HPCIC INNOVATION CHALLENGE
Designing Clever Buildings for the Future with Supercomputing
How ML and Data-Driven Analysis can help us turn Buildings into Carbon Sinks
Hi, I am Aleksei Kondratenko, PhD Researcher at Politecnico di Milano, and AI Intern at DBF. I am writing a series of blogs exploring how supercomputing and machine learning can be used to address the carbon impacts of the building industry. In this blog, we will discuss ways data-driven design can help make buildings carbon sinks.
Building as Carbon Sink: How do we do it?
In my first two blogs, we learned why buildings are a major carbon source and why we should act fast to rethink buildings could become carbon sinks. In this blog, I will talk about how we can actually do this using two key technologies I am exploring at Digital Blue Foam: Data-Driven design and Machine Learning (ML). Specifically, we will talk about ML, and a subset of ML called “Deep Learning” (DL).
So what is machine learning? Simply put, machine learning allows us to create tools that learn and improve from experience (data) without being explicitly programmed to do so. It requires a lot of data to “learn” the system/function and the relationship between given inputs and outputs (if there are any). Of course, data-driven design and ML are not the only technology that could help us to transform our buildings into carbon sinks, but in my opinion, they could be the most impactful. Moreover, high-performance computers (or supercomputers) can help us tremendously to bring these technologies into design practice.
Data-driven design and ML are not the only ways to transform our buildings into carbon sinks, but in my opinion, they could be be the most impactful .
Data-Driven Design and Operational Carbon
Looking at operational carbon, data-driven design can help in the following ways:
- Data-driven Daylight Analysis: Daylight analysis is extremely important in the passive design paradigm to reduce the consumption of electricity in our buildings, allowing as much natural daylight in them as possible. As mentioned, to train any ML model we need a lot of data, but there are not many open datasets available for the Architecture Engineering and Construction (AEC) industry. Therefore, the common approach is to generate synthetic data to train ML models by yourself using 3D modeling software. Usually, as more data we have as accurate ML predictions will be. Therefore, access to additional computing resources (like Supercomputer) can tremendously accelerate the data generation workflow. As a result, a quick exploration of the effect of possible building massing and orientation on the potential energy use is possible. It could help us to significantly reduce operational carbon in the new buildings.
- Data-driven Digital Twins: Algorithmically, generated data can help us create better digital twins for the energy management of existing buildings. A digital twin is basically the digital copy of our building and all its assets that is updated in real-time. It uses sensors and Internet of things (IoT) technology to collect the most up-to-date data from our building. After some energy data (air temperature, electricity, heating spent, etc.) collection (that could take some time), ML could “learn” the building system and help humans to optimize its energy use, and eventually even autonomously control the building without any human intervention through something called Reinforcement Learning that is also conventionally considered as a part of ML. Later, these kinds of buildings could communicate with each other and positively affect the whole grid they are connected to and other “smart” building systems (or just take over the world as an alternative).
- On-site Electricity Generation: Predicting the impact and generated energy of solar panels depending on building shape configuration and location is getting attention but is not commonly applied in the practice by now. In any case, we must remember that going to carbon sink buildings from the operational carbon perspective depends a lot on the source of electricity that has to do a lot with the energy industry and potential data-driven applications there.
Data-Driven Design and Embodied Carbon
Looking at embodied carbon, data-driven design can help in the following ways:
- Optimizing Building Superstructure for Embodied Carbon via “surrogate models”: The evaluation of multiple design scenarios in the conceptual structural design could have a significant effect on embodied carbon. Different materials, structural configurations, column spans, etc. could be considered. However, structural analysis in sophisticated finite element model (FEM) software is required to prove the feasibility of any considered option. It can tremendously increase the time for any structural system generation and consequently decrease the number of options considered. ML comes in handy since it can be trained to predict structural displacements, weight, etc. given the geometric, material, and load condition inputs and accelerate the creation of any structural solution drastically increasing so-called design space. The data to train ML, in this case, can be generated synthetically in the same way as for the data-driven daylight analysis I mentioned before (and supercomputer can also come in handy here). Then, optimization techniques such as Genetic Algorithm (GA) could help engineers efficiently navigate through the generated design space and optimize the generated structural configurations. As a result, using this generative design technique the structural engineer is able to choose the best possible building configuration from the embodied carbon perspective.
- Facilitating the reuse of material after the building’s end of life: As I mentioned before in the article, the production of structural material emits a lot of embodied carbon and it could be significantly reduced if some material from to-be-demolished or totally unused/abandoned buildings is reused in the new structures. Additionally, various ML techniques could be used to create a database of available building stock from images and then connect people who need building material with people who have it from the existing building at the end of its lifetime.
- Building refurbishment: Other potential ML applications to turn our buildings into carbon sinks from the embodied carbon perspective could be focused on enhancing the refurbishment of existing buildings instead of building new ones and improving the efficiency and accessibility of offsite construction methods (modular, precast, etc.).
Final Thought
Today we talked about a few ideas for applying Data-Driven design to address the challenge of carbon emissions in the built environment and how supercomputing can help us with that. However, we are just scratching the surface! Across the world, researchers in universities and innovative companies are exploring many novel ways of applying these technologies to bring the building industry into a sustainable future.
Next time: How is DBF using data-driven design and supercomputing to reduce embodied carbon in new buildings? To find out, stay tuned for the final part 4!
About the Author
Aleksei Kondratenko is an AI Design Intern at Digital Blue Foam and PhD candidate at Politecnico di Milano. With a background in structural engineering, he aims to improve the sustainability of the built environment through the most modern digital technologies. He currently works on AI and ML applications in structural engineering presenting his work at prestigious global conferences like AI in AEC and DigitalFutures.
About DBF
Digital Blue Foam (DBF) comprises an elite mix of designers and technologists from around the world who share a strong commitment to empowering a revolution in architecture, engineering, and construction (AEC) industries toward carbon-negative projects by leveraging data-driven, AI-powered, collaborative, and sustainable approaches. We embrace collaboration and sponsorship, and we thrive at offering customized solutions that make designing a hassle-free and intuitive process. To learn more about Digital Blue Foam, visit our website.
