Why Optimization is not Generative Design
We have arrived at an interesting point in the development of design technology. Designers and engineers are beginning to trust computers with the very sacred task of design. Computers — our computational beasts of burden seem to have suddenly acquired magical abilities— thanks to Generative Design.
Architectural theorists were among the first to fantasize about the creative potential of computers — but they did not have the capacity to realize it. Engineers on the other hand, created CAD tools to make their work easier. Much of their time was wasted on improving and optimizing design through a process of thoughtful fiddling — which can now be achieved without thought.
All you now need to do is throw in some random numbers
You can throw random numbers on any parametric design and generate millions of design variation. I did this too, some time a ago and was thrilled by what I saw. But that was 12 years ago. No real progress has been made since then. All what we are witnessing is a desperate repacking of old ideas and work processes on faster computers.
If you unpack this package you will find within it - good old optimization, simulation, parametric design, BIM, a bit of AI, AR, VR and any other seasonal item such as Robotic construction, 3D printing, drone surveying etc. to convince you of how much more you need to know.
If it can be optimized can we call it design?
We live in times where professions rarely engage in tasks that machines can. Engineering offices no longer have teams of slide rule pros to do calculations. They have computers. Most companies (except the far thinking ones that fail) create tools for companies to do stuff that they are already doing — faster and cheaper. Big CAD companies are no exception to this rule. Just that they are under pressure to add some jazz to to their outdated tools based on even more outdated thinking.
Most practicing engineers may not like to believe that the bulk of their life’s work is optimization. Don tell them that. You will not be forgiven. Very few designs are created from scratch. They are mostly derived from well established models with well established ways of optimizing them. Optimization works well if you are optimizing one criteria. But then, there are no real world problems that come with a single criteria. So engineers optimize designs for various criteria and pick the one that combines them favorably. This is the way they have been working since the birth of their profession. There is nothing new in this. The big excitement for CAD companies is that now their CAD programs are connected with optimization programs. So now you get two for the price of one.
Design is about complexity
If there is no complexity in design, computers can crank through it without human interaction. You only need humans involved only for things that machines cannot do. For now, a good part of creative design remains in this category. So we need methods of design that can complement human efforts and support human exportation of design possibilities. But that is different from setting up the problem and hitting the ‘go’ button — that is called optimization.
Design is about finding the problem as much as it is about solving it. Both go hand in hand and resolve themselves jointly. This is why you cannot set the problem up in a computer and hit the ‘go’ button. If you can do that, you are not doing design — you are executing a routine operation. CAD companies build tools to eliminate these mundane operations. But they now wish to move upstream — but do not know how.
Meanwhile, they are happy to convince you that you can. All they need to do, is to let you tweak those parameters hidden deep in CAD files with sliders or with random numbers and — voilà! you have generative design and with 3D printing, you can also print it. Problem solved. Or is it?
Puny design spaces
The pundits of parametric design rarely discuss the size of the design space that they are able to explore. The game Go will give you some idea of what design spaces may look like.
Design spaces are huge. Tweaking parametric designs can only cover puny regions of it. Parametric Generative design lives in this puny space — but even with it, it searches blindly.
Knowledge is generated during design
Most designers would agree that they know a lot more about the design at the end of the design process than at the beginning. That is because they have gained significant amount of knowledge about the design problem through the design process. So when you setup the awesome generative scheme and hit the ‘go’ button, you actually know very little about the design problem and all the issues involved.
Designs process are iterative, work across multiple domains and generate new and useful knowledge — that guides the development of the design. Yes, simulations can be used for creating some of this knowledge, but they are not accurate enough with early stage design models. You need completed designs to gain reliable results.
Many important factors in design are not numeratable
This is a kicker. Game stopper. But then there are workarounds. The human designer can make the important decisions eg. on the size and shape of an aircraft panel and get the generative code to do some optimization inside it to reduce weight, increase stiffness and manage manufacturing costs. This is what is being done. It is not a bad thing to do. But this is not generative design. It is good old optimization bolted onto CAD tools.
What then is Generative Design?
This is a tough question. Wikipedia takes a good shot at answering it. Typically it has :
- A design schema
- A means of creating variations
- A means of selecting desirable outcomes
This is a good start; but there are many ways of structuring these components. What we now have in the form of optimization bolt-ons to CAD is the most impoverished form of generative design. Can we do better ?
to be continued…in the next article “ Generative Design : Fundamentals”
