Análise da estabilidade de talude ferroviário sob influência de cargas e vibrações utilizando modelos de aprendizado de máquina

Abstract

Railways play a fundamental role in mining by promoting the transport of large ore volumes, and analyzing the stability of railway slopes is essential to ensure operational safety and continuous process integrity. Based on the Action Design Research (ADR) methodology, this study aims to evaluate the influence of loads and vibrations caused by trains on the stability of railway embankments, from the perspective of artificial intelligence – more specifically, machine learning. The analysis involved the construction of a database elaborated by carrying out stability analyzes with the variation of parameters and the subsequent training of four different algorithms: Decision Tree, K-Nearest Neighbors, Support Vector Machine and Voting Regressor. The evaluation of the algorithms by mean absolute error, SHAP plots and k-fold cross validation indicated that the KNN was the most suitable for the dataset. Finally, the developed artifact was applied with real data of the structure, where it was found that, although loads and vibrations have an impact on the calculated safety factor, the intrinsic variability of geotechnical parameters can cover these effects.