ENSEMBLE OF CLASSIFIERS FOR IF STEEL PLATES EVALUATION USING PULSED EDDY CURRENTS
Machine Learning. Pulsed Eddy Current. Pattern Recognition. Ensemble. Digital Signal Processing.
Interstitial Free (IF) steels are used in the manufacture of numerous components with complex shapes. They are used for stamping chassis in the automobile industry due to their high ductility and good mechanical strength. Since it is delivered in coil form, the material is exposed to additional loads of its own weight and may suffer from residual stresses generated. As a consequence, the steel's mechanical properties are susceptible to changes. Since the material is submitted to stamping equipment with pre-established parameters for desired conditions, identifying residual stresses beyond the tolerance range is an essential step in preventing defects in the manufactured parts. In this prospect, the current work proposes a set of intelligent techniques focused on feature extraction and signal classification, with the objective of identifying IF steel samples regarding the modifications of mechanical properties through non-destructive testing using the pulsed eddy current technique (PEC). For this purpose, different computational routines based on Digital Signal Processing techniques and several machine learning algorithms were analyzed: Support Vector Machine, K-nearest-neighbor, Decision Tree, Multilayer Perceptron, LightGBM, Logistic Regression, and Gaussian Naive Bayes. As an additional part of this research, a tensile test was conducted to investigate the behavior of specimens subjected to known stress for residual stress generation and its repercussion on the PEC test. Different ensemble models were evaluated for classification optimization, using the previously validated models as base classifiers. The results obtained indicate a satisfactory performance in the classification of the signals with the base models and show an improvement with the use of an ensemble.