Bridging the gap – how data is changing the future of building safety and resilience

In today’s world, the safety of buildings and bridges is essential.  

Whether it’s ensuring that skyscrapers can withstand the impacts of an earthquake, or that bridges remain steady during a storm, determining the response of buildings and bridges to different forces is crucial.  

But how do we predict the behaviour of something as complex as a bridge or a high-rise, especially when every structure is unique? 

That’s where Stefan Fuchs’ research comes into play, as part of the ‘Emerging Technologies’ theme led by Professor Robert Amor at the University of Auckland for the HERA-led and Endeavour-funded research to transform the construction sector of Aotearoa using Construction 4.0 approaches.

Stefan, who recently completed his PhD, is diving deep into the structural sensor data to help engineers better predict how buildings will react to the natural and man-made forces they face daily to ensure the safety and comfort of occupants, and longevity and maintainability of these structures as well.  

A data-centric approach to structural safety

Typically, engineers rely on physical models of the structures – like Finite Element (FE) models to simulate and capture the properties and behavioural characteristics of materials and connections. These models, while effective, can’t give 100% accuracy and require simplifications and assumptions to make the calculations manageable. In real life though, buildings are much more complex! 

Factors like material behaviour, how a building ages, and the unique forces acting on it over time can make it challenging to capture a complete picture of a structure’s health. 

Stefan’s research, however, takes a different path. Instead of solely relying on theoretical models, or approaches such as Bayesian updating of the FE models which are based on sensor data (e.g., measurements taken with accelerometers) obtained from monitored structures to increase accuracy of the model, he is exploring how data collected from the real world can be mined to tell us more.  

This approach is being explored as traditional models can be computationally intensive, often assume linear behaviour, be unavailable for existing structures, and become outdated due to changes in response characteristics caused by the natural deterioration of the structures.  

The future of smart structures

Stefan’s research approach has been successfully applied by Li et al. (2024) for simulated data for buildings with similar designs and structures, and builds on this by exploring the feasibility of this approach trained with real-world data. 

Using sensors – such as accelerometers installed in buildings, he is utilising the vibrations and movements that occur naturally, or during events like earthquakes, to establish a foundation model for structural response prediction.  

This in turn can be fine-tuned to predict how a structure will respond to future events. Think of it like giving the building a ‘health check,’ using data to forecast potential issues and better prepare for them. 

For our Ngākopa Construction 4.0 research, the work Stefan is doing is very exciting as it may improve structural response predictions for existing buildings or structures with no pre-existing models as many existing buildings weren’t constructed with modern technology, making it hard to assess their current safety.  

By using real-world data, Stefan’s models can “learn” from the ambient vibrations and movements data obtained from monitored existing structures, enabling engineers to forecast how they would behave in a disaster, even without a detailed physical model. 

Additionally, this research has the potential to make monitoring more dynamic without requiring continuous monitoring systems to be installed on buildings. As buildings naturally deteriorate over time, their responses to forces change too. Stefan’s data-centric approach allows for continuous updates, ensuring that the models stay relevant and accurate as the buildings age. This is particularly useful for infrastructure like bridges, which often serve communities for decades and need ongoing assessment to remain safe. 

Li, Z., Yang, Q., Deng, Q., Gong, Y., Tian, D., Su, P., and Teng, J. (2024). Time history seismic response prediction of multiple homogeneous building structures using only one deep learning-based structure temporal fusion network. Earthq. Eng. Struct. Dyn. 

About Stefan 

Stefan Fuchs has just completed his PhD at the School of Computer Science at the University of Auckland, New Zealand, under the supervision of Prof. Robert Amor, Prof. Michael Witbrock, and Dr. Johannes Dimyadi. His PhD thesis focuses on employing and training large language models to interpret building codes and make them applicable to automated compliance checking. In his capacity as a research assistant for the Construction 4.0 Project, he is connecting the research theme “Computing Technologies for Construction 4.0” with the research sub-program “Monitoring 4.0” by investigating the practical application of cutting-edge neural networks to analyse, predict, and forecast structural responses recorded by accelerometers.