Engineers practice simplicity. Even compared to tiny marine plankton, man-made constructions turn out to be simple. Filigree structures are impossible to draw traditionally, even more, impossible to manufacture traditionally, and cost time and money. But 3D printers would already allow us to add more complexity. In order to realize their potential, we need to overcome our lack of imagination and work with tools that allow objects to “grow” – just like in nature.
Josefine Lissner shares her insights on nature-inspired innovations in this piece published on 1E9.
Fifty-five years ago, a certain Gordon Moore introduced his observation of the transistor density on a computing chip doubling every 12 – 24 months. In doing so, he predicted the digital revolution that brought exponential growth to our world. The interesting question is whether the law was an intuition of deterministic logic or rather a prophecy whose claim to fulfillment has kept the tech industry on its toes for half a century. Personally, I am an advocate of the second explanation, because one would want to assume that mankind is still the author and not the puppet of its own history. But often it takes just such a Mr. Moore to show where the journey is headed.
The interplay between mentality and innovation potential often exhibits self-reinforcing mechanisms. Unfortunately, this also applies in the opposite sense. Where do you think we would be in terms of technology if we had never developed a sense of the scalability of computer technology and no aspirations for its implementation? How will we be future-proofed if we neglect to invest in education and technological advancement during today’s pandemic and economic crisis? In this somewhat dark context, I would like to discuss a topic that has fascinated me for some time and, in my opinion, could lend the future an exciting touch.
Because one thing is clear: without technical innovation, we will not be able to solve the global challenges we face. I am also convinced that to do so, we need to completely rethink the design and development of physical objects!
Now, I am working for Hyperganic, an international deep-tech company with offices in Munich, Singapore, and China. It’s also where I found Gordon Moore’s machine equivalent: It’s the 3D printer, which promises the potential to produce entirely new kinds of objects. The problem right now, though, is that traditional software and design tools are based on the assumption that engineers in the 21st century will still work the same way Leonardo Da Vinci once did: manually and visually. That doesn’t scale.
3D printers – complexity included!
3D printing, or additive manufacturing, has left hobbyists and prototyping behind now for a long time. Industrial 3D printing is proving to be increasingly competitive against traditional manufacturing methods in terms of production times and costs – and even outperforms everything that has gone before in one aspect: complexity is free! The production cost of a 3D-printed component depends primarily on the amount of material required. Whether I print a simple block or a highly complex rocket engine makes no difference, as long as both components have the same volume.
In classical subtractive manufacturing, you would start with an oversized block of material and remove everything that should not be part of the finished object by milling or drilling instead of building up an object layer by layer. Complex components are associated with a considerable amount of additional work. This is because only areas that can be accessed by tools from the outside can be machined. Areas below the surface are inaccessible. In addition, the material is not a variable, but a boundary condition. Everything else must be manufactured in individual parts and reassembled later.
Additive manufacturing owns an immense reservoir for physical innovation, especially in regard to complex, integral, highly optimized, and efficient designs.
The engineer as the bottleneck
Here, we are dealing with a paradigm shift in two respects: On the one hand, the framework for structures that are practically feasible to produce according to the state of the art has changed. On the other hand, this requires a fundamental change in the way we conceive and design geometry. It is naïve to assume that an engineer, even if he is an undisputed CAD virtuoso (and who is?), would still be able to realize highly complex designs in a manual, piece-by-piece fashion in the foreseeable future.
The automation of the geometry-generating process requires an algorithmic approach that can be fully computerized – that is what we call Algorithmic Engineering. However, the engineer is far from off the hook with this, because the design task itself will also change dramatically and will turn into developing algorithms and directing computers.
It frustrates me to see that this insight has not arrived at educational institutions and universities, yet. In my opinion, we need people who think through products and technical solutions once again, uninhibited by previous conventions and in a new, functional way.
Algorithmic designs for rapid diversity
In order for machines to generate objects automatically, we first need to leverage our knowledge of how to create geometry and elevate it to an algorithmic level. Then we would start to capture design logics programmatically so that they become accessible to computers. Precisely this means that an engineer will no longer deal with individual problems but with rather abstract problems, and cast design knowledge into generic algorithms, so that these can then be applied as versatilely as possible to a large number of concrete cases.
The more I deal with this, it becomes more obvious that the current working methods and industry standards are a waste of time. After all, once you have such a robust, parametric design description, a computer can generate an enormous diversity of variations in a very short time and without further human intervention. Moreover, each variation can be evaluated against a fitness function, simulation or an experimental test. After all, these findings can be fed back into the ongoing design process as feedback. Interestingly, mutation and fitness testing apply the same resources as natural evolution does.
A glance into nature
Taking a look at nature from an engineering perspective is definitely worthwhile because every organism in nature is incredibly complex and perfectly adapted to respective environmental factors. Every tree is topologically optimized with respect to the prevailing wind direction, slope, and solar radiation. Natural systems achieve their functionality with minimal material input and are also inherently resource-efficient. So what could be more favorable than to use nature as a model for technical innovation and to help us solve global problems?
But back to our engineering design problem: How do I even begin to describe something that is complex? If you had to come up with some incredibly elaborate algorithm to do so, you wouldn’t really be able to ban the human bottleneck from the process and limit the universality of such a solution approach.
Nature, however, manages to grow highly specialized structures from primitive clusters of cells while relying on a rather minimalist program code. By this, I mean that an organism’s genome, which represents its blueprint or construction logic, tends to be kept short in order to be less error-prone during reproduction.
Therefore, a natural growth process is often based on very simple rules, which only provide increasingly more complexity through their recursive application. Such simple rules can certainly be recreated in mathematical models. Envisioning to grow technical objects like in nature one day, I would like to present two simple, but potentially useful, mathematical production systems in the second part of the text. Don’t worry, there will be lots of pictures!
Lindenmayer systems
Lindenmayer or L-systems are the first production systems that were used in computer graphics to simulate fractal or plant-like structures. For example, one starts with a single plant stem element and then applies a simple rule with every iteration step, replacing each existing geometry element with more complicated segments. So, if I want to create a “plant,” my rule might be: “Replace each stem element from the previous step with two new elements at a time, enclosing an angle of 20 degrees to each other and having only half the stem length.”
What sounds simple and comprehensible leads to a doubling of the branching number with each repetition, and thus to exponential growth. My blade of grass from the beginning would have turned into a broad, finely branched bush after a few iterations. Such L-systems are wonderful to experiment with*. Small deviations in the definition of the growth parameters can lead to extremely diverse phenotypes. Results can look like this, for example:
Despite the striking similarity of these computer-generated structures to real plants – especially trees and grasses – the comparison to natural development processes is misleading. L-Systems execute the internal code – the “genome” – according to a strict set of rules without being subject to certain randomness or external factors.
In biology, however, it is considered proven that epigenetic factors of the environment, such as solar radiation or food supply, exert an equally important influence on growth. Moreover, with the exception of sheer surface maximization, such exponentially growing structures rarely prove to be technically relevant. It would be much better if this growth could be controlled in a targeted manner – in much the same way as a tree always tries to grow towards light and gravity.
Space Colonization
A mathematical model that illustrates this aspect very well is the biologically inspired space colonization algorithm. This was originally used to simulate the enforcement of foliage leaves with leaf veins and their branching. The principle is based on discrete concentrations of the plant hormone Auxin distributed evenly or randomly over the leaf area. The hormones act as attractors, influencing the growth direction of the surrounding leaf veins and “pulling” them in their direction. In this way, areas can be targeted according to an external stimulus in the form of hormone point distribution. As a visual stimulus, here are a few results of my (playful) gardening:
Beyond the simulation of leaf veins, it can be used to recreate natural textures. The following image shows a selection of my nature photographs in the top row. In the bottom row, I have matched each of them with a computer-generated equivalent.
The space colonization algorithm is thus externally controllable via an attractor distribution and should therefore have broader applicability in the technical domain than the L-systems described earlier. For example, what if the Auxin molecules were data points from a simulation? Or surface points on a geometry to be wired in an automated fashion? I think we can still learn numerous design principles from nature if we look at them systematically and specifically from this point of view.
A walk through the Munich South Cemetery the other day resulted in the following pictures, which I would like to share here for general entertainment. All these leaf veins, bark textures and ramifications present themselves in many ways. Nevertheless, I think they could be recreated very realistically with a handful of algorithms and some artificial randomness.
In the future, houses could grow like trees!
In the long term, biologically inspired software solutions could help us to find novel technical solutions and build better products. To a large extent, modern 3D printing already allows us to produce geometries that we have so far only been able to describe and design inadequately. To exploit the potential for innovation behind this, we need to start building a mathematical understanding and a repertoire of algorithms that will help us to mimic natural growth processes and control complexity in a targeted way. Then, one day, we may be able to build engineering products that will be limited only by the rules of physics, but no longer by human imagination and individual dexterity!
Houses could grow like trees according to environmental factors. The shape of an object would no longer be defined by a technical drawing, predefined points, and lines, but would result from the sum of environmental variables that directly influence the function of the respective object. Moreover, such solution principles could be applied across disciplines. Road planning and missile cooling ducts could be solved based on the same algorithms. Personally, I am very curious to see how this story will continue to unfold!
