Use of Random Forest to predict the accumulation of plastic strain at grain boundaries of a polycrystalline material

Resumo

The main motivation is the study of the accumulation of plastic strain in the grain scale, through
the use of machine learning. This alternative can be a significant contribution towards creating models capable
of predicting the accumulation of strains. In this way, machine learning becomes a tool capable of helping to
understand which physical parameters control damage accumulation. The objective of this study is to predict the
accumulation of plastic strains at grain boundaries using the Random Forest model. For all machine learning
models, it is necessary to perform effectiveness tests and in this study cross-validation was used. It is a numerical
work, based on machine learning, which uses the Random Forest algorithm and cross-validation to authenticate
the model. The metric used to measure the performance of the model was the coefficient of determination (R2).
Results for the predictions of the accumulation of plastic strains, when considering the same microstructure, are
coherent and of good quality. When comparing the results obtained in this work with the predictions found in the
literature, the results obtained are satisfactory. Concluded that the Random Forest model is reliable for predicting
the accumulation of plastic strains in grain boundaries of a polycrystalline material.