Steel type determination by spark test image processing with machine learning

Abstract The spark test method is a simple and low-cost method in which an operator with special skills observes the sparks emitted by a grinding wheel in contact with the steel in order to identify the material of the sample. However, the operator might classify erroneously two different steel materials with close spark characteristics. Therefore, we propose a method that extracts features from images captured from the spark test method and uses these features as input on machine learning models. The regression models predicted the carbon content of steel with 8% error while the classifiers had 82% of accuracy. The classifiers models had good results with few confusion points and regression models had low error. Regarding the confusion points, the regression algorithms could solve the misclassification by predicting the carbon content of the sample and increasing accuracy. The proposed method is suitable for real-time and shop floor applications.