A sandpit event was held 12-15 September at TUDelft on Geoenergy and Carbon Capture and Storage set up by Heriot-Watt GeoDataScience group and DARTS simulation group at TU Delft.

The workshop was organized by Dr Denis Voskov (TUDelft) and Prof Vasily Demyanov (HWU) with a keen support from Prof Sebastian Geiger and Dr Guillaume Rongier (TUDelft).

The sandpit brought together 7 PhD and 1 MSc students from TUDelft and 4 PhDs from Heriot-Watt to share expertise in reservoir characterization and modelling, explore new technologies and develop innovative solutions for energy transition. The event aimed to build synergy by combining  the complementary expertise between the two universities.

The first day set up the scene with the student talks who brought there research on the table to share with the room to establish the common tech ground:

• Object detection from seismic auto-segmentation by Quentin Corlay (HWU)
• Deep learning PDE proxy by George Hadjisotiriou  (TUD)
• Physics based ML in Geomechanics by Farah Rabie (HWU)
• Modelling Thermodynamics in DARTS by Michiel Wapperom (TUD)
• Coupled well and reservoir modelling by Kiarash Mansour Pour  (TUD)
• Learning geological patterns from outcrops with computer vision by  Athos Nathanail (HWU)
• Collocated Finite Volume scheme for scalability in geomechanical simulations by Alex Novikov (TUD)
• Geothermal Energy and implications of induced seismicity by  Ilshat Sifullin (TUD)
• UQ for EOR by  Gabriel Brandão de Miranda (Universidade Federal de Juiz de Fora, visiting student of TUD)
• GAN for fluvial facies modelling – learning from process models by Chao Sun (HWU)
• Improved understanding of NFR with Data Assimilation by Gabriel Serrao Seabra (TUD)
• Inverse modelling feature of DARTs based on the adjoint gradients by Xiaoming Tian  (TUD)
• Agent-based modelling for screening dynamics of subsurface systems, Quentin Corlay (HWU)

A series of expert  lectures from senior academics helped to focus vision and set up the landscape across the Energy transition field on the second day:

• Rapid Reservoir Modelling by Sebastian Geiger
• Introduction to Dynamic Simulations with DARTS by Denis Voskov
• Uncertainty Quantification in subsurface predictions by Vasily Demyanov
• Concepts of Machine learning and its application in subsurface workflows by Vasily Demyanov

The sandpit landscape was enriches by the number of subsurface data sets to be used though out the team work in the sandpit:

• Watt data set for AHM and UQ (HWU)
• Brugge data set for AHM and UQ (TNO)
• UNISIM I/III data sets for AHM (UNICAMP)
• Synthetic NFR data set (TUD)
• Fluid Flower lab experimental based flow simulation of CCS in heterogeneous porous media (TUD/UiB)
• DAP Well geothermal data set (TUD)

The students formed 3 teams to formulate challenging problems that requires synergy of the expertise between the two universities:

1. Multi-fidelity modelling for UQ with hybrid physics-driven and data-driven methods
2. Improve CCS flow prediction accuracy and geological realism with computer vision and fast proxy screening tools for UQ with high fidelity lab flow experiment data – Fluid Flower
3. Geothermal well placement optimisation under geological uncertainty for a fluvial system field using learning from physical process modelling.

We hope to conclude this activity in journal publications and joint future projects. The approach used in our workshop opens new opportunities in learning activities for graduate students.