Content uploaded by Wolfgang E. Lorenz

Author content

All content in this area was uploaded by Wolfgang E. Lorenz on Oct 01, 2017

Content may be subject to copyright.

Structuring of Teaching and Learning Situations in

Architectural Education

Using and Integrating Digital Analysis within Interactive Genetic

Algorithms

Matthias Kulcke1, Wolfgang E. Lorenz2, Gabriel Wurzer3

1Hamburg University of Technology/HafenCity University Hamburg 2,3TU Wien

1matthias.kulcke@tuhh.de 2,3{lorenz|wurzer}@iemar.tuwien.ac.at

BACKGROUND

In the search for a deeper understanding of aesthetic

quality in architecture, complexity of a design pro-

vides a possible area of comparison. In fur ther detail,

varying degrees of complexity may serve as compa-

rable characteristics regarding diﬀerent design pro-

posals, either observed in the design as a whole or

over diﬀerent scales (fractal characteristics). Thus,

using diﬀerent methods to determine complexities

can lead to a deeper insight concerning which role

each of them plays in the overall aesthetic quality of

a given design. One well established method to de-

scribe complexity represented as density and distri-

bution of signiﬁcant lines of a design across diﬀer-

ent scales is the fractal analysis method box-counting

(Lorenz 2012). The authors propose an additional

method, called gradient analysis, as part of a greater

palette to aid design aspect-analysis.

Gradient analysis

Gradient analysis is an additional method for archi-

tects, designers as well as students and scholars to

analyze and subsequently inﬂuence the degree of

proportion-complexity of a design (Kulcke et al 2015,

2016). This method is based on the assumption that

redundancy reduces the complexity of an objects’

appearance. However, the reduction of complexity is

not a quality per se; the matter at hand is for design-

ers to achieve a balance between redundancy, which

provides readability of form, and complexity to sus-

tain interest in it. Moreover, the authors are aware

that, concerning this balance, many diﬀerent levels

of comparison exist, including, among others, the va-

riety of material and colour combinations. However,

for the moment the authors decided to focus on pro-

portion, expressed by a gradient which is deﬁned by

two points (diagonal of a rectangle).

The basic idea of redundancy-analysis by

gradient-comparisons led to a computer program

called ‘gradient analysis’, which pairs every signiﬁ-

cant point of a design with each of the other points,

comparing the gradients of the lines deﬁned by each

pair. By doing so emphasis is put on similar relations

rather than speciﬁc relations (e.g. Wagner, 1981).

Moreover, relations are not weighted depending on

distance and/or visual highlighting e.g. in form of

material edges, since Gestalt perception does not

necessarily require edges (Kulcke et al, 2015). The

weighting of diﬀerent perceptual relations will be

subject of future studies.

The result of the measurement is a list of pairs

of points (coordinates of corners or intersections of

lines) and their gradients, given as angles. In addi-

tion to the angle redundancy quotient a length re-

dundancy quotient is also determined. The angle-

redundancy quotient Rαis deﬁned by the number

of diﬀerent angles Crdivided by the total number of

angles C(formula 1; Kulcke et al, 2015):

Rα=

Cr

C(1)

Draft - eCAADe 35 |1

Figure 1

Scan and retrieving

rectangle

measurements

with rbeing all repetitions and Cbeing calcu-

lated as follows (formula 2; Kulcke et al 2015):

C=(n∙n−1

2)(2)

with nbeing all single connections. The length re-

dundancy quotient is calculated in an analogue fash-

ion. In both cases, a tolerance coeﬃcient takes into

account that slightly diﬀering angles or lengths may

be perceived as similar and that the drawing of the

object may not be accurate. The tolerance coeﬃcient

includes all angles inside a certain range, i.e. angles

that stay within that range are counted as a repetition

of the angle they are compared to.

Interactive genetic algorithm

In order to create a responsive system to be used

within a design- and/or learning process the gradi-

ent analysis has been integrated in an interactive ge-

netic algorithm (IGA). Within the IGA it serves to de-

termine the ﬁtness value of a given object evaluat-

ing the degree of proportion complexity and to cre-

ate variations with an optimized degree of propor-

tion complexity. In principle, the algorithm consists

of the following steps:

1. Generating the ﬁrst population of n�parents:

Currently two diﬀerent designs each of which

is deﬁned by signiﬁcant points.

2. Encoding: The coordinates of all signiﬁcant

points are translated into binary code.

3. Crossover: Two encoded parents are cut at a

deﬁnite point and their parts are exchanged.

4. Mutation: Digits are swapped at a certain po-

sition.

5. Fitness�values: Angle redundancy quotient

and length redundancy quotient are calcu-

lated to determine the segment�size on Gold-

bergs weighted roulette wheel (step 7).

6. User-choice: The user can favour a result for

Goldbergs weighted roulette wheel.

7. Generating the next population of n�parents:

The next generation depends on the result

of Goldbergs weighted roulette wheel (Coates

2010, König 2010) and another user-choice.

Some of the steps above are determined by users’

choice and/or including randomization. The process

is repeated until a satisfying solution is reached.

IGA WITH GRADIENT ANALYSIS IN THE

WORKSHOP

In the workshop participants extensively test the in-

teractive genetic algorithm and thus also the inte-

grated gradient analysis in a given framework. Since

the software is undergoing a process of permanent

evolution, particular attention is paid to the analysis

and discussion of the results. Another focus is laid on

the discussion of possible integrations into learning

environments.

The presented method of analysis to evaluate a

certain layer of design quality is used exemplary to

test the integration of design analysis via an IGA into

a design learning context. Alternative methods for

evaluating design complexity, like easy access fractal

analysis (Wurzer and Lorenz 2017), may also be inte-

grated into a similar design process. Participants are

held to adapt the process to their analytical focus or

expertise within their own teaching environment.

The process

The design process as considered in the workshop

consists of three phases:

2|eCAADe 35 - Draft

1. Freehand form development by sketchingob -

jects into a given rectangular ﬁeld plus im-

age enhancement and measurement extrac-

tion (Figure 1).

2. Evaluating the intuitively chosen proportions

by using the gradient analysis IGA (Figure 2)

and semi-automatically producing alternative

versions on the basis of proportion analysis

(Figure 3).

3. Comparing and choosing the ﬁnal version,

thus optimizing or conﬁrming the initial de-

sign gained by sketching.

Figure 2

IGA user Interface

and ﬁrst population

of n�parents (n=2;

showing just one of

the two parent),

displaying ﬁtness

values

The workshop content

As preparatory steps the participants are familiarized

with the general theoretical approach of proportion

complexity and proportion optimization. Then they

are taught to use scanning and an image enhanc-

ing process allowing to feed the intuitively reached

object into the web application to apply the IGA for

proportion optimization. The actual task is presented

as a design task to develop a prefabricated concrete

façade module with two openings, especially focus-

ing on a pleasing design considering the placement

and size of the windows or doors. The height and

width of the desired element are given, participants

are asked to choose the openings boundaries as par-

allels to the rectangular boundary of the overall ob-

ject. The participants are advised to draw a mini-

mum of six variations and encouraged to optimize

a favoured ﬁrst design by iteration i.e. by sketch-

ing it again with alteration of measurement, but pre-

serving the general idea of the previous design. Af-

ter choosing two favourite designs from the sketch-

ing phase, these are scanned (Figure 1) and opened

within the image enhancing software. The sketched

and digitalized rectangles receive an overlay of dig-

ital rectangles as a means to retrieve the measure-

ments of the openings. These are chosen by the par-

ticipants who visually control the process, especially

regarding if the overlay rectangles still represents or

even betters their design aim. The measurements of

the favoured two variants can now be entered into

the user interface of the gradient analysis IGA to serve

as parents (Figure 2), as the starting point for the in-

teractive genetic algorithm. Again in several cycles

the proportions of the openings within the facade el-

ement are optimized, chosen by the numerical gra-

dient quotient and visual appearance (Figure 3). Af-

ter several cycles the participants receive an output

of the favoured solution of each cycle out of which

again two favoured designs can be chosen. These

are then put to comparison to the top design solution

from the ﬁrst phase of sketching. Finally each partic-

ipant presents his or her work and voices the choices

he or she has made on the way toward the ﬁnal de-

sign.

Advanced Task

The integration of user choices remains an important

focus as the developer intend the tool to be utilized

as a cognitive and analytic aid during the design pro-

cess and not as an automatic design generator. For

the workshop the software has been developed fur-

ther in order to allow users to manipulate elements

of the IGA themselves while producing design varia-

tions e.g. implementing their own cross over masks

(Coates 2010, König 2010).

To gather and evaluate the feedback (of stu-

dents) a qualitative approach is in development. The

status quo is using narrative interview technique and

computer aided qualitative content analysis. It is part

Draft - eCAADe 35 |3

of the workshop to discuss the optimization of and

alternatives to this feedback-process.

Figure 3

Two examples of

ﬁrst cycle and user

choice (bottom) for

next cycle,

displaying ﬁtness

values

PREVIOUS TEST CASES

The implementation in the learning envi-

ronment

The approach has undergone ﬁrst testing with stu-

dents in the winter semester 2016/17. The students’

feedback is taken into account to optimize the pro-

cess and the interactive algorithm (Poirson et al 2010)

for further implementation in design education. A

crucial point is the carefully adjusted use of diﬀer-

ent media (sketching in combination with analyz-

ing/choosing within the GUI of the IGA). On the basis

of testing the IGA within the learning environment

changes have been made to the algorithm to opti-

mize the design process. The comparability of the

task is signiﬁcantly changed by seemingly small al-

terations, respectively given rules and constraints, re-

garding the task.

To gather and evaluate student feedback a qual-

itative approach is in development. The status quo

is using narrative interview technique and computer

aided qualitative content analysis. It is part of the

workshop to discuss the optimization of and alterna-

tives to this feedback-process.

DISCUSSION

Allowing participants to use edges that are not paral-

lel to the boundaries of the rectangle, using rounded

corners or organic freeform is of course possible. But

this should be done in a way that every participant

is aware of that possibility, to ensure comparability

for later discussions of the results. A possible varia-

tion is also to allow for teamwork and to propose to

participants to marry their own favourite design from

sketching to one chosen from a colleague, thus build-

ing the initial pair for the IGA.

REFERENCES

Burke, A 2008 ’Reframing “Intelligence” in Computa-

tional Design Environments’, First International Con-

ference on Critical Digital: What Matter(s)?, Cam-

bridge (USA), pp. 359-366

Coates, P 2010, The Programming of Architecture, Rout-

ledge, New York

Herr, CM and Karakiewicz, J 2008 ’Towards an Under-

standing of Design Tutoring: A grounded study

of presentation materials used in tutorial conver-

sations’, Proceedings of CAADRIA 2008, Chiang mai,

Thailand, pp. 321-327

Kulcke, M and Lorenz, WE 2015 ’Gradient-Analysis:

Method and Software to Compare Diﬀerent Degrees

of Complexity in the Design of Architecture and De-

signobjects’, Proceedings of eCAADe 2015, Volume 1,

Vienna, pp. 415-424

Kulcke, M and Lorenz, WE 2016 ’Utilizing Gradient Anal-

ysis within Interactive Genetic Algorithms’, Proceed-

ings of eCAADe 2016, Volume 2, Oulu, pp. 359-364

König, R 2010, Simulation und Visualisierung der Dy-

namik räumlicher Prozesse, VS Verlag für Sozialwis-

senschaften, Wiesbaden

Lorenz, WE 2012 ’Estimating the Fractal Dimension of

Architecture: Using two Measurement Methods

implemented in AutoCAD by VBA’, Proceedings of

eCAADe 2012, Prague, pp. 505-513

Poirson, E, Petiot, J, Aliouat, E, Boivin, L and Blumenthal,

D 2010 ’Study of the convergence of Interactive Ge-

netic Algorithm in iterative user’s tests: application

to car dashboard design’, Proceedings of IDMME - Vir-

tual Concept 2010, Bordeaux

Wagner, FC 1981, Grundlagen der Gestaltung – plastis-

che und räumliche Gestaltungsmittel, Kohlhammer,

Stuttgart

Wurzer, G and Lorenz, WE 2017 ’Cell Phone Application

to Measure Box Counting Dimension’, Proceedings of

CAADRIA 2017, Suzhou, China, pp. 725-734

4|eCAADe 35 - Draft