Generative Design – Programmed Identity

Ulf C. Cadenbach, Henrik Hornung, Tobias O. Schelkes

Prof. Hans Dehlinger

School of Art, University of Kassel, Germany

e-mail: ulfcadenbach@web.de, henrikhornung@web.de, tobischelkes@web.de

Abstract

The following article sums up three different approaches, which have been developed by the context of the department of product design at the School of Art of Kassel. It is about 1. reflections of the “generative approach” in design from a student’s point of view, 2. an experiment on the identity of a fan-like object and 3. a concept for the generative creation of a lamp shade made of laser cut sheet metal.

All three texts deal with the methods, the results and the created identity of generative design. The contribution of Henrik Hornung is rather reflecting, whereas the one of Tobias O. Schelkes is experimental and the one of Ulf C. Cadenbach deals directly with the design of a lamp shade.

All three contributions cover fields (theory, experiment, design), which are significant teaching methods of the department of product design.

1. Reflections about computer generated Design

1.1 Designing and Programming

The processes of designing and programming are quite similar concerning their structure. Both deal with solving problems. First of all, the existing problem is to be analysed. If the circumstances are known, the requirements for the final product need to be defined. Now the process of designing or problem solving begins. This requires creativity, logical thinking, imagination and experience. Connected with the question of the realization one of the design objects is now implemented.

A programming language has to be learned similar to that of a foreign language, including vocabulary, orthography, syntax and grammar. But there is one important difference: The programming language does not accept a colloquial language. It has to be used correctly in order to be understood. In contrast to humans the computer is not fault-tolerant.

A programme is not built via mouse click. The instructions must be written down explicitly. But those proceedings do not have much in common with programming yet. Although the programme runs step by step a huge number of instructions can be written directly into one file and implemented from there in order to visualise the result.

At the beginning of a programme there is an idea. It relies on a thorough analysis of what seems to be possible. And it contains the thought that the result can be reached through a process.

By structuring work, e.g. organisation of procedures, the idea becomes a concept. Gradually a programme is developed, which specifies generative rules for the computational as well as the visual implementation within the rules of the used programming language. This is actually programming. And it has much in common with designing.

What we colloquially call „programming“ is defined much more precisely by computer scientists. For them the translation of a concept into algorithms is the most important thing.

Computer generated design can be understood as parametric design. Apart from many boundary conditions a certain quantity of parameters remain open. By manipulating them one can control the total behaviour of the final design.

1.2 Calculating and Designing

A huge asset of the computer is the generation of quantity. And this is the theoretical condition for the generation of variety. Within a given and a conceptually described solution space it is observed in experiments that computer programmes generate very surprising solutions.

What is the reason for that? The computer is ignorant. It can inevitably only do what it is dictated. Humans however are knowing. This knowledge affects their thinking and acting permanently in the background. From their experience humans build connections, draw parallels, associate incessantly without the possibility of taking distance. They possibly end up with different solutions than the computer. But those are mostly located outside the defined solution space.

The computer works with high speed. Where humans give up on fatigued, the computer calculates further. Where humans stop or change the process in another direction because they estimate the probable success as little, the computer calculates further. This weighs particularly if a random function is integrated into a programme. Systematically (or randomly) solutions are generated. Many of them are possibly useless, may it be for technical or aesthetical reasons. It can however lead to interesting surprises.

Frequently some kind of “accident” occurs while designing. Every designer might know these situations. Many of the big inventions of mankind have been the result of an accident. If one tries to program a solution for a design problem it is particularly interesting to recognise such cases. One even tries to provoke them, to cause them consciously, which is a logical paradox. How can one plan what occurs per definition only unplanned. Designing is however in any way and always imprisoned in such and similar paradoxes. A tiny programming error may cause either an error message or surprise with unexpected results. Even if such results are not directly useful, they can be inspiration for further considerations.

1.3 The Question about Identity

Döring ^[1] describes „identity“ as unmistakableness („individuality“) over the time („continuity“) in different situations („consistency“). Unmistakableness in the sense of a uniqueness is not only clearly recognisable with a computer generated design (see fig. 1.1), but always as well obviously and directly provable on the basis of numbers. Naturally solutions are generated that resemble each other strongly. But apart from the fact that this is the same with “handmade” design, exactly this is a further strength of the computer. Even if a design is already, good an appropriate programme still generates alternatives. The closer one gets towards its aim, the more one can restrict the computer’s scope, the more parameter settings can be fixed and the more precise and the more defined the design becomes.

Complex programmes are constantly tested during their developing. The designer / programmer constantly gets direct results during this process, which he must evaluate. Process and result are therefore subjected to a constant quality inspection. If necessary it can be intervened at any time. And that will always be necessary, because even if many decisions are assigned to the programme, even if design principles are defined as algorithms, it is the substantial task of the designer to select the best one from the multitude of results. This requires strength of judgement, sharp thinking and a good intuition.

2 Experiment on the Identity of a Fan-like Object

2.1 Explanation of the experiment

We describe an experiment in which we follow the question how the identity of an object can be generated, changed or exterminated by transforming a figure. We are structurally defining a kind of “wing” (see Fig. 2.1) which is divided into zones. These are restricted by a value facet. We are generating two kinds of wings by inputting these information and some random variables into a script.

(1) angular, polygonal (see Fig. 2.2)

(2) roundish (see Fig 2.3)

Now we are generating objects from these wings by rotating seven elements which are uniform around a lightly varying rotation angle (29° to °33). Being geometrically defined the axis is always the same. The similarity to a Chinese fan of the objects generated in this way is wanted. The identity of such a fan is highly been determined by the kind of elements and the way it is fan-like. There are far more characteristic items of real Chinese fans like the kind of folding, material properties and others.

Anyway we can prove experimentally that the „generator“, which should generate fan-like objects really generates fan-like objects.

Based on this initial situation we ask the following questions:

(a) What is the identity of the generated objects? Are they invariably of fan-like kind or appeared any new identities?

(b) Are there any possibilities to intensify the identity?

Do we have any possibility to generate new identities with this generator which differs from the fan-likes?

The generator has been programmed as Rhino Script. The conclusions are shown in the tables 1 and 2.

2.2 Discussing the Achievements

There is a distinct tendency to families of forms in both the polygonal and the round kind of objects. I divide the families by an instant overview. There are to mention four different groups or families in this experiment. The first group contains plane fan-like objects like result 2 and 8 of the roundish kind and 5 and 16 of the polygonal kind. The second family includes the lightly opened like result 6 and 7 of the roundish ones and 1 and 13 of the polygonal ones. The third group contains the more leavened forms (roundish: 1 and 3 / polygonal: 2 and 3). The fourth group are the ones which are opened in the upper area (roundish: 4 and 11 as polygonal: 8 and 10) and so make up the other extreme.

It is remarkable that the identity of the objects becomes more and more another with more openings but never loses its relationship to the fan-like kind.

By modifying the generator to more filigree results the fan-like identity becomes more obvious. If the modifying results in forms that are more massive in the base area and at the same time filigree in the upper area the identity increasingly departs from the fan-like impression.

2.3 Conclusion of the Experiment

The more waisted and / or symmetric the generated form is the more fan-like the objects get. Regarding the results with the number 10, 15, 14 in table 2.1 and the results with the number 2,11,18 in table 2.2 that becomes obvious. But at the same time there is to mention that there are borders where the conclusion obverts and so result is less a fan-like object. Therefore the objects with the numbers 11, 19, 23 on table 2.1 and the results with the numbers 9, 10, 19 on table 2.2. are an example which underline this.

The conclusion of this experiment is that by rotating a defined number of elements around a geometrically defined rotation axis the identity becomes more and more a different one but nevertheless still is remarkably fan-like.

3. Generation of a Lampshade from Laser cut Sheet Metal

3.1 Task

The basic idea was to design a lamp consisting of a number of sheet metal elements, which can be rotated around an axis. By using the same, but scaled shape for each element, similarities to the shell of a nautilus or pearlboat will appear.

We start with a sketched, closed curve, from which we extrude a surface. This surface is bent to a semicircular shape, copied, scaled, and rotated various times around a horizontal axis. By doing so we create a three-dimensional object.

The generative approach arised from trying to automatise iterating work processes. The shape was still sketched by hand but, to let the computer do the work of the iterating steps of copying, scaling and rotating, a small script was created for the CAD-Software Rhinoceros. Due to the fact that not every sketched shape could be processed with this one script (collisions appeared), parameters like number of elements, scale factor and rotation angle got varied and saved as stand-alone scripts. It was astonishing, which various appearances could be realised only by varying those three parameters (see fig. 3.2).

Next step was to define eight different versions of the script parameters, which represented the most characteristical appearances of the generated models (see fig. 3.1) With this “kind of generator” I tested the manually sketched shapes.

The characteristics of the models shown in fig. 3.1 could be summarised as:

- “less elements, little scale factor, large angle” versus “ many elements, large scale factor, little angle”

By using this procedure a large number of different appearances could be created, most of them were suitable for further processing. Some scripts were rejected, creating unrealisable models. For example the characteristic of “large number of elements, little scale factor (<0. 6), large rotation angle” lead to elements which were obviously too small for processing and even not visible in the model. But this helped to find out value sets for the later created generator, which almost exclusively create results suitable for processing.

In the next step a programme code was created (rhino-script), which as a generative system completely autonomous generates models of lampshades (see appendix 5.3)

This programme is fully functioning for relatively incomplex shapes. The shapes are generated by random, giving the designer or user the option to accept or decline the generated shapes for further processing (see fig. 3.4) The user also has the option to pick points by hand to create a shape and assign it to the generator for further processing (see fig. 3.3).

3.2 Operating Mode of the Generator

The generator creates eight coordinates by random. The value sets of this coordinates were chosen to avoid intersections of the generated curve. In a later version of the programme the value sets will change dynamically depending on the previous randomly created coordinates, to achieve a maximum variation of the curve.

The first and last coordinate points of the curve are located on the x-axis to get a closed shape by mirroring the curve. The following check routines make sure that the created shape is a closed curve, which is important for generating a surface and for processing the shape in a CNC lasercutter. If curve points are picked by hand, a check routine ensures that each curve point is assigned to a coordinate to avoid errors.

Now the curve is drawn, mirrored, joined and checked. A surface is now created from the emerging shape and bent semicircular.

That form is aligned parallel to the c-plane and the user is prompted to enter values for number of elements, scale factor and rotation angle. Optionally these values are generated semi-dynamically by random. From these values the CAD-Software builds the model and changes the view to perspective.

The next version will include some useful features, for example; loading of pregenerated shapes, saving of generated curves and models to files, an output of the processed values and coordinates and an option to export the shape as a .dxf-file for transferring the data directly to a CNC lasercutter (see fig. 3.6).

See fig. 3.7 for a flowchart, and appendix 5.3 for the complete program code

3.3 Conclusion

Generative design surely does not replace the classical design process. It is an interesting strategy for some concepts - like this one. In this concept a number of parameters can be varied without moving away from the basic idea.

A great advantage is to automatise iterating and tedious, but essential processes and get to significant results in a fraction of the time as compared to “handmade” models. In a given time-frame one can create and evaluate a much larger number of variations of a basic concept. Now it is possible to achieve results which in the classical design process would be declined due to lack of time.

Since the generator is deaf, dumb and blind, it will deliver concepts which the designer would never have processed nor considered due to his own preoccupation. Further on, completely new concepts arise from so called “accidents” or imprecisenesses defining parameters. These new concepts may vary from the basic idea but could be interesting by all means and lead to a completely different result.

Last but not least, it is important for students to deal with the concept of “mass customisation”, which surely show some similarities to generative / parametrical design processes.

The generative / parametrical design approach is considered a useful add-on to the classical design approaches. One should not be afraid to make use of this strategy if it seems to be useful. Consequently it is absolutely inevitable for designers to deal with this form of design approach.

4 References

[1] Döring, N.: „Sozialpsychologie des Internet. Die Bedeutung des Internet für Kommunikationsprozesse, Identitäten, soziale Beziehungen und Gruppen“, Hogrefe, Göttingen 1999, P. 255