Information Extraction from Time Series in the EDM Drilling Process

Electrical discharge machining (EDM) allows to obtain small holes with the high efficiency and high quality. Such features are most common in jet engine turbine airfoils. The main problem of the analysis is detection of a moment when the machine should stop the drilling process—the breakthrough detection. Machine learning applications requires that data and models to be prepared by specialists that can extract the most important information from an input data and choose most suitable Artificial Intelligence (AI) algorithm for particular case. This Article describes an experiment on how to extract valuable information from heterogeneous time series data with various sources (which is popular in an industry 4.0—Internet of Things) from the EDM drilling process. EDM process is conducted by a Computerized Numerical Control (CNC) drilling device that measures the position of the electrode, drilling speed, Additional sensor, monitors impulses current and voltage. The research is focused on classic AI algorithms (decision tree, random forest and eXtreme Gradient Boosting (XGBoost)) which provide fast training and possibility to check more hyperparameters in a time than neural networks algorithms. It is also described how feature extraction can high up AI algorithm predictions.