Can AI design the next Burj Khalifa?

The idea of (CAutoD) Computer-Automated Design (not to be confused with Computer Aided Design) has been well around since the 1960s. The first instance of it was probably mentioned in the IBM Journal of Research and Development, where a program was written to search for logic circuits within the constraints of hardware design. With recent advancements in the field of AI, it is not be a pipe dream to imagine a world where high rises are designed by pieces of software.

The AEC industry is not exactly a poster child for constant innovation. The industry has focused far too much effort into simply growing while setting aside limited resources to ‘growing smart’. Perhaps the last major breakthrough in building design came with the advent of CAD software. Moving the tedious task of drafting, from the table to the screen saved countless hours of manual drawing and more importantly re-drawing. With this ‘digital drawing board’, architects and engineers could now shift more focus onto planning out the views and designs. All the drawings were still however created independently. The pressure on the design sector to generate efficient models that adhere to a plethora of codes, constraints and requirements in shorter time frames has been ever growing. The key to the problem most likely lies in automated computational design.

The next stage in building design came in the form of BIM or Building Information Modelling. With BIM, designs got smarter. Designs are no longer based on grouping sets of perspective views to communicate the overall picture. BIM is a 3D information model-based process that allows collaboration of multiple trades that work on an AEC project. A central project file can have multiple architects and engineers of different trades working on it, adding value, and coordinating simultaneously. Any perceivable view can be generated off a single 3 dimensional ‘master model’ that houses all data related to the project. With BIM it is not difficult to see the direction in which we are moving forward as an industry in that BIM models act as repositories of data which can aid in advanced design processes. Afterall, big data will only mean big value in the future.

Being a market leader, Autodesk has striven to be in the forefront of development in automated design. With ‘Project Dreamcatcher’ they aim to achieve just that. It is a system based on the principles of generative design in which designers assign goals with associated constraints. Based on this information, multiple design solutions that meet the criteria are produced for the designer’s perusal. The scope of generative design is seemingly endless. Design objectives or constraints may include functions, materials, modes of manufacturing, performance, cost, etc. Once the set of possible solutions are explored, the user may then choose to add more details or forward the solutions directly to manufacturing.

One possible application of the system was demonstrated with a project made in collaboration with Volkswagen. With the use of generative design, the team was able to model quirky looking designs for wheels, mirror mounts and steering wheel mounts. These odd structures were results of AI algorithms with constraints placed such as structural integrity, cost, and weight, among others. According to VW, these novel designs had saved 18% the weight of earlier designs. Of course, it would be naïve to think that we have reached a stage where generative design can model completely new pieces of automotive equipment that factor in every conceivable variable, but it isn’t too far down the road. With advancements in the field, designers will be able to add in more and more constraints and generate optimum designs that match the criteria.

With the onset of generative modelling in the AEC industry, we are now at the cusp of a revolution in building design. In the age of AI, building designs will be generated rather than drawn or plotted. Skeptics of the idea challenge that architecture and engineering design involve an intangible element of ‘artistry’ which cannot be replicated by a computer. This theory is not completely groundless, and we may in fact be far off from creating robots that can paint another Mona Lisa. However, advocates are of the notion that nearly all buildings in existence are in fact standard to a large degree. Most buildings are barely varied combinations of set boxes that forms rooms or spaces. A standard plan of a residential apartment will consist of a hall, a kitchen, a few bedrooms, adjoining bathrooms and connecting corridors. What seems like a seemingly endless set of possible layout options at first glance can be reduced to a few optimal solutions by an iterative engine. It’s tricky for a computer to be creative, but a computer is exceptionally good at performing iterative tasks. If the design process of a space is broken down, the same can hypothetically be carried out in its entirety by an automation algorithm. The challenge is to figure out a standardized workflow pattern that can cover all aspects that affect the process. Perhaps one of the best resources we have in this regard is the vast collection of data that already exists in the form of drawings and project files available in the industry. Machine learning algorithms can make use of these massive datasets and draw out patterns that obey cultural and legal codes.

The need of the hour is for energy efficient design. According to the 2019 Global Status Report for Buildings and Construction published by the UN Environment and the International Energy Agency, buildings and construction activity produce over 39% of global CO2 emissions, however action still trails far behind opportunity. Dramatic improvements in design, construction and operation are required if the AEC industry is to play its role in meeting the goals of the Paris Agreement. Efficiency must be part of the picture in the design stage itself. With the involvement of several analysis techniques and automation processes, energy efficient buildings can take form in the drawing room.  

Automated design is not limited in application to structural layouts. The principles of generative design can essentially be applied to all trades that would otherwise require teams of engineers and draftsmen. Novel shapes of skyscrapers that fit our functions can be generated in accordance with topographical data. Platforms like Hypar do just that. Once structural layouts are fixed, they can be fit out with HVAC, plumbing and electrical systems automatically. Moreover, as software can be ‘trained’ to consider thousands of design options, optimal solutions can be reached to maximize efficiency. Imagine a system that carries out CFD analysis of all possible duct network layouts and chooses the one that maximizes performance by creating a homogeneous flow field. It does not even stop there. Automation has extended into interior design and space planning as well. Through generative design modules that come with Autodesk Revit 2021, decisions can be made on optimal arrangement choices of tables in banquet halls to ensure effective utilization of space while maximizing other parameters. Tiles that fit a floor can be pre-cut and brought to the location saving time, material, and money. The possibilities are limitless.

In a future where machines design buildings, no last-minute client requirement is too drastic.  Increasing the energy efficiency of a structure is not an afterthought. We do not need weeks to generate plans for residential complexes or malls. The AI train is coming for the AEC industry and it will shift the trade into the new age of construction optimization. It is time we get on board.