| Resumo: | This work aims to evaluate the performance of various machine learning algorithms in the prediction of metal forming defects, particularly the occurrence of edge cracking. To this end, seven different single classifiers and two types of ensemble models (majority voting and stacking) were used to make predictions, based on a dataset generated from the results of two types of mechanical tests: the uniaxial tensile test and the hole expansion test. The performance evaluation was based on four metrics: accuracy, recall, precision and F-score, with the F-score being considered the most relevant. The best performances were achieved by the majority voting models. The ROC curve of a majority voting model was also evaluated, in order to confirm the predictive capabilities of the model. Globally, ML algorithms are able to predict the occurrence of edge cracking satisfactorily
|
|---|