Hèrm Hofmeyer

'The AI team’ delivers optimal building designs

Climate change, environmental pollution and resource depletion force us to build better and smarter buildings. 'The AI Team’ comes to the rescue. Hèrm Hofmeyer and his colleagues at the Built Environment department create ‘AI versions’ of the experts who play a role in the design of a building to have the design process take place virtually. These have already shown to improve structures beyond human capabilities. A brief introduction in five questions and answers.

What is your key research question?

With a growing world population, the need for strong and stable houses and buildings only increases. In addition, requirements are numerous: buildings must be energy-efficient, long-lasting, flexible in use, and this all with an increasing depletion of resources. In other words, we simply need much better buildings.

We have long seen a role for computers to support this. To optimize a building’s design, for example, by having a computer to determine the optimal dimensions of various structural components. But during the design, you are not only dealing with the structure, but also with architecture, building physics, lighting, heating and cooling, noise, and the flow of people and transport in and around a building. If you would predict the optimum design for all these disciplines, you end up with four or five different designs that really do not match.

Current studies make more complex and pleasant building geometries possible.

Therefore, together with my colleagues Bauke de Vries and Pieter Pauwels, and Michael Emmerich from LIACS Leiden, we work at TU/e on using AI to improve buildings and that what the building delivers: the design process. The process in which all the experts continually consult with each other in order to arrive at a better design. AI should bundle all the information and extract precisely those things that will help a design team to move forward.

What is the main challenge in your work?

Designing is really a field of its own. It can be compared to music: not a science in itself, but you can look at it scientifically though. And that's what we do. It’s a very difficult process to grasp, as designing a building involves so many different disciplines.

We try to simulate this process computationally. How does that work? For each discipline we develop a piece of software that pretends to be a person, a so-called AI agent. Armored with pattern recognition, knowledge based systems, rule mining, and neural networks, such an agent can take part in the design process.

For example, you start with a spatial design made by the architect, with only some shapes and lines. The AI ​​agent for structural engineering starts working with the design and determines where to put the walls, floors, and columns. Then there's an agent that will design the building from a building physics point of view, looking at things like insulation, lighting and so on.

Interestingly, these disciplines will subsequently 'virtually compete' with each other. They come up with contradictory ideas and after competing for a while they come up with a compromise and an alternative spatial design – just like in the real world.

Then the architect's AI agent reenters the picture and again it has his own demands, such as another organization of rooms, a different appearance, or simply having the front door facing the street. It therefore also enters into the ‘discussion’ with the other disciplines in order to arrive at a better design. This cycle takes place, as in practice, simulated several times in the computer.

AI pattern recognition allows for zoning, which provides better AI automatically stabilized structures.

Currently, these simulations lead to sensible results: structures are realistic, stable, and so they do what they are supposed to do. The related AI has shown to improve the structures beyond human capabilities. Worldwide, there is no other system that can do this, and it is open-source.

Of course, it remains a simplified version of reality. Thus, ‘softer elements’ such as aesthetics are missing: how appealing is a building? Something like that is difficult to capture in a mathematical approach. That is why we also join research projects to let AI lead to creativity.

What are the practical applications of your research? How does it benefit society?

Now that we can simulate the design process, we can start experimenting with the order of the disciplines in the process. Does it work better if the architect and building physicist go through the cycle with each other first before the structural engineer joins, or vice versa? This allows us to advise students and design teams in practice for a better outcome.

This is all the more important now that there are more and more cuts in the financing of design processes. While sometimes, with a design process that costs a hundred thousand euros more, you could make better choices and save up to a million euros on the construction or energy bills of the building itself. So, if we can make the design process more efficient, you quickly win a lot on the product side too.

Another interesting application is to run the simulations in parallel with practice, and to compare the designs with each other. Then the system becomes an assistant, a design support system. Thanks to AI, this is taking off, as the systems are getting better and better. Whereas architects in the 1980s didn't like simulated designs at all, now we can deliver building designs that are real architecture and that appeal to people too.


How do you see the development of AI in the future?

A long time ago I used to hear about the idea that you can use computers to replace a lot of people's tasks, but now I think that's outdated. AI works on hardware and software that has been made and invented by humans. Therefore, I can’t believe that AI systems can transcend themselves and become so smart that they would fabricate or design themselves. AI is fantastic and has tremendous potential, but likely will always remain subject to the conditions that humans give it.


Why would any AI researcher want to work at TU/e?

What I really like is the will to cooperate within TU/e. Not only is this stimulated from above, but also from below, cross-pollinations between different disciplines occur continuously. For example, on the advice of my supervisor at that time, I started taking courses in Computer Science during my studies. In addition, last year the AI institute EAISI was founded within the TU/e, which really gives AI research a tremendous boost.

Hèrm Hofmeyer (1972) is an associate professor in the chair of Applied Mechanics at the Built Environment department. He graduated from TU/e in 1994 on ‘AI in structural design’, using AI language Prolog, then already starting the line of research that he still follows today, almost 30 years later.

Hofmeyer was therefore the first researcher at TU/e to involve computer science in structural design, and thus began working with AI as early as the 1990s. The line of research, now together with LIACS in Leiden, has so far been financed with 5 PhD students.

Barry van der Meer
(Head of Department)

More Artificial Intelligence