The AI journey in architectural engineering with computational design optimization

Buildings and the built environment form a complex system where multiple factors shape how we live and behave in those spaces. As a result, architectural design is a demanding process that requires an understanding of how design choices both building users and the environment. The performance demands are many and diverse. Highly performing buildings are increasingly necessary, especially in light of urgent sustainability goals. Achieving this level of performance requires more than just solid engineering — it requires an integral design approach from the earliest stages.

Today, computational design methods make it easier to incorporate different criteria into integrated models. These methods help multidisciplinary design teams analyze and compare different scenarios in which various criteria are prioritized. Among the most widely used methods are parametric design, computational performance assessment, computational optimization and data analytics. Parametric design enables the generation of geometric design alternatives, which can then be evaluated using measurable criteria computed with computational performance assessment techniques such as digital simulation. Optimization algorithms and data analytics are used to explore optimal and sub-optimal parametric solutions to support design decisions and extract design insights.

Computational design: the intersection of architectural design and building technology

At the Chair of Design Informatics at Delft University of Technology (TU Delft), computational design is an active area of research, according to Associate Professor Michela Turrin. “We develop computational methods - including the use of modeFRONTIER software - to support scientific investigations into building performance from the earliest stages of the design process, when the most impactful decisions are made,” Turrin said. “In particular, our research focuses on putting forward computational design in order to integrate engineering aspects into the very early stage of architectural design conception and trigger design innovation with sound fulfillment of performance criteria.”

This integrated approach to architectural design and the engineering process involves three main components, each incorporating different computational methods and techniques:

• Form generation: parametric modeling and AI-based form generation.
• Performance assessment: computational simulations and AI-based surrogate models.
• Optimization for multidisciplinary design exploration: computational design exploration and optimization.

Architectural design is inherently multidisciplinary. Different disciplines may have completely different priorities and speak different languages. When it comes to data, each discipline may analyze data differently. As a result, there’s a need to establish a common language that can be understood across disciplines. Multidisciplinary design exploration refers to anything that helps designers understand the consequences of their decisions - it’s essentially a scenario-based approach. Data plays a critical role in understanding “what if” scenarios, revealing how certain design choices impact building performance. Multidisciplinary design exploration refers to anything that helps designers understand the consequences of their design decisions - it’s essentially a scenario-based approach. Data plays a critical role in understanding “what if” scenarios, revealing how certain design choices impact building performance. Every building design is an iterative process that balances intuitive, creative and diverging exploration with rational analysis, synthesis and convergence. In this articulated process, computational design optimization supports the investigation of multiple objectives that influence overall building performance. It allows designers to explore a large number of alternatives under different simulated conditions and identify optimal or preferred design solutions.

At the intersection of architectural design and engineering, computational design helps clarify the related options and their implications. Unlike traditional engineering optimization — which often focuses on producing a single mono-disciplinary solution — this approach embraces multiple trade offs. Combining multi-objective optimization algorithms with data analytics provides clearer insights into the relationships between design parameters and their corresponding performance. Multi-objective methods are especially relevant given the wide range of conflicting performance criteria that must be considered. Data analytics techniques are also essential when handling a high number of variables, which can make life difficult for designers seeking to understand correlations. Despite its advantages, this process relies heavily on performance assessment of numerous design alternatives — often based on digital simulations. These simulations can be time-consuming, sometimes taking hours even for simplified models. The use of machine learning to create surrogate models significantly speeds up this process, making it more suitable for the fast timelines of early design stages, when changes are elaborated quickly.

Integrated architectural design and engineering research projects at TU Delft

“Shaping the built environment toward the design of highly performing buildings is the ultimate goal of our research activity at TU Delft” added Michela Turrin. “By shaping, we refer to the design process across various scales — from individual building components to the broader urban scale. In this context, one of the crucial points is always the collaboration aspect and the way performance data has been generated from computational workflows. We try to use these data to enhance discussion across disciplines as a way to facilitate brainstorming and create a common ground that can be used to negotiate across different ones on the different pros and cons of design scenarios. Indeed, collaboration based on building performance data is the foundation of our MEGA course for our students.”

MEGA course projects: conceptual and design development phases of a multi-functional high-rise building

MEGA is a MSc 2 course running at the TU Delft Faculty of Architecture. The course centres on the collaborative, multi-disciplinary redesign of a large or tall building. Teams of 5 to 7 students develop an integrated design that spans the disciplines of architectural design, including climate design, computational design, facade design, structural design and management.

This collaborative and multi-disciplinary process is supported by a digital design environment that includes 3D modeling through parametric design, Business Information Modeling (BIM), performance assessment and computational design optimization software such as modeFRONTIER. As part of the MEGA course, ESTECO application and support engineers lead a workshop for students that focuses on computational optimization techniques to explore design alternatives.

Each student takes responsibility for a specific role and is guided by an expert in that discipline. Working as a team, students share a computational workflow designed to facilitate collaboration across different fields. This shared workflow enables each discipline to analyze the design and assess the impacts of specific design decisions, recognizing that different disciplines may have competing criteria and priorities. As a result, the workflow becomes a platform for negotiation, discussion, and interdisciplinary brainstorming. Here’s a glimpse of some design challenges tackled by MEGA course students over the years, where interdisciplinary computational processes play a big role.

1) Structural design of an atrium roof while considering the impact of microclimate conditions created underneath the roof

2) Overall building massing in relation to multiple functions, including housing, hotel, offices, data center, fab lab, and distribution center

3) Distribution of floor area across the building’s levels to meet functional requirements

4) Direct connection to the facade based on different building massing configurations

5) Solar radiation received by each facade module compared to indoor climate conditions influenced by solar radiation and daylight

6) Leveraging complex geometry to reduce costs, conserve materials, and improve sustainability in high-rise building construction

From shaping to understanding build environment

Techniques such as optimization, surrogate models, correlation, and cluster analysis are often tailored to the specific requirements of individual design cases. However, there is a growing need for a broader understanding of the features of the built environment to gain insights that extend beyond one-off designs and carry general value. This challenge was explored by Evangelia Despoina Giouri in her master’s thesis which focused on understanding the impact of design decisions towards zero-energy high-rise buildings specifically under different climate scenarios. By coupling the Rhino/Grasshopper modeling environment with EnergyPlus software to simulate energy consumption and daylight illuminance levels within a modeFRONTIER workflow, she was able to define which construction parameters have the highest impact on annual energy demand and thermal comfort in the building.

Stephanie Mumjani's thesis examined how different height scenarios in residential high-rise buildings affect compliance with the BENG 2020 regulations — a Dutch standard for nearly energy-neutral buildings. The study focused on energy demand and local energy generation, aiming to determine whether these regulations impose height limitations despite the use of optimal design solutions. By evaluating a wide range of design scenarios for a residential high-rise in a temperate climate, she assessed energy performance, energy loads, and occupant thermal comfort. She developed an integrated computational methodology that combines parametric modeling (Grasshopper), energy simulation analysis and modeFRONTIER design optimization software to explore and optimize building design parameters effectively.

In his master’s thesis, Francisco Guzman examined how different properties of green facades influence the surrounding microclimate and the thermal performance of building walls. His research involved parametrizing both green facade and living wall systems to assess their responses under different initial conditions. He developed a cutting-edge computational workflow by integrating ENVI-met, Rhino/Grasshopper and modeFRONTIER, along with Python 3 scripting, to evaluate the performance of vertical greenery systems,

In another example of an indoor application, Kees Jan Hendriks explored ways to improve the thermal energy performance of a phase change material (PCM) Trombe wall and develop an economically feasible product. This passive system lowers the total cost of ownership for the energy system within an office building in the Netherlands by decreasing both the energy demand and the peak loads on the mechanical system. He integrated a MATLAB/Simulink simulation model into a modeFRONTIER workflow to perform multi-objective design optimization. The goal was to design an optimal system that minimizes energy consumption while using the least amount of PCM material necessary to achieve thermodynamic efficiency and cost-effectiveness.

Strategic partnership between TU Delft and ESTECO: key focus areas

Our collaboration with TU Delft aims at advancing computational workflows for architectural design through five key pillars. Here’s a breakdown of each focus area and the projects where we’re involved in.

1. Generative design and (re)parameterization, including computer vision.
2. Fast performance assessment.
3. Extracting information from large datasets.
4. Integration of quantitative and qualitative criteria.
5. Interrelation with materialization.

The first focus area addresses the challenge of parametrization — specifically, identifying which parameters are most meaningful to include in the design process and determining a reasonable number of parameters to incorporate. This issue was explored in the PhD research of Ding Yang, titled "Design as Exploration: Multi-Objective and Multi-Disciplinary Optimization of Indoor Sports Halls". Yang demonstrated the integration of computational optimization techniques by linking Grasshopper, Daysim, EnergyPlus, and Karamba3D analyses within the modeFRONTIER workflow. His work focused on achieving a balance between structural performance and daylighting quality through the optimization of roof geometry. He also developed methods to simplify complex multi-objective problems by strategically reducing design variables while preserving essential performance criteria.

The second key focus area, particularly relevant during the early design phase, is to enable rapid evaluation of multiple performance criteria to guide decision-making and prevent expensive modifications later in the process. At TU Delft, the goal is to use computational simulations to develop AI-based surrogate models. These models will allow designers to quickly assess performance across various design problems, operating within the validity limits of each surrogate model. ESTECO supplied the TU Delft research team with modeFRONTIER and its AI-based surrogate models for the project “Integrated Approaches for the Energy Transition in Existing Buildings - Digitizing the Renovation Process”. The objective of the project was to create an automated process for optimizing and manufacturing building components that support passive design strategies aligned with energy transition goals. Beginning with digital models of existing buildings, the workflow integrated parametric design, optimization, and 3D printing, leveraging machine learning neural network-based thermal simulations for rapid performance assessment.

The third and fourth pillars of our partnership reflect the need to extract information from large datasets and make it accessible to different disciplines, while also integrating both quantitative and qualitative criteria. In his master’s thesis “Visual Analytics for Generative Design Exploration”, Jamal Van Kastel proposed a highly interactive, game-like data environment. This tool visualizes high-dimensional data alongside building geometries and processes for a wide range of sports hall design alternatives. As part of the computational design workflow, modeFRONTIER was used to create bidimensional representations of complex multi-dimensional data with its self-organizing map (SOM) post-processing tool.

The final focus area involves the connection to materialization. We participated in the Double Face 2.0 project, where the TU Delft research team aimed to design a prototype that passively enhances thermal comfort in indoor and semi-indoor spaces. The design used lightweight materials for latent heat storage while maximizing daylight transmission, achieved through additive manufacturing techniques.

Potential impact of computer science on the future of architectural design

In summary, we will continue to support TU Delft’s Architectural Engineering and Technology department by providing our process automation and design optimization technologies to tackle one of the key challenges in the architecture, engineering, and construction industry: delivering higher-quality buildings while conserving the planet’s finite resources. TU Delft’s research approach focuses on advancing and applying cutting-edge computational design and manufacturing methods, digital tools, and innovative technologies to create sustainable solutions for both new construction and the revitalization of outdated buildings. Computer science is central to this mission, enabling multidisciplinary collaboration, fostering iterative and exploratory design processes, and integrating both qualitative and quantitative criteria into architectural solutions. Through computational design, building performance can be optimized, resource efficiency improved, and resilient, environmentally responsible architecture for the future can be realized.