Predicting Physical Properties Using Neural Networks: A Case Study from Ludvika Mines-Central Sweden : Machine-learning applications for predicting unmeasured rock physical properties

This MSc project aims to leverage the power of machine learning solutions to predict subsurface wave velocities for other measured properties, a crucial parameter in seismic exploration and characterization of mineral deposits. Specifically, the project will develop a neural network model that utilizes rock properties, lithology, and depth information extracted from borehole data from the Blötberget mine in the Ludvika region in central Sweden. By training the model on these parameters, it seeks to generate reliable sonic velocity predictions for other boreholes within the same area.There are six boreholes where these properties are measured and the goal is to predict the velocities fo rborehole where only limited other information is available.The significance of sonic velocity lies in its direct correlation with the subsurface geological structure,which is accountable in identifying potential mineralized zones but also in their use in for seismic imaging applications such as for migration and time-to-depth conversion. Traditional methods of obtaining these velocities, such as direct measurement or empirical correlations, are often time-consuming, costly, and subject to significant uncertainties. By adopting a machine-learning approach, this MSc thesis endeavors to bypass these limitations, offering a cost-effective, accurate, and efficient alternative in the exploration phase.