Wear detection for a cutting tool based on feature extraction and multivariate regression

In this paper, a method for detecting the wear of the cutting tool in laminate production is proposed. First, principal component analysis (PCA) for dimensionality reduction and clustering are used to determine from the measurement data whether a data set was recorded during production or during idling. Then, using only the data sets from actual production, a model for the wear is trained in a feature-based approach. The most relevant features for detecting wear are selected using a filter feature selection approach. Afterwards, an estimator for the wear is determined from the selected features by multivariate regression. A comparison of the results of two different sensor systems shows, that the sensor data already available for process monitoring can be reused for this purpose and that no additional sensor system needs to be installed.