Prediction of rockfall hazard in open pit mines using a regression based machine learning model

This study investigates the feasibility of implementing simple Machine Learning models to make fast and reliable predictions of rockfall energies and run-outs at the base of highwalls. Probabilistic rockfall simulations are performed to generate a synthetic dataset of rockfall trajectories using high-resolution 3D photogrammetric models of fifteen highwalls from open pit coal mines. An automated software solution is developed to extract 2D profiles along the full strike length of the highwalls and meaningful geometrical features are identified and quantified. Four representative highwalls are considered for the model calibration and the remaining walls are used for validation. The block release position, slope local roughness and average slope angle are chosen as input parameters, whereas the kinetic energy at the first impact, the first impact position and the final run-out of the blocks at the base of the highwall are used as target parameters in developing predictive models. The application of various regression models is investigated for each target parameter and their performances are compared. A multi-linear regression model shows the best predictions for the kinetic energy at the first impact at the base of the highwall, while the first impact position and the final rockfall run-out are better predicted by a multi-nonlinear regression model. Overall, the models perform very similar. The results show reasonable applicability of the approach for a fast prediction of rockfall hazard for arbitrary highwalls based on the fully automatised extraction of geometrical features from 3D photogrammetric data.