Foster + Partners is a global studio for architecture, urbanism, and design rooted in sustainability, which was founded over fifty years ago in 1967 by Lord Norman Foster. Since then, the practice has established a worldwide reputation for thoughtful and pioneering design, working as a single studio that is both ethnically and culturally diverse. The studio integrates the skills of architecture with structural and environmental engineering, urban, interior, and industrial design, model and film making, aeronautics, fabrication, and many more – our collegiate working environment is similar to a compact university. These diverse skills make us capable of tackling a wide range of projects, particularly those of considerable complexity and scale. Design is at the core of everything that we do. We design buildings, spaces, and cities; we listen, we question, and we innovate.
Technology forms an integral part of our workflow and often is one of the catalysts for innovation. Any idea usually leverages some form of technology to make it real. Several interdisciplinary groups at the practice are involved in areas such as computational design, building physics, performance analysis, optimisation, fabrication, and interaction design. We conduct state-of-the-art research and development in collaboration with universities and industry partners, exploring far-reaching ideas from bio-inspired engineering to extra-planetary 3D printing.
We developed a web application called Hermes which facilitates sharing of design data in real-time across different applications, computers, disciplines, organisations, and locations around the world. Client plugins for Grasshopper, Dynamo, Unity, and Excel among others – communicate with the application to send and receive data (Figure 1). The Hermes approach to interoperability is message-based rather than model-based. This ensures data consistency through the transfer of relatively small amounts of data that are used as the basis for downstream computation by the subscribing client applications.
The AEC industry has increasingly embraced parametric design in the last decade to the point where it has become a critical part of the digital design toolbox in many practices. Parametric computation often takes the form of ‘data flow’ programming where data is transformed through a series of algorithms in a flow specified by a designer. Hermes extends this concept to bridge across parametric models (Figure 2): the output of one parametric model becomes the input for another (serial flow) or several models constructed from one source (parallel flow).
One of the most valuable contributions Hermes has made to our design process is to use this extended data flow capability to eliminate the need to manually rebuild models for different purposes: architectural design, structural and environmental analysis, visualisation, design documentation. Each of these disciplines has its own parametric definition that can be automatically updated by a ‘recipe’ provided by another discipline through Hermes.
For instance, in a traditional workflow, structural engineers might rebuild an analysis model based on a wireframe received from architects which they would then analyse and convert into a BIM model with appropriately dimensioned beams, columns and slabs. Using Hermes, when the engineer receives a wireframe ‘message’ from the architects, their parametric model automatically constructs a model suitable for analysis. The engineer can focus on analysis and evaluation rather than tediously rebuilding the model before they can begin their work. Once their analysis is completed, they can send centrelines, dimensions, and profile information to another parametric model to generate the structural BIM model.
Because Hermes is a web application, it can be integrated with other web tools and data stores to promote useful workflows. Safely and securely enrolling external consultants is possible, and we have a very successful interface with Microsoft Teams which publishes notifications when new Hermes messages are available and facilitates discussion between all the collaborators (Figure 3).
In collaboration with Autodesk in 2017, we used Machine Learning to predict how a passively actuated façade would react to temperature change.
The idea for this project came from constant advances in material science, more precisely, the developments around passively actuated materials. Those materials have the capability of deforming not as a result of a mechanical actuation, but rather based on changes to their physical environment e.g. thermal fluctuations, changes in light conditions and even humidity. These materials have incredible potential in the future: imagine a façade that doesn’t have any shading devices but self-deforms based on external conditions to accommodate shading or privacy (Figure 4).
Working very closely with a team at Autodesk, our Applied Research + Development team has been investigating how laminates behave (Figure 5). The behaviour of the material was the primary focus. As architects, we’re interested in defining the start and end state of material – specifying, thus, its morphological deformation. But as these materials do not comply to a linear deformation, it’s very hard to predict what their initial layering should be in order to deliver a certain deformation under desired conditions. One way this problem could have been approached would be to run Finite Element Analysis (FEA) on a particularly layered thermoactivated material, check its structural deformation, slightly change the layering, analyse again and repeat until the desired results were met. However, this would have been extremely time-consuming and would only yield grossly approximated results.
Therefore, we decided to experiment and solve the inverse design problem, where the structural deformation would be the input, rather than the output. This process could provide the desired layering. Since obtaining an analytical solution was not possible, we decided to use Machine Learning and build a predictive model. As every predictive model requires high quality data to learn from, we used Foster + Partners’ in-house parallelisation engine, called Hydra, which allowed us to run hundreds of thousands of non-linear FEA for a plethora of sampled laminate layerings tens of times faster than any out-of-the-box software. Subsequently, we used the dataset created (the deformations that derived from initial states) and fed it into a generative adversarial neural network trained on pairs of input nonlinear FEA simulation results and cut-out patterns of laminate layers which caused a given deformation (Figure 6).
A trained predictive model was able to generate material layerings for a given deformation within seconds. This allowed us to prototype a simple application where anyone could experiment with various deformations interactively and be presented with the cut-out patterns almost instantaneously. This workflow not only challenged the way such laminates are currently being designed but also offers the possibility of reducing the prototyping phase to minimum. These results proved the effectiveness of the proposed workflow, especially at an early design stage, which could be further extended to the design of any nonlinear mechanical deformation. The details of this research are published in a peer-reviewed article [A. Abdel-Rahman, M. Kosicki, P. Michalatos, and M. Tsigkari, “Design of thermally deformable laminates using machine learning,” in Advances in Engineering Materials, Structures and Systems: Innovations, Mechanics and Applications, A. Zingoni, Ed. London, UK: CRC Press, 2019, pp. 1016–1021.] which is available to the scientific community or curious readers.
Foster + Partners won first prize at a competition staged by NASA to design a habitat for astronauts living on Mars (Figure 7). This project takes construction, materiality, and design to an extreme by building in a remote landscape on an uninhabited planet. Resilience and robustness are therefore major factors, with material optimisation being essential as everything has to be transported from Earth.
We explored the use of autonomous robots, inspired by swarming termites, to build habitats without human input.
The research focused on the overall mission architecture, improving human comfort, materiality, robot autonomy, and the 3D printing methodology, all of which are uniquely suitable for an autonomously constructed habitat in an extreme environment far from Earth. Distributing risk across multiple simpler units working in parallel can improve chances of success, rather than the traditional consolidation into a single complex unit.
The multi-robot swarm includes agents with different functions that additively sinter layers of regolith (the layer of unconsolidated rocky material covering solid rock) into a protective shield over an inflatable pressurised module. This occurs prior to human arrival using local materials to work in conjunction with other un-manned deliveries of habitat elements. The lightweight inhabitable element offers the high-tech aspects of life-support, manufactured in controlled environments on Earth (Figure 8), whilst the heavy-weight shield acts as protection from radiation, micro-meteorites, and dust storms.
The work carried out when designing structural elements using autonomous robots led us to reappraise how we design for strength; also, working with limited resources taught us how to design in a more sustainable way on Earth.
Following our research on the Paris Agreement and Sustainability rating systems worldwide, we realised that embodied carbon is an important issue that requires a more comprehensive approach that is embedded into the design process.
The embodied carbon viewer is an online platform to view, analyse and report on embodied carbon (Figure 9). The tool has two primary inputs; firstly, the building geometry and information, in the form of Revit models is uploaded. In addition, the tool uses a global database of materials and products, which include the carbon impacts associated with materials extraction, transport, processing and manufacturing, all of which are localised to the project region. The user can rapidly move around their design, which is fully visualised, assigning products and switching them for alternatives to understand potential carbon savings through materials sourcing.
The tool integrates Revit models from different disciplines including structural, MEP, Architectural and Landscaping, automating a process that would usually take weeks into just a few hours (Figure 10). This relatively new platform is already being used across the practice, allowing all project stakeholders to understand the embodied carbon breakdown of their project and empowering them to make the critical decisions to help in the fight against climate change.
Algorithms, of course, play a big part in our every-day working environment. Using generative design in Rhino/Grasshopper combined with custom tools developed in-house, we have converted design models into full featured BIM models and optimised the arrangement of apartment typologies in a residential tower (Figure 11). Custom tools were developed using Dynamo to read data from the Rhino design model.
The first script reads a spreadsheet generated from the Rhino model which is a matrix showing apartment typologies arranged by level and orientation. The script then automatically generates a BIM model by placing, mirroring and rotating instances of linked apartment Revit models. The second script directly reads model data (curves) from Rhino and automatically instantiates balustrades at the correct location inside the BIM model
Each project creates its bespoke ‘ecosystem’ of tools using the most suitable software for each task – in a masterplan project, we built in automation using Dynamo to create the BIM model, and then built tools to translate the BIM model to GIS (Geographic Information System). The automation included a tool to outline the plot boundaries from CAD files, annotate sheet views for each plot (in excess of 25,000 individual plots), and to automatically number and annotate plots, subdivisions, and blocks on plots (Figure 12). The GIS translation tool is completely unique to Foster + Partners and allowed the direct translation of the BIM model to a fully featured GIS Shapefile for geospatial analysis purposes by the Urban Design Group.
A key part of designing at the city scale is understanding the different contexts we are working in and being able to quantify different aspects of the urban environment to assess the impact of different design strategies (Figure 13). To facilitate this process, we have developed a mix of tools that take advantage of open data APIs and leverage the capabilities of ArcPro SDK, ArcPY, and other geospatial libraries for Python. The three main tools include 1) an urban data extraction and pre-processing add-in for ArcPro, 2) a walkability analysis web-app, and 3) a series of urban form analysis algorithms. Together these tools help to create context analysis for a diverse range of projects, as well as quantify and benchmark different aspects of the built environment to inform design decisions.
The urban data extraction and pre-processing add-in for ArcPro automatically fetches data from various API’s to construct geospatial datasets related to buildings, street and transport networks, natural features, amenities, satellite imagery, and digital elevation models. These datasets are automatically formatted and visualised to be used for urban context analysis, including 3D models that incorporate information on land uses (Figure 14).
The walkability web-app gives a proxy for propensity to walk in any given location. The web-app is built on top of the urban data extraction tool, and adds additional capabilities by modelling street-networks as graphs and calculating distances from a given point to all key amenities such as schools, libraries, banks, restaurants, cafes, etc. The tool calculates a walk-score by assigning weights to different amenity types, incorporating a distance decay function, as well as additional information about block size and morphology (F1gure 15). The urban form analysis scripts are also built on top of the urban data extraction and pre-processing tools, and provide additional capabilities to quantify urban form. This includes street network metrics and urban form analysis that can be used to benchmark different urban contexts.
Technology, in many cases, allow us to not only realise ideas but also to work smarter. As a practice, we are continuously assessing how we work. Innovation results in incremental improvement. While some of the technologies seem futuristic, it is only because by the time our designs have come to fruition, we have reached the future.
I would like to thank all those who have contributed to this article including Francis Aish, Martha Tsigkari, Adam Davis and Marcin Kosicki from our Applied Research + Development Group; Irene Gallou, Joshua Mason and Sam Wilkinson from our Specialist Modelling Group; Chris Trott and William Allen from our Sustainability Group; Bruno Moser and Mateo Neira from our Urban Design Group; and Cliff Green and Boleslaw Musierowicz from the BIM & Design Systems Group.
Han Shi is the Head of BIM & Design Systems at Foster + Partners, an award-winning British architectural design and engineering firm. Han directs and leads the implementation of BIM across the 1,400+ employee global firm. An architect by profession with a passion for technology, Han has 40+ years’ international experience implementing creative technology solutions to increase productivity, efficiency and provide new services in AEC companies across the world, whilst always ensuring optimal design and operational effectiveness. He has implemented BIM and CAD methodologies in different size firms ranging from a 2,500+ employee multi-disciplinary firm in the Middle East, to a mid-size US based architectural firm, to international Hong Kong based interior designers, to small architects in England, and has managed projects of different sizes ranging from small residential properties and palaces, to multi-use complexes, luxury hotels and resorts.
Have comments or feedback on this article? Visit its AECbytes blog posting to share them with other readers or see what others have to say.
AECbytes content should not be reproduced on any other website, blog, print publication, or newsletter without permission.
