1.Introduction

Which visual representation can efficiently carry the architect’s reasoning during its first conceptual moments? What types of tools are available to him to support him in his task? What link exists between the interface of theses tools and the internal representation of the architectural object in conception?

When looking at the software currently available to professionals, we must note that it follows the formula summarizing Histoire d’un dessinateur "to draw is to see, and to see is to know"(Viollet-le-Duc). Indeed all existing architectural design environments are based on the visual aspect of their interface to support the designer’s reasoning. Like the paper it replaces, the screen of each CAD tool proposes a new support to collect the graphic traces expressing the results of the designer’s reflexion. In the same way, the three-dimensional pattern makers substitute the balsa models with a virtual representation, showing a constructive model of the future building floating in computer space.

While both types of representation increase tenfold their author’s productivity by raising his speed, precision and adaptation capacity, they can also support the final steps of his design by optimizing various common performances concerning the real building: construction costs, energy needs, acoustic quality, stability or lighting features, for example, which are estimated thanks to a number of stimulators based on the data issuing from the architectural models.

Nevertheless, these representations are only composed visually by the architect on the computer. In light of the second proposition, "to see is to know," it is tempting to believe that the digital system gains the architect’s knowledge by "seeing" the representation he has given it. This representational space sends back an image of the imagined architectural space to be built. In that respect, it is seductive and reassuring for the professional and for his client but in no way does it constitute a design support. Here are two obvious reasons.

 

1. These representations do not cover architectural semantics. How many pattern makers actually address the notion of space? Most of them only take into consideration the constructive representation of architecture, i.e. the material features which physically and visually compose the building (walls, tiles, windows, etc). However, architecture uses other more important notions for the designer, especially in the earlier steps of design functional space, the border and the topological network addressing the concepts of the adjacency, accessibility and level of privacy.
2. There is no "conscience" of the depicted scene. The constructive objects composing it are unaware of the ensemble they are participating in. They are present in themselves, fixed at the moment of their creation on shared geometric points. They do not address any notion of their role, their continuity, nor of their relationship to the other components of the building.

In order to enter a real design space, we must acceed to a semantic and functional model. This model requires a complete detailed description of its space components, of their state, their role and relationships.

This description is always tedious and not very compatible with the necessarily nimble and pliable thought process of the designer in the complexities of his creative reasoning. (Leclercq, 1994a; Azar, 1994). A design model must then be able to interpret the work data used by the architect in order to allow him to move comfortably in the field of implicit knowledge (Discry, 1991).

Our thesis (Leclercq, 1994a) proposed to use the architectural sketch as a priviledged means of communication between the architect and his virtual model. Indeed, the graphic sketch contains, according to Do and Gross (Do and Gross, 1996), the essential part of composition:

 

• as the first trace of the mind’s emerging ideas, it constitutes the best reflection of the creative intention;
• it works at a projection and topology level: it expresses the relationships of its components while giving them a level of pliability far superior to a geometric model;
• throughout its progression, the sketch also expresses the operating models of the thinking mind.

We have rated and demonstrated the potentials of the sketch as a direct means of gaining semantic information for an architectural project (Leclercq, 1997).

The main result was: the sketch can efficiently carry spatial information. Although it is made up of a limited layout - the borders - it is the seat of implicit deductions since it favors, in the construction, the emergence of functional spaces. These spaces are rarely drawn as such but in our studies of architects' sketches, they still represent a quarter of their content.

The chronological character of free-hand drawing following the line of description, will in fact be the priviledged carrier for our interpretation. Indeed, the precise knowledge of the context in which each line is put down on the paper indicates the way of reading it and allows us to understand the second degree of the designer’s intent.

2. Objective

Thus our objective is to propose a tool whose interface is compatible with the designer’s working technique. It must be able to assist him during the designing phase and not during a long evaluation gap which interrupts the development of his ideas.

Concretely, we wish to recreate the existing conditions during the meeting of two architects associates, when, in front of a drink, one explains one of his recent projects to the other. His means are very limited, even rudimentary: he uses a simple pen and a beer coaster; his sketch is very vague and incomplete. Still, the other architect captures the project , understands its options, its functioning and various particularities. He could even, from this simple sketch, take over the design and carry it to its production stage.

This is made possible by three discoursing knowledgeable professionals they share the same knowledge of architecture, they use similar graphic representations and they follow, step by step, or line by line, a common reasoning during the chronological development of the sketch.

Those three conditions are for us the basis of any agent meant to aid the designing process. (Leclercq, 1994a; Leclercq, 1996b). We have attempted to apply them to our EsQUIsE prototype, whose concept with an example of application is presented in this article.

3. The procedural technique

The EsQUIsE prototype is a geometric interpretor of descriptive architectural sketches. Since it only disposes of drawing procedures and lacks the manipulating procedures of the drawn elements, our prototype is currently limiting its ambition to reading and interpreting the descriptive sketches (not yet conceptual).

It was developed in symbolic language, here the Common Lisp, and functions with a digital tablet Wacom on a Macintosh or an AIX station.

It is composed of modules distributed on two levels of temporal intervention. First the processing module for the graphic signal which operates in real time on the synthesis of the drawn line. Then several procedures are currently launched one after another, waiting for an implementation in parallel processes, and which compose and deduce the topological relationships of the described architectural project.

Figure 1. The procedural diagram.

Other reading procedures of captions and of semantic attribution finally complete and improve the pertinence of the deducing process operated on the composition of spaces. The architectural representation thus designed can be made available to the classical performance evaluations or can access a case base of architectural designs.

Here is a description of each module in its order of intervention.

4. The Acquisition Module

The first module achieves the capture and synthesis of the lines drawn on the digital tablet. These lines can have two colors, used to indicate the degree of transparency of the partitions: opaque or glass (a third color is available for non-significant features of the sketch, such as hatching or a shrub, for example).

Operating at an acquisition frequency of at least 100 points per second, this module sends to the main program an uninterrupted flow of coordinates which must be reduced in order to keep the execution of the process in real time.

A synthesis process, constituted of successive filters, "distils" the crucial data: the difficulty of this operation consists of reducing as much as possible the weight in memory of the representation of each line and, at the same time, of keeping a graphic synthesis close enough to the initial drawing. Aside from the fact that these objectives are obviously in contradiction, they both have recourse to time-consuming algorithms while we are trying to achieve the operation in real time.

In its present state, our system is able to send back the reliable synthesis of a line in less than 1/10th of a second - that is the time to move one’s pencil to start the next line - while keeping only, on the average, 15% of the flow of coordinates transmitted by the tablet.

5. The Module for Extracting Captions

After achieving the synthesis of the lines composing the sketch, but before composing the representation of the spaces they create, it is useful to detect and interpret the captions written in the drawing: these texts are so many necessary indications to establish the architectural model. How could the model deduce the sense of the functioning determined by the author of the sketch without first knowing at least the prescribed use of each room?

The technique we have proposed here is based on the facilities furnished by the module processing the lines mentioned earlier. It gives access to the reading of characters - in a simplified version, which is enough for us - and, this way, to recognizing the words, in terms of a specific dictionary edited for that purpose.

Figure 2. Original sketch, drawn on digital tablet,
the reading of captions and the synthesized sketch.

Thanks to this limited dictionary, the identification of the captions turns out to be very efficient because stable and performing, despite a relative rate of success in the reading of characters (from 60 to 70%).

Through the decoded caption, the program is able to name the functional spaces, to fix the characteristics normally associated with them and to use these in the feeding of the evaluations it must serve.

6. The Determination of the Contours, the Composition of Occupied

Spaces and Spatial Parametrization

From the study of the contacts between the synthesized lines, EsQUIsE can deduce the spaces delimited by them. These procedures of implementation are fairly classical in the field of infographics where they are close to the techniques for determining the contours in boolean operations applied on co-planar polygons. The difficulty in our case concerns the imprecise nature of architectural sketches and, especially, the conditions of defining the lines’ extremities which are imperfect since always interrupted too soon or too late!

By going past this imprecision - not by correcting it because the creative aspect of a sketch depends precisely on its unfinished character (Gross and Do, 1996) - EsQUIsE materializes the second level of architectural perception , i.e. the spaces to be occupied.

Figure 3. List and topology of the spaces and partitions interpreted, 
completed with their deduced characteristics.

These are rarely specifically drawn by the architect (in the shape of blobs, for example) but they are rather described by the particular organization of their respective or shared limits. Yet, in the designer’s mind, the architectural sketch first represents a set of functional spaces. Thus, EsQUIsE attempts to perceive their underlying organization and can supply their adjacent relationship thanks to a deduced architectural model.

As complement, it provides all the characteristics of the borders which separate them: dimensions and orientation of each wall are completed with choices of technological implementation, implicitly decided thanks to the conditions separated by each partition. For example, between a bedroom and its adjacent sanitary space, EsQUIsE will choose a wall sufficiently thick for acoustic comfort and a water-proof wall covering.

6.1. The Semantic architectural model

As illustrated in figure 4, EsQUIsE takes part, during the realisation of the sketch, in the relations interwoven between functional spaces, boundaries and detailed boundaries.

Starting from the operations drawn, which are the "capturable" lines expressing boundaries, it elaborates both the identification of these functional spaces and the establishment of the relations between these spatial entities and their dividing elements. These topological relations are suitably adapted to support the semantic information. They enable some sub-models of work to be isolated, thereby making it easier to monitor and maintain coherence:

 

• the structure of the functional spaces permits an abstract semantic representation of the architecture, which supports the general controls of functionality (for example, accessibility, trajectories and privacy limits);
• the boundaries level is the place where the flow control, which is part of the atmosphere definition occurs (thermics, acoustics, ... );
• the detailed boundaries represent the first step towards the product model, established here in the more consistant sense, from the semantic aspect to the constructive characteristics.

This structure preserves the relational model for a longer time. It facilitates the rise towards abstraction, thereby easing the constraints of a problem, as previously shown in the diagram of the design process. The geometric model, built upon the detailed boundaries level, is also verified at many different degrees of representation. This gives it enhanced consistency.

Figure 4. The emergence of the functional spaces.

The generalisation and specialisation movements thus authorised make it possible to carry out design work linked to the analogue approach. A model which uses this multistage representation meets the conditions of creativity: it provides the means with which to work on the analogical and adaptational approach.

7. EsQUIsE Application: Evaluation of a project’s energy needs

Starting from the characteristics of delimited spaces, EsQUIsE interprets and then translates the semantic representation of the project drawn on the tablet, into various parameters destined to the evaluations.

Figure 5. The project is sketched and receives, a few seconds later, the estimation of its energy needs.

For example, the margin of thermal conductivity of each partition is directly provided (in W/m2K) taking into account the spaces to separate. In the same way, the prescribed temperature for each occupied space is fixed according to its function as described in the captions: 18°C for the kitchen, 22°C for the bathroom, etc.

The application now tested in the late stages of EsQUIsE is a classical module of multi-zone evaluation of the building’s energy needs, called MZS: from each window’s surface and orientation, it estimates the sun supply for each room and calculates for the whole building (viewed as a network of spaces) the balancing of its heating or air conditioning needs.

If these energy appraising methods have been known for several decades, their application in this context demonstrates the advantage of using a man-machine interface based on the semantic analysis of an architectural sketch. We consistently use the same matrix calculator of thermal calculations but while awaiting the possible development of EsQUIsE for bigger projects, we labor each time on the fastidious measuring work of the architect’s blueprint. As opposed to the two or three days usually required to measure and encode, the energy performances are, in this case, supplied only a few seconds after the drawing of the last line of the sketch.

8. Developments and Conclusions

Currently we are studying the prototype in terms of, on the one hand, extending its capacity to deduce spaces to three-dimensional volumes and no longer 2D1/2: this requires space generating algorithms from the combining of plane and section drawings.

 Figure 6. 2D1/2 views of the sketch, created by EsQUIsE in Quicktime3D

On the other hand, our research is studying the user conditions of EsQUIsE to formulate - implicitly - requests based on cases in order to access a directly useful knowledge in computer-aided design.

Finally, EsQUIsE is felt to be a priviledged interface tool for more complex estimation applications than the thermal balancing mentioned in this article, like the qualitative evaluation of interior lighting for example.

With its ability to recognize roles and relationships between the components of an architectural sketch, EsQUIsE proposes an interface active in the process of architectural design, making it at once more coherent and more creative.

Lebahar has given a very good definition of the architectural sketch (Lebahar, 1983). We think we have completed that approach by bringing it a concrete, working formalism which opens a new field of development.

Moved from the paper to a virtual support, the sketch can keep the semantic expression of its symbolic and relational abstraction for a longer period of time and it allows the reflecting designer to better build and manage his knowledge.

Information provided by: http://www.arch.usyd.edu.au