requires as input the architectural model of a plant to start with. Such models are built using a generic plant-items database and are closely related to the Process & Instrumentation Diagrams (P&ID). The objects for these plantitems are derived from certain existing sub-model classes. Optimal sensor placement New sensor architecture
Diagnosability results
Diagnosability analysis
Tchnical specifications
Process
Generate a dynamic and formal models
This paper presents a model builder software for the design of FDI algorithms for thermofluid processes. The methodology is based on the analysis of bond graph, structural and functional models. The system dynamics model, and the residuals associated with Analytical Redundancy Relations (ARRs) are generated under a symbolic format. Thanks to a developed generic item database which consists of a set of predefined process, controllers and sensor classes, and has been incorporated as capsules in the software, the designer can easily build the dynamic and functional models of most thermofluid processes from the Process and Instrumentation Diagram (P&I D) of the plant, and automatically generate symbolic dynamic models and ARRs . It also provides structural monitoring ability (ability to detect and to isolate a faults) analysis and isolation properties
Generate a formal residuals Formal Residuals
P&ID
• Fonctional analysis • BG • Physics
Sensors Data from sensors
• Structural analysis • Causal and structural propeerties of BG
Online implementation
Numerical residuals
Fig. 1. Main activities of model builder These models are then parsed and connection syntax is validated. User specified nomenclature is applied to all known inputs and also to all measured quantities. The fault signature matrix, isolation ability and monitoring ability are output as a single compact matrix. The input model and all the outputs are exported in XML form to be used by other toolboxes in the supervision platform. The outputs are all shown in form of equations and matrices in the Graphical User
647
Interface (GUI). All the sub-models are created through bond-graph modeling language and are linked by the implementation to create a global model in memory to start equation derivation. When the signature faults is generated, the user can add a new sensors graphically to satisfy the imposed technical specifications. Finally the numerical residuals can be implemented on line for supervision.
3. SOFTWARE PRESENTATION The overall Graphical User Interface (GUI) of the toolbox is shown figure 2. Fig. 3. Generic data base The generic database consists of fixed causality and generic causality capsules. While storage tank is modelled as a fixed causality capsule, which has a flow input and effort output, the valve is modelled as a variable causality generic capsule. Valve sub-model capsule can have either effort or flow as the input and the flow through the valve is the output. Capsules are well modelled subsystems with their partly derived equations. Their equations are linked together in the main model and then reduced to result in compact behavioral equations. The internal model of the tank capsule is shown in figure 4. Fig. 2. Graphical User Interface
3.1 Generic plant Item Data Base The system architecture is a component oriented model, which directly describes the process plant as a network of interconnected plant items (valve, pump, pipe, ...). The Piping and Instrumentation Diagrams (P&ID) is used for the visual description of the system architecture. A plant item is a quantity of matter or space that is, or is intended to be, a part of the process plant, e.g. pump, valve, pipe, ... A process plant is a functional assembly of plant items. The user build the P&ID of the supervised process based on implemented generic plant item data base. The generic plant item data base consists of a set of predefined process classes. The structure of the library (based on functional and bond graph theory (Ould-Bouamama, 2003)) is divided into three main parts :the thermofluid process class ((boiler, valve, pipe, tank,...), the energy sources (pumps, heater, ...), the controller class (P, PI, PID, ...) and sensor class which represents the available instrumentation. The generic plant item data base (fig. 3is displayed under icon model library familiar to the industrial user. Based on the P&ID of the process plant, the architectural model can be built.
Fig. 4. Internal Bond Graph Model of the Tank The elements connected to bonds 1, 3, 4, 5 and 6 are external flow input ports. Elements connected to bonds 7-11 are external effort output ports. Bond number 12 is connected to an effort output port which may be connected to measurement devices. All these external ports have an attribute that specifies them as optional ports. Optional ports may or may not be connected in the main model. For example, only three ports of this capsule are connected to other capsules in the illustrated model. The C- element for storage of
648
liquid appears in bond 2. The parameter for the element is C2. One or more representative icons and some help text can be embedded as resources to the capsule object.
FDI system. This has been implemented using a routine to create the Global System Model. The behavioral model for the system is created using preferred integral causality. The part equations of the capsules are linked together to arrive at the compact set of state equations. If the residuals for the global model has already been derived, they are also integrated to these equations. There is an option to derive the equations for faulty systems. Each capsule may be associated with another hidden capsule that has fault models embedded to it. As an example, in the capsule for storage tank, a leakage flow is added as a negative flow source and stored as the corresponding faulty mode capsule. The non-faulty mode capsule is used to derive the residuals whereas the hidden capsule with embedded faults may be used for the behavioral model. If no corresponding faultymode capsule is found, the non-faulty capsule is used instead. Linking residuals with the behavioral model requires automatic creation of certain relations to obtain results of integrated and differentiated terms appearing in the residuals.
3.2 Connectivity rules The connectivity of capsules is checked for validity in the main model. Any errors found there are reported. All post-processing is suspended till the model is syntactically valid. As an example, if two tanks are connected directly without a transportation process such as valve or pipe between them, then the network process is not valid as discussed in(Bouamama et al., 2004). This invalidity is displayed by the software as shown in figure 5. In thermofluid systems, the connections are vector bonds. The input and output side vector dimensions of each bond is checked for consistency of dimensions. A syntactically valid model can be linked to create a global model. While creating global models, consistent input and output ports of capsules are assimilated and a memory model of the junction structure is created.
By default, the Software assigns integral causalities. To assign differential causalities, all measurements are converted initially to reverse sources and all I and C elements are dualized. After integral causality assignment on partially dualized graph, the dualized elements are again dualized to return to normal graph. The global model in preferred differential causalities and inverted measurements is now ready for residual derivation algorithm and monitoring ability analysis (fig.6).
(a)
(b)
Fig. 5. Valid (a) and invalid architectural model
4. ARRS AND BEHAVIORAL MODEL GENERATION It has been implemented using SYMBOLS 2000 ((Mukherjee and Samantaray, 2001)) interface and symbolic manipulation routines. The model is drawn using sub-models only from the set of generic object database based on thermofluid defined classes previously. They contain the bond graph and equation model for the component and interface constraints. These sub-components when connected to create architectural model for a system create syntactically valid models. The software using the power directions, causal properties and vector dimensions for each connection checks the model validity automatically.
Fig. 6. Monitoring ability analysis interface The behavioral model can be used to create a simulation model in the software. The model can also be exported as S-function and then simulated in the Matlab-Simulink environment. Additionally, the Model builder simulator module provides slider variables whereby faults may be simulated interactively by the user at the run-time (fig.7). Such interactive variation of parameters leads to
However, for generation of residuals we need a global model on which causal properties are analyzed and altered to create a proper computable
649
availability of options to try out several combinations and to make subjective observations from the online plotting of the residuals.
Medjaher, K., A.K. Samantaray, B. Ould Bouamama and M. Staroswiecki (2005). Supervision of an industrial steam generator. part II: On line implementation. Control Engineering Practice (CEP) 14(1), 85—96. Moyse, A., L. Jourda, J. M. Le-Lann and X. Joulia (1999). Process, hydraulic and control devices in odysseo, an object-oriented framework for process dynamic simulation.. Monpellier. Mukherjee, A. and A. K. Samantaray (2001). System modelling through bond graph objects on SYMBOLS 2000. In: Int. Conf. Bond Graph Modeling and Simulation. Vol. 33. SCS. Phoenix, Arizona. pp. 164—170. Ould-Bouamama, B. (2003). Bond graph appraoch as analysis tool in thermofluid model library conception. Journal of Franklin Institute 340(1), 1—23. Ould-Bouamama, B., K. Medjaher, A.K. Samantaray and M. Staroswiecki (2005). Supervision of an industrial steam generator. part 1: Bond graph modelling. Control Engineering Practice (CEP) 14(1), 71—86. Simulink (1997). Dynamic System Simulation for Matlab. Math Works. France. Westerberg, L. T. and B. R. Benjamin (1985). Thoughts on a future equation-oriented flowsheeting system. Computers and chemical Engineering 9(5), 517—526. Wollhaf, C., K. Schultz and S. Engell (1996). BaSiP-batch process simulation with dynamically reconfigured process dynamics. Computer and Chemical Engineering supplement pp. 1281—1286.
Fig. 7. Simulation Interface
5. CONCLUSION The software aims at creating an integrated interface to easily create models for plants and derive fault diagnostics specific results from it in symbolic equation forms. It also aims at providing means for the validation of these results through offline simulations and exporting the results for on-line implementation by other toolboxes when they run together in the supervision platform. All these are implemented in a single integrated development environment and significantly reduce the time and cost of modeling and manual derivation of equations and residual.
REFERENCES A.K.Samantaray, K. Medjaher, B. Ould-Bouamama and M. Staroswiecki (2004). Component based modelling of thermofluid systems for sensor placement and fault detection. SIMULATION : Transactions of SCS 80(Issue 7-8), 381—398. Bouamama, B. Ould, K Medjaher, A.K. Samantaray and G. Dauphin-Tanguy (2004). Model builder using functional and bond graph tools for FDI design. submitted to Control Engineering Practice (CEP) journal 13(7), 875— 891. Briegler, L. T. (1989). Chemical process simulation. Chemical Engineering pp. 50—61. "CHEM", European Project (2001-2004). Advanced Decision Support System (DSS) for Chemical/Petrochemical Manufacturing Processes. g1rd-ct-2001-00466 ed. Elmqvist, H., F. E. Cellier and M. Otter (1993). Object oriented modelling of hybrid systems. In: ESS’93 European Simulation Symposium. The Netherlands.
650