Integration of geoscientific knowledge and explainability in machine-learning algorithms for the assisted lithological interpretation of well data

Integration of geoscientific knowledge and explainability in machine-learning algorithms for the assisted lithological interpretation of well data
In the context of the energy transition, the characterization of subsurface rocks is essential for CO₂ sequestration, geothermal energy and resource exploration (for example lithium or natural hydrogen). Leveraging existing data and automation through artificial intelligence (AI) are key drivers despite the complexity of geological formations and the challenges related to trust and explainability. The objective of this PhD work is to develop innovative machine-learning (ML) approaches that integrate geoscientific knowledge (geoscience-aware AI) for the automatic interpretation of well data (logs and high-resolution images) and the characterization of geological formations.
The PhD project is structured in particular around three main areas.
•    Spatial modeling of geological formations: use of sequential models such as Recurrent Neural Networks, Convolutional Neural Networks and Hidden Markov Models to capture spatial dependencies in the data, with rich and augmented datasets to strengthen robustness.
•    Incorporation of geoscientific knowledge: integration of knowledge graphs and regularizations based on physical principles to guide the algorithms and ensure the geological consistency of predictions.
•    Improvement of model interpretability: implementation of techniques such as Grad-CAM, t-SNE and uncertainty estimation to address trust issues and encourage the adoption of ML tools.
Supervision will remain structured yet flexible, allowing the doctoral candidate to explore different approaches and to develop a complete Python tool intended for external users. Methodological advances may lead to high-impact publications, notably because the issues addressed arise in many domains involving AI (for example physics-informed machine learning).

Keywords: machine learning, knowledge graphs, explainability, geophysical uncertainty, well logs, well images.

• Academic supervisor    Dr Marianne CLAUSEL, U Lorraine, ORCID : 0000-0002-5329-0801
• Doctoral School    ED077 IAEM, Université de Lorraine
• IFPEN supervisor Dr Francesco PATACCHINI, ORCID : 0009-0002-0660-7893
• PhD location    IFPEN, Rueil-Malmaison, France  
• Duration and start date    3 years, starting in the fourth quarter 2026 (November 2)
• Employer    IFPEN
• Academic requirements    University Master degree in Applied Mathematics or equivalent
• Language requirements    English level B2 (CEFR)    
• Other requirements    Python programming

To apply, please send your cover letter and CV to the IFPEN supervisor indicated here below.