DACAS: a Distributed Architecture for Changeable and ... - CiteSeerX

2 downloads 2215 Views 142KB Size Report
IA-Sim a nd a g reenhouse d esign t ool, DAMOCIA-Design. W. § e will describe the use of DACAS in the former one, specifying its distributed model of execution ...
DACAS: a Distributed Architecture for Changeable and Adaptable Simulation 

Bienvenido, J.F.1; Marin, R.2; Guirado, R.1; Corral, A.1 1

Departamento de Lenguajes y Computación. Universidad de Almería. E-04120 Almería (Spain) e-mail: [email protected] 2

Departamento Informática y Sistemas. Universidad de Murcia. E-30071 Murcia (Spain)



Keywords: DAI, distributed simulation, adaptive control, reactive vs deliberated planning.



Abstract:





This p aper d escribes a d omain in dependent d istributed s oftware ar chitecture, D ACAS (Distributed Architecture for Changeable and Adaptable Simulation), that can be applied efficiently onto engineering design and simulation tasks with a high level of changeability. This architecture includes a general Control Block that works as a facilitator using a declarative Behaviour Definition, which acts as a blackboard where it is described the control flow. The functionality of the software is concentrated in a set of Execution and Evaluation Processes. These can ask the Control Process for a modification of the Behaviour Definition (changing the control structure or parameters). With this architecture, we proposed a c ompletely s eparated u ser in terface, f reeing th e ap plication p rocesses o f the execution in a determined machine. This paper describes first this general ar chitecture, highlighting its main characteristics. We have used this architecture with a greenhouse simulation tool, DAMOCIA-Sim a nd a g reenhouse d esign t ool, DAMOCIA-Design. We will describe the use of DACAS in the former one, specifying its distributed model of execution. These tools were developed as part of the project DAMOCIA. 



1.- Introduction

• Product customization. It is necessary an easy versions generation into the design and simulation processes. The different versions must be adequately specified. • Integration of diverse models and submodels in complex systems. We work with complex systems that integrate different subsystems, instead of absolutely separate subsystems. • Integration of w ell know n and pr oved tools and techniques. Obtaining a reduction of the work time, a shorter number of errors (supposed we integrate proved tools) and a high functionality. • Adaptability of the processes. The introduction of new technologies, manufacturing methods, elements and pieces, mathematical and physical models and specific computing technologies requires that our complex design and simulation processes can take these improvements in easily and quickly. • Integration of t he de sign and s imulation processes with the m anufacturing pr ocess. This can be reached deriving, whatever it is possible, the manufacturing software and/or parameters from the design and simulation specifications. 



The ap plication o f th e d esign an d s imulation software techniques has experienced a huge expansion last years, due to the reduction of the computer costs, the relative cost increase of the physical prototypes and the extension of the known methods and techniques to other fields of application. This expansion to new fields and the increasing demand of results have put on light or underlined a set of problems that the most classical simulation and design system can not resolve straightforward, between these are:







• Reduction of the response time to the market changes. It is necessary to incorporate the new technologies and new consumer requirements in the shortest time possible. • More accurate simulations or detailed designs. (E.g. reducing the size of the finite elements in materials simulation.) • Standards ac complishment. There is an increasing set of “de jure” standards that generates the necessity o f s ome regional customization. The in time “de facto” standards adoption requires a great capacity of adaptation.









All these requirements can be summarized on these few technical points:

♦ ♦ ♦ ♦

Reduced development time. Intensive calculus processes. Customization (variability) of products. Adaptability to changes (of technology, norms or infrastructure). ♦ Integration of external tools. ♦ Integration on global production schemes. 

2.- Proposed architecture One of the possible ways of facing the general challenge previously described, it is the use of new distributed architectures with these characteristics:

1) Elaboration of a formalized and detailed definition of the diverse production objectives and their correspondence with the greenhouse conditions required. 2) Development of a set of alternative systems of acquisition and control adapted to prototypes with different automation levels. 3) Implementation of a greenhouse design tool, which would elaborate customized construction projects including sets of normalized plans, budgets and structural analysis. This tool generates a formal description of the proposed structures that can be use by the simulation tool. 4) Implementation of a g reenhouse s imulation tool, centered on the simulation of the structure as a radiation captor. Its specific objective is to compare different structure proposals before their building. 5) Building of a set of specific models of structures related with the agronomic optimums and ready for their commercialization.

• Differentiation and functional independence between th e d ifferent s ubmodels an d subsystems, using sets of independent software processes. • Parameterization and reusability of the software processes related with th e d ifferent submodels an d subsystems. • Plain integration of the different subsystems and submodels processes using soft links based on the interchange of standard data structures. • Multilevel distribution of the processing (hierarchy of executive software modules). • Declarative control, possible via blackboards. • Main deliberative control with reactive options, some modules can change the previously established behaviour (modifying the blackboard data). • Alternative subsystems and submodels applicable under d ifferent c onditions, generating multiple valid results. • Separate user interface, isolating the data demanding and results presentation on a given system and freeing the executive modules of the interface conditions and its execution environment. • Machine independence of the different work modules, so it is possible to increase the computing power w ithout alte rations in to the executive modules. • Sets of default values, offering a wide product variability into simple design and simulation processes. • Possible incorporation of supervisor and counselor modules. Some of these can include specific experts systems and some sort of intelligent behaviour (mainly on the design and evaluation processes). • Integration of external tools (as distributed elements of work), using the adequate interfaces. • Standardization of the outputs, using whatever would be possible external standards, making possible th e in terconnection w ith later processes (manufacturing). E.g. DXF plans.

It was working on the worklines 3 and 4 where we found all the requirements previously described in a general way. Next, we will describe how we have faced this situation in a general way using a distributed architecture, DACAS, and how it has worked implementing DAMOCIA-Sim.

Summing up, we will distribute the functionality on separate modules (integrating specifically in some cases external tools), that could be use in alternative and reusable w ays, is olating th e d ifferent s ubsystems and submodels. These modules will be linked between them via standardized data structures, making an integrated

These were the conditions we found working on the project DAMOC IA, f inanced b y t he UE (ESPRIT Special Action P7510 PACE) and the Spanish Ministry of Industry (PATI PC-191). This is an interdisciplinary project w hose main o bjective is to imp rove the behaviour of t he p lastic c overed greenhouses used mainly in the Mediterranean climate regions, and specially the Almeria province. (More than 20,000 h of greenhouses, the world greatest concentration.) 









With a global objective of a “ Computer aided design of customized and controlled greenhouses, and the development of specific marketable prototypes”, the project was structured on these worklines:





















• Working Data Space. It stores all the domain specific problem data. These can be distributed onto several machines and hardware elements. These distributed working blocks can reside on memory, disk or any other storage device. The control process offers the addresses of their working blocks to the execution and evaluation processes. This working data space is filled with the Interchange Data S tructures, which constitute the data blocks that interchange the different executive processes (Execution and Evaluation Processes). These have standard formats, being used in some cases by different alternative processes. The Control Block recognizes o nly th eir e xistence, w ith a n ame, location and size, managing the use of the Working Data Space. The interpretation of these IDS is taken ahead by the Executive and Evaluation Processes. • Execution p rocesses. They apply the domain specific tasks to sets of input data stored into the working data space (IDSs). They are formally described by using a clear input-process-output structure. Internally, an execution process can be implemented directly as a computer process or as a higher level execution block, constituting in this case a multilevel distributed system. • Evaluation processes. They are execution processes that modify the control data or the behaviour definition. • User interface. It provides the input data, activates the execution block and shows the execution evolution and primary results.

environment. The separation of the user interface frees the computing modules of the input data formats and the interacting environment. The use of a general declarative control eases modifications and permits both a deliberative an d r eactive c ontrol. Th e e xistence of computing alternatives (with different computation models of a problem implemented) and default values alternatives eases the elaboration of multiple product alternatives and their adaptability.





In response to this idea, we propose a general distributed architecture DACAS (Distributed Architecture for Changeable and Adaptable Simulation) that is described in Figure 1.









Figure 1. DACAS General Architecture. The main elements of the proposed architecture are: • Control Process. It is based onto a collection of Behaviour Definitions, selecting one that it is instantiated applying the control parameters defined into a Control Data Block. It evaluates the sequence of execution of the different evaluation and execution processes and their points of execution, sending them the location of their input data and assigning them the required storage resources. • Behaviour Definitions. It holds a declarative representation of a set of problem-solving methods using a specific language that it is interpreted at r un time. This language includes capabilities of parallel execution, evaluation of alternative methods and correctness tracking. • Control data block. It includes the control parameters used f or in stantiating th e b ehaviour definition. These data are supplied, first, by the user interface, being modified by th e C ontrol P rocess or the Evaluation Processes when it is required. 







The Control Data Block and the Behaviour Definition act as a control blackboard where is annotated the control flow to be executed. The former one is constituted by a set of parameters and the later by a set of extended rules. (It is configured like an interpretable program for the control of the global process). The initial and alternative Behaviour Definitions are assembled analyzing the functionality required and

Figure 2, including sequential, alternative, parallel and cyclic execution of the different Execution Processes.

The proposed architecture presents two ways of realtime control adaptability: a) Reactivity: the choice between a c ollection o f alternative problem-solving methods changes in response to modifications of the control parameters; these are executed by an evaluation process; b ) Deliberated planning: the evaluation process c hanges th e p lan its elf b y modifying the behaviour d efinitions, th at is , th e p rocess s equence, their execution criteria (point of execution or resources assignment), and the available alternatives or the selection criteria. Moreover, the system performs dynamic allotment of the computer resources, using the rules stored into the declarative behaviour definition and the resources availability information stored into the control data block. The main assigned resources are: location of the working data space blocks and CPU where the different execution or evaluation processes will be executed. 





3. Application case Here we are going to describe an application case of the proposed distributed architecture, in this case the actual version of the DAMOCIA-Sim simulation tool. The main objective of this tool is modeling the radiation behaviour of s pecific g reenhouse s tructures i n a general way. This tool is complementary with DAMOCIA-Design, a specific greenhouse projects generator. Figure 3 shows the tool general diagram. 

Figure 2 . G raphic r epresentation of the processes behaviour possibilities.



Design Tool Specific Greenhouse Definitions

Simulation User Interface

Radiation Plans

Radiation Simulator Experiment Definition

Solar Radiation Simulation

Shadows Simulation

Compatible Experiment Data Model

Output

Plastic Simulation

Forms Builtder

Outputs Demand



Figure 3. DAMOCIA-Sim General Diagram. elaborating a diagram that shows the possible alternatives of execution. This graphic representation is transformed in the behaviour definition (defined with ordered rules and execution blocks that acts as a declarative control program). The main elements we use with this g raphic r epresentation ar e s howed in 

Reports



The radiation simulator uses as input the desired simulation profile, th at it is called “Experiment Definition”, and the formal definition of the greenhouse generated by the design tool.

The experiment definition includes the selection of the greenhouse structure to be simulated (picked over a list of specific available structures, previously defined with DAMOCIADesign), the period of simulation (e.g. a set of days distributed along a year, a complete farming campaign, an instant , etc.), the climatic conditions (there are two possibilities, we can u se e xperimental s ets o f d ata obtained with meteorological stations, or sets of simulated data obtained using multiple weather

all the results of the simulation process. These data are showed to the user using a special module/process of presentation, th e “ Compatible O utput Forms Builder”, that receives the required presentation format from the user interface (“Output Demand”). Th is is a g eneral module that is used with other tools of the DAMOCIATools set. The presentation of the simulation results generated by the COFB is similar to that generated with the experimental system used to evaluate the results of the simulation.

patterns) an d th e s imulation conditions like the preselection b etween alte rnative s imulation mo dels (when there are more than one) or the size of the different elements when we use techniques of simulation based on finite elements. It includes, of course, the demanded results, like the global accumulated radiation, a map of the accumulated radiation over a given surface into the greenhouse or a set of maps of the incident radiation over a surface along a time period. 







Extraterrestrial radiation Submodel A

Sun position Submodel B



The Radiation S imulator h as been implemented u sing th e proposed D ACAS a rchitecture. In this case, we started elaborating a functional diagram of the global process, defining all the subprocesses required to carry out the simulation. In this schema, we analyzed the Interchange Data Structures that are interchanged between the different Execution Processes. Figure 4 shows part o f t his diagram.

XA1

Computing of curve surfaces

C1

Greenhouse structure

XA2

XA3

Sun position

Radiation

Kind of plastic

XA4

Computing initial shadow characteristics

Shadows

C2

Computing of geometric characteristics of the greenhouse surfaces

XA5

Geometric data

Updating of the relation table surfaces

XA6

D1

Computing the projections that generate shadows

C3

Computing the bitmaps of initial illumination of surfaces C4



Plastic 1

F1

Plastic 2

F2

* * *

Plastic n

Fn 

D2

Updating the illumination bit maps

Computing the plastic transmissivity F0

From this, we fixed the different subsystems to take account of, analyzing the possible XA7 XA8 alternatives (of physical model, Illumination implementation o r used Transmissivity bit maps resources), l ogical sequence, possibilities o f p arallelization and cycles required (as it is XA13 u sual w orking w ith f inite Radiation map of Computing the internal Shadows the analyzed elements). Afterward, we radiation on a surface E G surface determined the set of executive processes to b e imp lemented as in dependent s oftware Figure 4. Part of the functional diagram of the modules, simulation agents, listed in table 1. Radiation Simulation. (Surface analysis). With this, we assembled a graphic representation of the The greenhouse description is a precise representation proposed B ehaviour D efinition, w hich is shown on of all the elements that constitute a specific structure. Figure 5. On this diagram, we can observe multiple This is located in the space (with latitude and an selections (as that of the diverse possible plastics orientation). The description includes the exact simulation models, the use of experimental vs placement o f all th e g reenhouse elements, like the calculated sun radiation values or the absorption concrete foundations, piles, wires, plastic fastening models), some parallelizable processes (as the meshes, mosquito nets, cover plastics, doors, pipes, evaluation of the shadows of the solids and the roller fastenings, etc. The greenhouse definition is radiation maps on the external surfaces), and cycles. formalized using a declarative specific language, which (The evaluation of the radiation into the greenhouse by is generated automatically by the design tool. successive reflections and refractions on the surfaces.) Bit maps





D3



















The radiation simulation output is the “Experiment Data Model”, a file (in disk or memory) that contents

This is a graphical representation of the basic Behaviour Definition stored on memory and managed





by the (general) Control Process. Next, we incorporated the Evaluation Processes that permit to change the previously assigned control flow, changing

EXPERIMENT DEFINITION

P1 S1

XP1 &

XP1 XP2 XP3 XP4 XP5 XP6 XP71,2,n XP8 XP9 XP10 XP11 XP12 XP13 XP14 XP15 XP16 XP17 XP18 XP19 XP20 XP21 XP22 XP23 XP24 XP25 XP26 XP27 XP28 XP29 XP30 XP31 XP32 XP33

!





























#

"

!

S1 S2 S3 S4 S5 S6 S7 S8 ES1 ES2 ES3 ES4 ES5 ES6 ES7 ES8 %

"



!

$



Sun position submodel Extraterrestrial radiation submodel Experimental data adjustment Direct radiation estimate Computing of curved piles Computing of curved surfaces Different behaviour models of the plastic Computing initial shadow characteristics Computing of the surface geometric characteristics Updating of the relation table surfaces Computing the projections that generate shadows Computing bitmaps of surfaces initial illumination Updating the illumination bit maps Percentage map computing of external wire mesh Percentage map computing of the internal wire mesh Percentage map computing of the piles near surfaces Computing the radiation map in the external side Computing the external reflections between surfaces Computing the radiation map in the internal side Discretizing each surface in macro-elements Classification of the piles of the internal structure Computing the absorption of the volumetric elements Macro-elements projection on the receiving surfaces Volumetric models Absorption by layers Uniform absorption No absorption Calculation of the direction of the reflected radiation Radiation distribution among the finite elements Calculation of the radiation reflected by each element of the receiving surfaces Compaction of the generated macro-elements lists Radiation map generated from the macro-elements Interpolation of the discontinuities generated on the radiation map by the surfaces edges Selection of the radiation data type Selection depending on curved posts Selection depending on cover curved surfaces Selection of the plastic kind Selection depending on external reflections Selection depending on volumetric absorption model Selection of iterative task ending Selection of the absorption model Global radiation results Piles computing results Surfaces computing results Transmissivity results by type of plastic Reflection computing results Absorption for each finite element Inner reflections between the surfaces Final result of the absorption models

&

XP2

XP3

ES1 &

XP4

EP1

P2 P3 ' ' '

S2

S3 & &

&

XP5

XP6

XP71

&

XP8 P4

&

&

XP9

XP10

P5 EP4

EP3

'

P6 &

P7 &

&

XP16

XP15 &

&

S5

XP12 &

XP14

XP11

ES5

EP7

EP5 &

XP17

XP13

EP6 '

S6 &

XP18

ES6 EP2 &

XP19 P7 &

'

&

XP22 &

XP23

S7

XP20

ES7 EP7 R1 &

XP21

P8 '

&

XP24

&

S8 &

XP25

XP26

&

XP27 &

XP28

ES8 &

XP29 EP8 &

XP30 &

XP31

R1 &

XP32 &

XP33

EXPERIMENT DATA MODEL 

Table 1. Possible independent subsystems. 

S4

XP72 ES4

ES3

ES2

&

Figure 5. Behaviour Definition Diagram.

&

XP7n

different parameters (as that of the selection blocks) or the main control flow (offering in this case alternative sections of control flow). In our case, we detect these evaluation processes: EVP1

• The control in these distributed architectures can be improved using declarative descriptions as blackboards w here t he d ifferent p rocesses ca n act over the global control flow (with an adequate centralized supervision). This supposes the possibility of a r eal-time control ad aptability w ith deliberative and reactive planning capacities.

Processes p arallelization d eactivator. I t a cts w hen there are no parallel resources available. 1EVP2 Macroelements u se d eactivator. W hen it detects a great discontinuity of the radiation values over the greenhouse surfaces, it orders to use directly the elemental finite elements. EVP3 Reducer of the number of finite elements. It evaluates if it is p ossible o r n ot to evaluate the proposed elements with the available resources, increasing the size of the finite elements. EVP4 Change of the plastic modeling alternative. It acts when the evaluation time with the selected model surpasses that permitted by the available resources. EVP5 Reduction of num ber of cycles of t he i nner r adiation evaluation. It increases the threshold of the radiation evaluation into the greenhouse when it evaluates a too long computing time. 

(





• DAMOCIA-Sim and DAMOCIA-Design have proved to give the r equired r esults, showing a wide functionality and adaptability. We have incorporated later new greenhouse structures and physical models with a reduced effort. 

!

)



!

(



(

*



The g eneral s imulation p rocess h as b een p roved w ith some work conditions effectively. We are now evaluating the results versus the experimental data. 



This Radiation Simulator is seen like a process (that actives other processes) from the DAMOCIA-Sim tool point o f v iew. A s a matte r o f f act, the integration between the user interface, the Radiation Simulator and the Compatible Output Forms Builder constitutes another simplified application of the DACAS architecture (in an upper level) with a extremely reduced Behaviour Definition (a sequential execution of the Radiation Simulator and the COFB for each simulation demand).

+



4. Conclusions and further works From th e ac complish w ork, the following conclusions can be highlighted: • The use of distributed software architectures can improve the computer aided design and simulation processes, r educing th e d evelopment time , increasing the adaptability of the processes and products, mak ing e asier th e d evelopment o f multiple versions, increasing the computing capacity and integrating external tools. 



• Our distributed architecture DACAS has proved to implement effectively this work method. The proposed ar chitecture is ap plicable to multiple simulation and design environments. 

,

Future issues include a backtracking utility for process results reuse, the implementation of an evaluation process f or r esources as signment b y au ction, the application of DACAS in other fields as the experimental data treatment and the integration of reduced e xperts s ystems as e valuators an d counselors on the design and simulation processes.

5. Bibliography 1) Bienvenido, J.F. et al.: DAMOCIA-Sim, a generic tool for radiation simulation into mild winter region greenhouses. Proceedings of the First European Conference for Information Technology in Agriculture, 1997. 2) Cuena, J.: Knowledge architectures for real-time decision support. In Second Generation Expert Systems, JM. David, JP. Krivine and R. Simmons (eds.), Springer-Verlag, 1993. 3) Guida, G. and Zanella, M.: Knowledge-based design using the multi-modeling approach. In Second Generation Expert Systems, JM. David, JP. Krivine and R. Simmons (eds.), Springer-Verlag, 1993. 4) Kirchner, T.B.: Distributed processing applied to ecological modeling. In Simulation Practice and Theory, Vol. 5-1, 1997. 5) Kurihara, S., Aoyagi, S. and Onai, R.: Adaptative Selection of Reactive/Deliberate Planning for the Dynamic Environment. In Multi-Agent Rationality, M. Boman and W. Van de Welde (eds.), Springer Verlag, 1997. 6) Lalis, I. and Menhart, P.: Object oriented toolset for sequential and distributed simulation. In Proc. European Simulation Multiconference ESM’95, Praha 1995. -

Suggest Documents