AI-Based Prediction of Rock Mechanical Properties Using XRD, XRF, andGamma Ray Data

This study explores the use of machine learning to predict mechanical properties of subsurface rocks,such as indentation-based Elastic modulus, using input features from X-ray diffraction (XRD), X-rayfluorescence (XRF), and gamma ray measurements. Traditional mechanical testing is costly, time-intensive,and often unfeasible for large datasets. Our goal is to develop a predictive tool that leverages accessible fieldand laboratory data to estimate geomechanical behavior, enabling real-time formation evaluation duringdrilling. By integrating in-situ measurements such as gamma ray logs while drilling and surface XRD/XRFanalysis from cuttings, this approach supports live monitoring of the drilling process and improves on-sitedecision making without the need for routine lab-based mechanical testing. The dataset for this study wasdeveloped from direct experimental measurements on rock samples obtained during coring operations. Allmeasurements were carried out in the laboratory: gamma ray signals were acquired through core scanning,mineralogical composition was determined using XRD, and elemental content was measured with XRF.Mechanical properties were then quantified through depth-resolved micro-indentation testing, ensuringthat both geophysical and geochemical data were consistently tied to the same set of core samples. Toevaluate predictive performance, we trained and validated multiple machine learning models, includingRandom Forest, XGBoost, Gradient Boosting, Support Vector Machines (SVM), K-Nearest Neighbors(KNN), and Decision Trees. Models were tuned using cross-validation and assessed with standard regressionmetrics (R , RMSE). Feature importance analysis was carried out to identify the most influential inputvariables, highlighting gamma ray values (TotalGamAPI), selected oxides (SIO2, CaO), elements (Si) andmineral phases (quartz, carbonates and clays) as key predictors of mechanical behavior. Among all testedmethods, the best predictive performance was obtained by XGBoost and Gradient Boosting (R2>0.8),which consistently outperformed the other models in terms of accuracy and generalization. The featureranking from these models provided clear guidance on the most relevant parameters, offering the potentialto reduce the number of required laboratory tests. By integrating this approach into the drilling workflow,operators can infer formation stiffness and strength in near real time using surface cuttings and while-drilling gamma ray data, improving operational efficiency and reducing reliance on time-consuming mechanicaltesting. This work demonstrates the feasibility of using AI models trained on multi-source geochemicaland geophysical data to estimate mechanical properties relevant to wellbore stability and geomechanicalmodeling. The combination of micro-indentation data and accessible field measurements offers a practical,scalable alternative to traditional testing, with direct applications in real-time drilling optimization, wellboreintegrity assessment, and formation evaluation.