Introduction
Civil engineers must maintain existing, ageing transportation infrastructure to ensure that it is safe and resilient, while minimising resource use.
This can be achieved by accounting for uncertainty and variability in ground and structural behaviour, for which traditional analysis techniques in civil engineering are poorly suited. This leads to over-conservatism in design and assessment that is unsustainable in terms of time, energy and cost. This project seeks to answer the central question: Can modern AI tools be used to design stable structures with more sustainable levels of conservatism?
Statistical methods have been proposed and developed for civil engineering analyses for many decades, but they have not been widely adopted, largely due to a lack of training and/or validation data. More recently, it has become much easier and cheaper to gather measurements from laboratory experiments and field monitoring that are suitable for statistical analyses. The Infrastructure Group at Southampton has gathered and published such data (Trinidad González et al. 2023; Huang et al. 2024; Briggs et al. 2024). Simultaneously, the increasing prominence of Statistical Machine Learning and AI has resulted in an influx of novel approaches to uncertainty quantification that could have significant impact in civil engineering. Members of the Statistics Research Group at Southampton have a track record in such methods (e.g. Cockayne et al. 2019; Oates et al. 2019).
This interdisciplinary PhD will explore whether the training of statistical models with appropriate input and validation data can be used to improve ground characterisation and the forecasting of structural performance. A stochastic finite element model will be developed to interpret uncertain ground conditions beneath civil engineering structures and forecast their long-term behaviour. These will include a trial embankment and a deep cutting in weathered clay. In each case, there are long-term ground monitoring datasets and associated ground investigation data available within the Infrastructure Group at Southampton to train and validate the model.
Stochastic back-analyses will enable engineers to determine the most likely composition of soil and rock strata providing the foundation of civil engineering structures. Statistical emulation will enable extrapolation of this approach for loading regimes (e.g. traffic, weather), ageing and ground materials (e.g. geology) applicable to a transportation network.
Research Aims & Methodolgy
Initially we will focus on established approaches based on Gaussian processes, such as the Bayesian finite element method (Poot et al. 2024; Girolami et al. 2021) or emulation techniques (Kennedy & O’Hagan 2021).
These are designed to accelerate components of the inference scheme described above, allowing for the computationally challenging exploration of design space to be accelerated, while still providing rigorous uncertainty quantification.
The research will be used to quantify uncertainties when assessing and managing ageing infrastructure, to enable sustainable decision making and reduce over-conservatism.
References:
Briggs, K.M., Trinidad González, Y., Powrie, W., Butler, S. and Sartain, N., 2024. Quantifying CPT cone factors in clays derived from weathered mudstone. Quarterly Journal of Engineering Geology and Hydrogeology, 57(1), pp.qjegh2023-014.
Cockayne, J., Oates, C. J., Sullivan, T. J., & Girolami, M. (2019). Bayesian Probabilistic Numerical Methods. SIAM Review, 61(3), 756–789.
Huang, W., Loveridge, F.A., Briggs, K.M., Smethurst, J.A., Saffari, N. and Thomson, F., 2024. Forecast climate change impact on porewater pressure regimes for the design and assessment of clay earthworks. Quarterly Journal of Engineering Geology and Hydrogeology, 57(1), pp.qjegh2023-015.
Kennedy, M. C., & O’Hagan, A. (2001). Bayesian Calibration of Computer Models. Journal of the Royal Statistical Society Series B: Statistical Methodology, 63(3), 425–464.
Girolami, M., Febrianto, E., Yin, G., & Cirak, F. (2021). The statistical finite element method (statFEM) for coherent synthesis of observation data and model predictions. Computer Methods in Applied Mechanics and Engineering, 375, 113533.
Oates, C. J., Cockayne, J., Aykroyd, R. G., & Girolami, M. (2019). Bayesian Probabilistic Numerical Methods in Time-Dependent State Estimation for Industrial Hydrocyclone Equipment. Journal of the American Statistical Association, 114(528), 1518–1531.
Poot, A., Kerfriden, P., Rocha, I., & van der Meer, F. (2024). A Bayesian approach to modeling finite element discretization error. Statistics and Computing, 34(5).
Trinidad González, Y., Briggs, K.M., Svalova, A. and Glendinning, S., 2023. Evaluating the likelihood of slope failure in ageing earthworks using Bayesian updating. Infrastructure Asset Management, 10(4), pp.207-222.