Abstract
To advance sustainable underground development, current practices must evolve toward the next level: the Digital Twin (DT), a real-time, single source of truth that integrates data from disparate systems, optimises design, and enhances productivity. However, several challenges and knowledge gaps remain before DTs can be effectively deployed. These include the lack of generalised, scalable, automated approaches for digital modelling, limited data and issues with integration of real-time monitoring within DTs, and the need for coupling with advanced assessment tools and holistic evaluation frameworks.
The good news is that recent advancements in information technology, particularly in computational design, machine learning, and computer vision, have opened new opportunities to address these gaps. These technologies are driving higher levels of automation in the creation, execution, and maintenance of digital twins for underground infrastructure.
This talk will focus on the effective utilisation of recent developments in machine learning and scientific computing to automate the processes of DT recognition, execution, and update. As data lies at the core of each digital twin, I will emphasise the following strategies: i) leveraging the generalisation capabilities of pretrained models to minimise data requirements; ii) balancing the quantity and quality of data; iii) optimal generation of synthetic data; and iv) combining real-world and synthetic data to maximise the effectiveness of ML in supporting decision-making.
On the DT predictive side, I will demonstrate how the seamless integration of digital data and high-fidelity numerical models can transform geotechnical project planning through efficient, computationally enhanced decision-making. Finally, I will present how machine learning–based surrogate models, embedded within DTs, can enable real-time design assessment and process control during construction. Together, these advancements facilitate the creation of a living digital replica of the asset, enabling more effective and proactive asset maintenance.