An AI-based shell for linking thermal and form-making considerations

Post on 21-Jan-2023

1 views 0 download

Transcript of An AI-based shell for linking thermal and form-making considerations

Ž .Automation in Construction 8 1999 455–462

An AI-based shell for linking thermal and form-makingconsiderations

Emmanuel-George Vakalo ), Ali M. Malkawi, Samir S. EmdanatCollege of Architecture and Urban Planning, The UniÕersity of Michigan, Ann Arbor, MI 48109-2069 USA

Abstract

Over the past few years, our team has developed several computer-based models in the areas of architecturalform-making and thermal analysis. These programs were designed to deal with specific problems and use a range oftechniques including machine vision, knowledge-based systems, and artificial intelligence techniques. Recently, a projectthat integrates these systems was initiated. Its objective is to design an intelligent computer shell that forms the basis for thisintegration in the domain of architecture. The paper discusses the development of the shell and its use to analyze and studyarchitectural form and its determinants. The shell accommodates modules that link the morphological structure of

Žarchitectural design with more of its determinants e.g., structural, acoustical, and lighting considerations, as well as code.requirements . The paper presents and discusses the background of the shell, its structure, its methods of knowledge

representation, and an example of its use. q 1999 Elsevier Science B.V. All rights reserved.

Keywords: AI-based shell; Form-making; Thermal analysis

1. Introduction

ŽArchitects use representations of buildings e.g.,.drawings and models to reason about the different

aspects of a design problem. They manipulate theserepresentations using compositional principles such

Ž .as symmetry e.g., translational, rotational androrŽ .Cartesian transformations e.g., similarity, topology .

The resulting compositions exhibit some desirableformal attributes such as balance, symmetry, axiality,proportion, rhythm, and hierarchy. These composi-tions should be functional and livable. They shouldbe structurally stable and thermally comfortable. Al-

) Corresponding author

though suited for solving the former type of problem,architectural drawings are incapable of capturing,among other things, the knowledge that relates to thebehavior of a form statically or its level of comfort.

Lacking a comprehensive method of representa-tion for their buildings, architects are required toincorporate and integrate a large body of knowledgeand knowledge sources to generate a satisfactoryartifact. In most cases, the complexity of the problemand the contradictory nature of its constraints makesit difficult for designers to accommodate all thedifferent requirements of a design problem.

Some of the problems that arise during the designof a building are problems that relate to construction,layout, movement, structural analysis, thermal per-formance, code requirements, acoustics, lighting,mechanical systems, and aesthetic preferences. Gen-

0926-5805r99r$ - see front matter q 1999 Elsevier Science B.V. All rights reserved.Ž .PII: S0926-5805 98 00092-2

( )E.-G. Vakalo et al.rAutomation in Construction 8 1999 455–462456

eral integrated computational tools for architecturalform-making are difficult to achieve because the

w xproblems they are associated with are ill-defined 2 .A building can be described at a number of levels ofabstraction. For instance, one can discuss a buildingin terms of its physical elements such as walls,columns, doors, windows, floors, and roofs. Anotherway of discussing a building is to describe its con-struction technologies such as structural systems andmechanical systems. In addition, a building can bedescribed in terms of its performance and use-type.

In architecture, computer applications have tar-Žgeted its production aspects e.g., creating com-

.puter-aided drafting and design tools . These applica-tions do not address fundamental form-making prob-lems, such as reasoning, to build computer-aidedsystems. Further, computers have rarely been usedfor analyzing the structure of architectural form. Inmost cases, morphological studies have been con-ducted manually. This can be attributed to two rea-sons. First, the people who have been traditionallyinterested in conducting such studies have not beencomputer-literate. Second, available computer tech-nologies as well as programming techniques madethe use of computers in the conduct of such studiescumbersome, if not impossible.

During the past two decades there has been no-ticeable advances in both computer technology andprogramming techniques. Of particular interest arethe advancements in the area of artificial intelli-gence. Artificial intelligence problem-solving tech-niques and programming methods make possible theexploration of solutions for complex activities suchas form-making. There are two reasons for this.First, they reduce the space and time complexityissues involved in solving such complex problems.Second, they can derive solutions to complex prob-lems by facilitating the exploration of the structureof an object through knowledge representation tech-niques.

Morphological analysis can be enhanced consider-ably through the use of such programming tech-niques. Once an adequate representational languagefor the attributes of a form is achieved, it will bepossible to create integrated knowledge-bases thatdescribe the relationship between the morphologicalattributes of an architectural form such as symmetry,hierarchy, and axiality, and its determinants such as

thermal behavior and structural performance. In ret-rospect, it would be possible to adapt many of theexisting artificial intelligence techniques, heuristicsand search methods to describe, analyze, and evalu-ate form-making solutions, as well as to explore newones.

In thermal design, computers have been usedprimarily to conduct simulations. Computer simula-tions aimed at optimizing the thermal performance ofa building require constructing a model of the build-ing’s envelope and subjecting it to analysis that will

w xdetermine its performance 15 . In most cases, theresults of these computer-based simulations have notbeen directly applicable to form-making. To enhancethe conduct of simulations with an evaluative com-ponent, several studies took advantage of knowl-edge-based systems that emerged out of the artificial

w x w xintelligence field, Brown 1 , Gero and Qian 3 ,w x w x w xHaung and Degelman 4 , Hitchcock 5 , Malkawi 7

w xand, McConkey and Case 8 . The systems theseauthors proposed, do not act as independent agents.In other words, they are not intended to replacehuman experts in a given situation. Rather, theyfunction as intelligent assistants augmenting or sup-plementing human expertise while, at the same time,increasing productivity. The integration of such sys-tems with systems that analyze and describe themorphological structure of buildings will allow de-signers not only to explore their form-making ideas,but also to get important feedback and advice on thethermal performance of the architectural forms theyare considering.

2. The conceptual structure of the shell

The shell is structured to integrate a variety ofbuilding representations. Initially, these include amorphological analysis module and a thermal predic-tion module. Other modules for handling structuralbehavior, lighting, acoustics, among others, are to beincorporated at later stages. The shell is designed tomake the most use of object-oriented programmingtechniques. This includes schema representations thatallow object inheritance and hierarchical class struc-ture to be modeled. The underlying hierarchical classstructure of the shell captures the dependency rela-

( )E.-G. Vakalo et al.rAutomation in Construction 8 1999 455–462 457

tionships between the elements of a building andŽ .their properties Fig. 1 , some of which are described

below.In the shell, architectural compositions are de-

rived from design elements such as points, lines,planes, and volumes. The class point is formallydefined by a direction and an angle or by threecoordinates in a Cartesian coordinate system. It indi-cates a position in space. The class Line is derivedfrom the class Point. It is a point extended in somedirection. A line inherits the position of one of itsendpoints from the class Point. In addition, the classLine has the properties of length, direction, thick-ness, and orientation. A plane is a line extending in adirection other than its extension. Finally, a volumeis a plane extending in the direction of its normalvector. The class Volume has the properties of length,width, height, form, surface, orientation, and posi-

tion. Volume encloses space. In addition to theirgeometric properties, spaces emerge from the dispo-sition of the other class elements such as walls,columns, and slabs.

Class elements have a number of associated prop-erties that are also represented in the class hierarchyŽe.g., texture, color, structural properties, and ther-

.mal characteristics of materials . This allows externalmodules to use the property values in their evalua-tion of the various aspects of a design including itsthermal behavior. This representation is carriedthroughout the structure of the shell to capture thedependency relationship between the elements of anarchitectural form.

A large portion of architectural knowledge isdocumented in the form of drawings of plans, sec-tions, and elevations. The shell attempts to transformthese drawings into vector drawings based on points,

Fig. 1. The class hierarachy.

( )E.-G. Vakalo et al.rAutomation in Construction 8 1999 455–462458

lines, planes, and volumes that are compatible withmost modeling environments as well as to provideanalyses of the morphological properties of theseforms. Moreover, it allows for additional informationabout a building that includes material properties tobe integrated with the morphological properties toallow the inclusion of thermal performance mea-sures. By implementing this representation that inte-grates various aspects of the knowledge concerningarchitectural form, design development and evalua-tion become more efficient in terms of time andquality.

3. Implementation

The shell is based on a set of functions associatedŽ .with a flexible interface see Fig. 2 that allows

external communications with stand-alone objectsŽ .using Object Linking and Embedding OLE . The

shell is being designed for the MS Windows 95rNTenvironment and is implemented using VB 5.0 andVisual Cqq 5.0 utilizing OLE automation. It in-

corporates a sketch pad, a computer vision module, ashape grammar generator, and a thermal analysismodule. Communications between the shell and itsmodules as well as with other CAD applicationssuch as AutoCAD are accomplished through OLE.

3.1. The sketch pad

The sketch pad component of the shell readspixel-based images and allows the user to modifytheir properties such as color tables, size, orientation,brightness among others. It also supports some ele-mentary image processing operations such as noisereduction through median filtering. In addition, theuser can create new bitmap images. As an OLEcomponent, this module communicates with othersoftware such as AutoCAD and Windows Paint pro-grams.

3.2. The Õision module

This module works on representations created oropened using the sketch pad module and allows theuser to extract the vector data of these images. In

Fig. 2. The shell interface.

( )E.-G. Vakalo et al.rAutomation in Construction 8 1999 455–462 459

Fig. 3. Sample output from the vision module.

addition, this module structures the extracted infor-mation in a form that is compatible with the gram-mar module and the thermal analysis module. Someof the vision techniques that are implemented in-clude boundary detection, corner detection, room

w xdetection, and openings detection 14 . After thedetection process, the user can modify the vector

Ždata to add attributes e.g., material properties, wall.details interactively. Moreover, the user will be able

to add or delete extracted elements, or edit theirŽ .alignment among other things Fig. 3 .

3.3. Shape grammar generator

Universal shape grammars and universal architec-tural grammars are used for the analysis and deriva-tion of the morphological structure of architecturalform. They assume that the morphological structureof architectural form can be analyzed and derived attwo levels of abstraction: the level of the geometricstructure and the level of the spatial structure. Linesconstitute the elements of the geometric structure,while walls and columns constitute the elements of

w xthe spatial structure 6 . The grammar module is auniversal shape grammar that interprets the extractedvector data according to a set of pre-defined rules toderive their adjacency graphs and derivation trees.The adjacency graphs capture the proximity relationsbetween the spaces of a building formally. Thederivation trees record the decision structure for theapplication of the rules in a hierarchical manner.

ŽSpatial relations such as corner conditions wall-to-.wall relations are also encoded.

3.4. Thermal analysis module

The thermal analysis procedures utilized in theŽ .shell use the Transfer Function Method TFM that

was first introduced by Mitalas and StephensonŽ . w x1967 10,12 . This procedure is based on responsefactors and the interplay of heat exchange between

w xvarious surfaces and sources of heat gain 11 . Trans-fer functions are based on two concepts: the Conduc-

Ž .tion Transfer Factors CTF and the Weighting Fac-Ž .tors WF . The CTF are used to describe the heat

flux at the inside wall, roof, partition, ceiling or flooras a function of previous values of the heat flux andprevious values of inside and outside temperaturesw x9 . The WF are used to translate the zone heat gain

w xinto cooling loads 13 . These functions are derivedmainly from response factors. These response factorsare defined as an ‘infinite series that relates a currentvariable to past values of other variables at discretetime intervals. A transfer function converts the theo-retically infinite set of response factors into a finitenumber of terms that multiply both past values of thevariable of interest and past values of other vari-

w xables’ 9 .

4. Use

The shell can be used to analyze the morphologi-cal structure as well as the thermal performance ofan existing architectural design or to evaluate adeveloping form-making solution. In the former situ-ation, designers can use the computer vision moduleto convert an architectural drawing into a model thatconforms to the shell’s class structure. In the latter,designers can use the shell to construct new form

Ž .solutions either in pixel form bitmap or directly asvector data. In both cases, once the internal represen-tation is constructed, the other modules of the shellŽi.e., the thermal analysis module and the shape

.grammar module can be used to derive additional

( )E.-G. Vakalo et al.rAutomation in Construction 8 1999 455–462460

information about the form solution such as its adja-cency relationships or thermal performance.

The results of the analysis are integrated throughthe class structure underlying of the shell and design-ers can interactively access and modify the attributesof a particular design element and observe thechanges that it would induce on the overall designperformance. Alternatively, designers can modify

Žsome of the evaluative criteria i.e., the thermal.comfort measures of a particular space and the shell

will identify the problematic areas. This will result ina better understanding of the emerging design solu-tion and its performance which, in turn, will allowusers to make more intelligent form-making deci-sions.

5. An example

Using the system, designers can access and mod-ify the attributes of a particular design element inter-actively and observe the changes it would engenderon the overall design performance. Alternatively,designers can modify some of the evaluative crite-

Žrion i.e., the thermal comfort measures of a particu-.lar space and the shell will identify the problematic

areas. This will result in a better understanding of theemerging design solution and its performance which,in turn, will allow users to make more intelligentform-making decisions.

As an illustration of the use of the system as ananalysis module, a building was selected. It was

Ž .modeled in AutoCAD r.13 see Fig. 4A and thentranslated into the system. The computer vision mod-ule was used to analyze the morphological structure

Fig. 4. The vision module.

Fig. 5. Building morphological properties.

of its plans. Spatial information was extracted andrecorded into the shared database that is maintainedwithin the shell of the main module. This informa-tion includes corner, wall, and room detection. PartB of Fig. 4 illustrates the original plan of the image

Ž .turned into a raster-based i.e., a bitmap representa-tion to be used in this module. The first operation isto recognize the walls. The pixels that make up the

Ž .plan are expanded Fig. 4C to plug the openings.Then the plan is reduced back to its original size.The result of this operation is a representation of the

Ž .plan that contains only the walls see Fig. 4D . Aw xpixel difference operation 14 is then carried out on

the original plan and the one shown in Fig. 4D toobtain the openings. At this stage, the boundary of

Ž .the plan is detected Fig. 4E and used for roomdetection. This is done by applying the pixel differ-ence operation between that and the plan shown inFig. 4D.

The detected information is now stored in theshell database and the user can view this information

Ž .through the interface see Fig. 5 . In addition to this,the user has several options to evaluate. For exam-ple, when the user chooses to evaluate the spatialstructure of the building, the system will display its

Žanalysis pertaining to the corner conditions Fig..6B,C,D and the adjacency relationships among theŽ .rooms see Fig. 6E,F .

To evaluate the design thermally, the shell pro-vides the user with an intuitive interface to assignnon-geometric properties to the extracted elements ofthe design. Information about the location and orien-tation of the building is entered through the interface.

( )E.-G. Vakalo et al.rAutomation in Construction 8 1999 455–462 461

Fig. 6. Example of the use of the system.

The user can also click on an element and modifythe properties associated with it. This is establishedby utilizing the vision module’s detection that hasalready been stored in the shell’s internal representa-tion. For instance, if the user double clicks on a wall,a space, or an opening, the system retrieves all theinformation from the database about the selectedelement and displays the appropriate properties dia-log for the user. Fig. 7 shows the material propertiesdialog of a wall or a wall or a set of walls.

After all the properties are assigned, the user canchoose to run the thermal simulation to check thethermal performance of the building. Fig. 8 illus-trates the room temperatures and their associated

Fig. 7. The material selection dialog for a wall element.

Fig. 8. Building thermal properties.

thermal loads. At any time, the user can ask formorphological or thermal analysis and the systemwill display an instant feedback.

6. Conclusions

The shell contributes methodologically to the sys-tematic and efficient derivation of architectural form.Specifically, the shell constitutes an initial attempt tobuild high level descriptions for languages of archi-tectural form. This entails building an intelligentintegrated computational system capable of recogniz-ing and reasoning about selected morphological at-tributes and determinants. This can help designersexplore and test their form-making ideas. Moreover,the shell integrates morphology and thermal perfor-mance uniquely. The modularity of OLE and itsinternal structure has allowed the integration of inde-pendent modules that specializes in different aspectsof the design problem. Breaking the design processinto these modules allows for the study of theireffects independently and in relation to other mod-ules.

It is envisioned that, when fully implemented, theshell will be useful, robust, efficient, and more im-portantly, capable of being used by practitioners,researchers, educators, students, critics, and othersnot only to analyze the morphological structure ofbuildings but also to explore their thermal behavior.

( )E.-G. Vakalo et al.rAutomation in Construction 8 1999 455–462462

References

w x1 G.Z. Brown, Desirable interface characteristics of knowl-edge-based energy software used by architects, ASHRAETransactions, vol. 91, Pt. 2, 1990, pp. 550–555.

w x2 C. Eastman, A data model for design knowledge, in: G.Ž .Carrara, Y.E. Kalay Eds. , Knowledge-Based Computer-

Aided Architectural Design, Elsevier Science B.V., 1994, pp.95–122.

w x3 J.S. Gero, L. Qian, A design support system using analogy,Ž .in: J.S. Gero Ed. , Artificial Intelligence in Design 92,

Kluwer Academic Publishers, Netherlands, 1992, pp. 795–813.

w x4 T.K. Haung, L.O. Degelman, ENERGRAPH, an Energy-Ori-ented Computer Assisted Design Package, ACADIA, Octo-ber, 1992.

w x5 R.J. Hitchcock, Knowledge-based system design guide tools,ASHRAE Transactions, vol. 97, Pt. 2, 1991, pp. 676–683.

w x6 S.-R. Liou, A Computer-Based Framework for Analyzingand Deriving the Morphological Structure of ArchitecturalDesigns, PhD Dissertation, The University of Michigan, AnnArbor, 1992.

w x7 A. Malkawi, Simulation and reasoning: Intelligent buildingthermal problem detection, in: Proceedings of the 4th Interna-

tional Building Simulation Conference, Madison, WI, 1995,pp. 176–182.

w x8 I. McConkey, M.P. Case, Artificial intelligence as integrationtechnology, ASHRAE Transactions, vol. 97, Pt. 2, 1991, pp.761–766.

w x9 F.C. McQuiston, J.D. Spitler, Cooling and Heating LoadCalculation Manual, 2nd edn., American Society of Heating,Refrigerating and Air Conditioning Engineers, Atlanta, 1992.

w x10 G.P. Mitalas, D.G. Stephenson, Room thermal response fac-tors, ASHRAE Transactions, vol. 73, Pt. 1, 1967.

w x11 T.B. Romine, Jr., Cooling Load Calculation: Art or Science?,ASHRAE Journal, January, 1992, pp. 14–24.

w x12 D.G. Stephenson, G.P. Mitalas, Cooling load calculations bythermal response factor method, ASHRAE Transactions, vol.73, Pt. 1, 1967.

w x13 E.F. Sowell, D.C. Chiles, Zone descriptions and responsecharacterization for CLFrCLTD calculations, ASHRAETransactions, vol. 91, Pt. 2A, 1984, pp. 179–200.

w x14 C. Terzidis, Computer-Aided Extraction of MorphologicalInformation from Architectural Drawings, PhD Dissertation,The University of Michigan, Ann Arbor, 1994.

w x15 K.W. Tham, H.S. Lee, J. Gero, Building envelope designusing design prototypes, ASHRAE Transactions, vol. 96, Pt.2, 1990, pp. 508–520.