Despite the overwhelming interest in artificial intelligence, a key hurdle remains for the large-scale adoption of ML-solutions; this is the “black-box” nature of the predictions where a model returns a value but with no obvious reasoning or tracking process of how the value was obtained. In this project, we shall focus on looking inside the black-box and producing not only the model’s prediction but also the reasoning the model took to deliver a given predicted value. This project aims to address this challenge through the use of eXplainable Artificial Intelligence (XAI) methodologies. The selected student will get the opportunity to work at the forefront of artificial intelligence methods, identifying how models come up with the solutions they offer and utilising that to further enhance the methods. To illustrate the benefit of XAI, we will focus on the use case of interpreting natural and induced damage zones at the wellbore wall and in the near wellbore region (e.g., natural fractures, drilling-induced tensile fractures, wellbore breakouts) from electrical, optical and acoustic well image logs. Naturally occurring fracture and fault systems play a key role in governing subsurface flow. Often, major flow paths are associated with large faults and their damage zones. Thus, detailed characterization of pre-existing fault and fracture networks, in terms of orientation, intensity and aperture, is vitally important for several industrial applications, such as water supply, hydrocarbon production, geothermal energy production, radioactive waste management, and carbon capture and storage. Drilling induced damage (drilling-induced tensile fractures and stress-induced borehole breakouts) can be of primary importance to depict the subsurface in-situ stress orientation and magnitude and, thus, for an optimal planning of well trajectory and monitoring of the wellbore stability. The numerical analysis will begin by training multiple ML models to predict the presence of a damaged zone and whether it is an open or closed fault. A range of models will be considered from highly interpretable decision trees to highly complex convolutional neural networks. Numerous XAI components will be investigated to provide a reasoning behind each models’ predicted values. This reasoning will be compared against our knowledge of the geological systems, allowing us to analyse if there is a physical meaning behind the model. This aids our understanding if and when we can trust a model’s prediction and the value of using it. The workflow developed in this project will be directly applicable to all ML-assisted log interpretation algorithms. A key outcome from this internship is to provide us with a clear understanding of which ML method is most suitable for wellbore damage zone prediction from wellbore image logs.