Generative and Evolutionary Design in

Design Education

 

Dipl.Des. Markus Schein (PhD cand.)

Teaching Associate, University of Kassel, Department of Product Design,

Generative Design Lab, Germany

e-mail:digitalpool@uni-kassel.de

 

Dipl.Des. Ole Werner

Teaching Associate, University of Kassel, Department of Product Design,

Generative Design Lab, Germany

e-mail:ole_werner@web.de

 

 

Abstract

 

This paper discusses the experience made when starting to establish “Generative and Evolutionary Design” as a new teaching subject in the curriculum of our design department. It discusses different approaches, successes and shortcomings in teaching this subject by presenting some exemplarily projects of last years courses.

 

1. Why to teach Generative and Evolutionary Design

 

The generative Design lab in Kassel was set up in  2001 as part of the Asia-Link program. Subsequently we started to integrate Generative and Evolutionary Design as a new teaching subject in course scheme of our design department. Even though this field wasn't completely new in our department – at least not since Prof. Dr. Hans Dehlinger started to teach in Kassel  (1) –  we had to gain new experiences to establish it as a teaching subject with beneficial effects for the design education as a whole. First of all why should we teach Generative and Evolutionary Design?

 

Looking at the divers field of Generative / Evolutionary Design and the wide variety of theoretical and methodical contributions we had to find some common ground on which to base. We define Generative / Evolutionary Design in a more open way, as rule based design (2) which is as well common in ‘normal’ design processes. This allows to meet students at their actual level of knowledge and to bridge between "normal" Design and Generative / Evolutionary Design. To discover and to formulate rule systems for design problems trains design thinking and acting because it focuses on the generating aspects of design practice. Especially in early stages of design projects one tends to make decisions that have far-reaching consequences and may endanger design thinking. These decisions are often unconsciously based on personal views , historico-cultural or ideological background and narrow down the view on the problem without reflected reasons. Additionally not only designers but human beings in general also tend to conceive unknown and future situations on the base of already known situations. They extrapolate something known instead of inventing something new (3).

 

Teaching Generative / Evolutionary Design offers a good opportunity to stimulate the conscious debate on design processes and design decisions. Students gain new insights and become aware of certain particularities that are typical for so-called wicked problems (4). It's very difficult to describe design processes or to locate oneself within the process while one is still in progress.  Of course one is able to justify decisions but just after the fact. To become more aware of the process one needs to bring every single step to external, every single decision even though the step seems to be unimportant for the moment.This is exactly the point where the difficulties in using generative procedures occur for most of the students. It is not the result they have to focus on first but the description of the process. And this description needs to be comprehensible not only for themself but for others as well.

 

„In other words, we must learn to design with our hands tied. To design without the hand, using logic alone. The task is to „bring into the open“ the (subconscious) thought processes at work in the brain, and make them available to (verifiable by) anyone.“

Makoto Sei Watanabe, 2002 (5)

 

By setting up these description or rule systems the benefit for the students gets obvious: they better understand the basic methodical difficulties and properties of a design process and start to develop their analytical competence in addition to their aesthetic one.

 

Despite these aspects there are many more that recommend teaching Generative and Evolutionary Design, the keywords are: continuous digital workflow, paradigm shift in modern science, increasing interest in dynamic systems, parallel development, changed human perspective due to global interconnection, bottom-up design strategies ...

 

One aspect, we also would like to point out in this context, is the development of mass customized products (6). The enhancement of computer aided communication and production technologies as well as economic changes during the last ten years led the way to this new type of products. Within certain boundaries the design of these products (material and immaterial) can be influenced by the customers suggestions and needs. In an ideal case it gets tailored on the base of individual measurements and aesthetic preferences. Mass customized products demand design concepts, that imply a range of different instances and not one static design solution. Generative design methods are very useful to fulfill this task because they always refer to solution spaces.

 

2. How to teach Generative and Evolutionary Design

 

Upon discussing why we should teach Generative and Evolutionary Design it is time to ask how to teach and how to integrate the new teaching subject in the course scheme of our design department. As we mentioned before the field of Generative / Evolutionary Design is very wide. It includes too many approaches, ideas, concepts, methods and tools to be taught just in one seminar and would cause confusion rather then understanding.

 

The experiences form the first seminar “Generative Prozesse und Verfahren” in 2001 made clear, that the expectation to bridge the gap between understanding a new design concept and applying it successfully to a design subject was somehow too optimistic (7). This first seminar was discussing various examples of different methods in different fields of application (design, art, architecture), accompanied by some subjects of broader interest like e.g. methodical problems of Generative and Evolutionary Design. To complete the seminar students should apply gained knowledge experimentally to a design subject. Although not required, they have been encouraged to implement their approaches as Ideas-macro-scripts.

Results and student feedback have shown that this was just too much to deal with within only one seminar. It turned out that it was especially necessary to draw a clear borderline between the Principles of Generative / Evolutionary Design and its implementation on computers and applying Generative / Evolutionary Design in real design projects.

 

Consequently we split the wide field into three different groups (8):

 

1. concept and principles of Generative / Evolutionary Design

2. implementation on computers (for example scripting courses)

3. applied projects

 

2.1 Teaching concepts and principles of Generative / Evolutionary Design

Seminars of this first group can be seen as a introduction into the field of Generative / Evolutionary Design. It imparts the basic principles and methods with the help of various examples. The students gain an overview and learn more about frequently used methods like Cellular Automata, Genetic Algorithms, L-Systems, Neuronal Networks and others. To link knowledge and practical  experiences the participants of the seminar have to apply their knowledge to different small exercises like the Christmas Tree Generator (9): 

 

The task of this exercise was to design a rule which will result in the generation of a Christmas tree when being applied. Students have been free to decide what method they want to use, how (and how strictly or vaguely) it should be interpreted in the generative process. The goal of this exercise was to give the students the opportunity to make a first experience with rule based design. The results have been surprisingly playful and versatile as you can see below [Fig. 1].

Fig. 1: Christmas Tree Generator (Bruno Winter)

 

Even though it's likely to implement generative rules on a computer, it's not necessary at all. The conception of rule systems is much more important. To invent rules means to grasp somebodys intention. The implementation may adopt differnt ways. It's often already sufficient to generate some instances by hand and  to explain the system using illustrated flow charts.

 

To develop generative ideas, to analyse and understand existing rule systems and to perceive Generative / Evolutionary Design as an enrichment of ones design education have been the primary goals of this seminar. Additionally the students should be encouraged to use their new knowledge adequately in their future design projects.

To teach the unit described above, it does not always need a seminar on its own. The content - not in it's total depth but partly - may also be integrated in other courses like Prof. Dr. Hans Dehlinger did during last years course on "theories and methods of planning".

2.2 Teaching the implementation on computers / Scripting courses

Although it's not implicitly necessary and although it's very time consuming, there are advantages of teaching the implementation of generative concepts on computers. It extraordinary helps (sometimes forces) the students to train their analytical and logical competence. It can be compared with learning a foreign language. It includes new vocabularies, orthography, syntax and grammar. But the implementation on computers demands the most strict notation and does not excuse the smallest inaccurateness. Due to this the descriptive rule system needs to be worked out much more carefully and detailed then students are used to. When they start to implement on a computer they realize shortcomings in their rule systems they never thought of, even though they were convinced to have thought of everything before. They would never have noticed the potenential gaps without giving it a try on a computer.

 

Another important point is something we call, aha experience. The workflow of generative design projects differ from what students are used to: it needs quite a lot of time until first results can be seen. A lot of thinking needs to be done to set up the right rules  and constraints and it's drawn-out work to implement these rules. These project stages are hard to overcome and afford a lot of patience. But the minute the program runs, a bulk of models can be produced.

 

In "normal" design processes students percive the progress as slow but continious

- apart from some up and downs - while in generative design processes they do not really see any progress for a long period of time until they are finally confronted with this bulk of models. This aha experience it strongly related to the power / speed of computers processing the rule system. It's not possible to achieve the same strong impression and the same amount of generated models by performing the rule  system manually.

 

The CAD-education in the Department of Product design / University of Kassel is among others based on the software platform Rhinocerous. Teaching is done on two different levels: basic / advanced courses and upper grade seminars.

 

To support students exploring Generative / Evolutionary Design a new course, exclusively focusing on scripting the used CAD - software was set up since 2003. The application programming interface of Rhinocerous enables the user to write his own programs using Visual Basic Script or C++.

 

During this seminar students are able to gain basic knowledge about scripting a CAD construction using Visual Basic Script and are asked to apply this knowledge. According to a given task they have to write a small program generating for example recursive line structures or pattern generators [Fig. 2+3].

 

 

Fig. 2+3: Scripting results: recursive line structures and pattern (printed on coffee cups).

 

Teaching goes along a web based tutorial that was developed especially for this course using a lot of examples scripts gradually introducing the world of programming. But to avoid frustration and increase the efficiency of teaching, students are permanently supported by one or two tutors.

This tutorial is considered as a tutorial in progress. It continuously changes  depending on experiences made during the courses.

 

Besides the benefits for Generative / Evolutionary Design projects the students gain much better understanding of CAD -tools and the mode of operation behind the graphical interface and profit from this for all their projects.

 

2.3 Applied projects

Generative / Evolutionary Design projects may have a strong tendency to become a goal in itself.

Not only the resulting digital models may loose their relation to the physical world; the design of a design "machine" may even become the main purpose of the design process, completely neglecting the former goal.To avoid this trap we introduced a third type of teaching, the applied projects, These projects are following a tight outline and have to end up in physical models or prototypes. They intend to close the appearing gap between working on the abstract level of code but designing for physical reality. The following section briefly presents two of these projects that have been carried out this year.

 

3. Examples

 

3.1 Shelfgenerator

The shelfgenerator project was initiated during the winter term 2003/04. The students were asked to develop a generative design concept for book shelves out of sheet material using different principles and starting points and to detail it up to prototypes. Nine students finally came up with 5 different concepts, mostly parametric systems with some random influence.

 

The shelfgenerator "DimIt" is based on a flexible grid system. The user can change the settings of the grid as well as parameters like the total size of the shelf, the upper and lower boundaries of the compartment size or the type of the compartments. As a function of these settings and random input the program starts to allocate the compartments. The aim of this generator was not the design of one special shelf but to solve problems of space layout and division that can be applied to different aesthetic appearances and and subjective views (10) of a shelf. The picture below [Fig. 4] showes one possible the shelf design and construction. It remains flexible because its standardized elements can be assembled in more than just one way. So you may run the program over and over again and look if it's possible to build a new instance from your elements.

Fig. 4: Shelfgenerator "DimIt", prototyp and CAD drawings (Emmy Galle, Frank Bayer)

 

Not random but music uses the following example. This Shelfgenerator transfers midi-files into a structure of differently sized boxes lined up in two rows on a wall [Fig. 5]. The user chooses a piece of music and then selects two of several voices - one for each row of boxes. Subsequently the program transfers length of a note, tone pitch and volume into a three dimensional shelf structure. While the tone pitch and the length of a note affect the size, the volume determines the color of the box.  In the end the user can select an arbitrary sequence of desired length from the generated shelf. Similar to the first example, the problem to be solved is not only the design of a shelf but proportioning in general. The results may be transferred to other design tasks like glasses or vases as well.

Fig. 5: Generated shelf structure form the song "spanish caravan" (Sibylle Manß, Ben Kossmann)

 

2.3  Sweet Greens

Sweet Greens is a small lawn landscape partially covered from roofs made out of fabric membranes supported by pneumatic structures. It was realized as a permanent installation on an area of about 600 square meters next to our department of product design in Kassel [Fig. 6].

Fig. 6: Sweet Greens, computergenerated lawn landscape, 2004

 

The task of the project was not to design a lawn landscape but to develop a program for the design of the lawn landscape (11). Within a two-week-workshop the students should develop a generative idea of a possible lawn landscape from some sort of vague description towards an intended program implementation. The finally implemented program is based on parametric design with some simple survive-or-die mechanisms and a certain random influence.

 

To run the Sweet Greens Generator the user defines (mostly geometrical) constraints  and selects the desired plot. Most of the constraints can be limited with an upper and a lower bound like the green radius range, the slope angle etc, others require only a minimum or a maximum value.

 

Next the program loads and represents the chosen ground plot, that consists of an area outline and restricted areas [Fig. 7]. No part of the landscape is allowed to interfere these areas. Immediately the generator places circles on the given plot and validates them by a series of checks. This validation decides whether the circle remains or gets removed. This procedure lasts until the desired amount of circles or placement attempts is reached. Based on the position of the generated circles and limited by the chosen settings, the program creates the landscape using the loft operation. Now the program moves on to process the roof structure on top. When finished, the generator draws a simplified, shaded version of the landscape with it's roof structure and calculates the required amount of soil, an important criteria because the students had to build the final landscape by themself.

Fig. 7: Sweet Greens Generator, round plot and different stages of generation.

 

Looking at possible results the program is able to generate something between non-usable fantasy landscapes and very boring structures. But this mainly depends on the chosen settings and the ground plot. To use the generator one has to experiment with the bounds to develop settings which are tight enough to avoid the generation of nonsense and produce a variety of results at the same time.

 

4. Outlook and Final Remarks

 

The main goal of projects is to experiment with generative procedures in a real design occasion. The particular interests have been to experiment with programming as a method for structuring and supporting design processes and to search for a generative solution which can be applied in different environments and eventually transferred to other design tasks.

 

It has shown that teaching Generative and Evolutionary Design is a highly valuable subject for design education. After taking the first barriers, they start to clarify for themselves how they perform design, they better understand the basic methodical difficulties and properties of a design process and start to develop new competences.

 

To illustrate this, there is one encouraging example to be pointed out: During a design project - that wasn't intended to be a generative or evolutionary at all - one student picked up the generative idea, he learned about in a different seminar one or two years before. His design of a lampshade got stuck and he couldn't get any further until he gradually defined parameters and rules for his design. The he implemented them as Rhino script and came up with an astonishing amount of new solutions. These solutions not only showed much more variety but were of higher interest as well.

Fig. 8: Lampshade Generator (Ulf Cadenbach).

 

To give a brief outlook the next applied project is already on its way: Computer generated in:ex:terior structures. This time it's not soil and grass, the students will have to deal with, but concrete. In addition to form generation we will try to integrate evolutionary structural optimization in this project. For those who are interested in this project, see (12). 

 

References

 

(1)Dehlinger, Hans (Hrsg.): Linienspiele: Mit dem Computer erzeugte Strichstrukturen. Gesamshochschule Kassel Fachbereich Produktdesign. 1984

(2)Galanter, P.: What is Generative Art? Complexity Theory as a Context for Art Theory.

In: Soddu, C: Proceedings of 6th international conference Generative Art. Generative Design Lab. Polytecnico di Milano. Mailand. 2003. p. 219.

(3)Dörner, Dietrich: Die Logik des Mißlingens - Strategisches denken in komlexen

    Situationen. Rowohlt Verlag GmbH, Hamburg.1989. p. 190

(4)Rittel, H. W. J.: Planen Entwerfen Design. Kohlhammer Verlag. Stuttgart. 1992. p. 77 and following.

(5)Makoto Sei Watanabe: Induction Design. A Method for Evolutionary Design.

Birkhäuser Verlag. Basel. 2002. p.7.

(6) Piller, Frank T.: Kundenindividuelle Massenproduktion. Carl Hanser Verlag München Wien. 1998.

(7)Schein Markus: Next Generations Computer Aided Design - Next Generation CAD-Education? In: iF International Forum Design GmbH: ICSID and Education Conference Critical Motivations and New Dimensions. Hannover. 2003

(8)same as (7)

(9)www.generativedesign.de (last visit: 11_04)

(10)Soddu, C: Proceedings of 5th international conference Generative Art. Generative Design Lab. Polytecnico di Milano. Mailand. 2002.

(11)Markus Schein, Gregor Zimmermann, Ole Werner (Hrsg.): Sweet Greens - A

comuter generated landscape. Books on Demand, Norderstedt. 2004

(12)http://www.uni-kassel.de/fb12/wwtwl/ (last visit: 11_04)