Method for predicting cased wellbore characteristics using machine learning

A method for well integrity assessment across a depth interval of a cased wellbore having one or more casing strings using machine learning, comprising:

• a. collecting scattered X-ray signals from a plurality of X-ray detectors arranged within a well logging tool while that tool operates in one or more wellbores with known wellbore characteristics;
• b. associating the known wellbore characteristics with the collected scattered X-ray signals;
• c. training a machine learning model using the collected scattered X-ray signals and associated wellbore characteristics to produce a trained model that predicts wellbore characteristics from scattered X-ray signals;
• d. collecting further scattered X-ray signals from the plurality of X-ray detectors arranged within a well logging tool while that tool operates in a wellbore with unknown wellbore characteristics;
• e. applying the trained machine learning model to the collected further scattered X-ray signals to predict unknown wellbore characteristics;
• and
• f. assessing the well integrity using the predicted wellbore, characteristics,
• g. wherein the steps of collecting the scattered X-ray signals and collecting the further scattered X-ray signals each comprise collecting the signals synchronously from all of the plurality of detectors, and
• h. wherein the steps of training the machine learning model using the scattered X-ray signals arising from X-rays above one or more energy thresholds and applying the trained machine learning model to the further scattered X-ray signals arising from X-rays above one or more energy thresholds comprise training the machine learning model and applying the trained machine learning model using X-ray signals from two or more energy thresholds in conjunction.