In the past two decades, the development of cutting-edge soft-computing technologies and their application to engineering problems has demonstrated huge potential to simulate complex non-linear problems. Machine learning and computer vision can play a significant role in supporting decision-making for infrastructure design, construction, and operation in several ways. They are often use for automated design and real-time optimization, virtual control of the construction process, to support real-time monitoring to make informed decisions regarding asset maintenance, repair, or replacement, estimation of the environmental impacts and optimisation of resources or processes to promote sustainability, and so on.
However, the development of robust prediction tools based on Machine Learning (ML) techniques requires the availability of complete, consistent, accurate, and large datasets. The application of ML in structural engineering has been limited because, although real-size experiments provide complete and accurate data, they are time-consuming and expensive. If we look at large infrastructure projects, the available data is often incomplete and associated with uncertainties or is difficult to interpret. Over the past decades, a vast amount of data has been collected about the condition of our structures and stored in asset management systems in reports, however, this data was collected in an unsystematic manner and often presented in a highly subjective way. The average data scientist spends more than 60% of their time on collecting, organizing, and cleaning data instead of the actual analysis. This is why there is an increasing trend of producing synthetic data. While synthetic data offers benefits compared to real-world data (e.g., increased data quality, scalability and interoperability), it is limited mostly due to bias, lack of realism and accuracy, and the inability to represent the response of complex systems.
In this talk, I will discuss the balance of real-word and synthetic data and how to best leverage the strengths of both to maximise the potential of ML to support decision making for design, construction and maintenance of structures and infrastructure. I will reflect on how ML algorithms and their application in structural engineering have evolved over the past decade, their potential and limitations, and the way forward. Finally, I will present several examples of how ML can be used to optimise structural design [1,2], to virtually control the construction process and minimise the impact on the existing environment [3], and to support visual inspection and maintenance of structures, providing a high level of consistency and automation [4].
References
[1] Cabrera, M., Ninic, J. and Tizani, W., 2023.. Eng with Comp, pp.1-19.
[2] Ninić, J., Gamra, A., Ghiassi, B., 2023. Underground Sp.
[3] Ninić, J., et al., 2017. Tun and Underground SpTech, 63, pp.12-28.
[4] Bush, J. et al., 2021. EG-ICE, Berlin, Germany (pp. 421-431).
