Under CopyrightEnge-Rosenblatt, OlafOlafEnge-RosenblattBayer, ChristianChristianBayerBrüning, AndreasAndreasBrüning2022-03-134.8.20162016https://publica.fraunhofer.de/handle/publica/39268710.24406/publica-fhg-392687This contribution presents the idea of using different sources of data in order to effectively monitor and control the manufacturing process. For application in classical Condition Monitoring we developed a data analysis flow which is equipped with self-learning properties. In this flow we use all available sensor data to extract relevant features (in the sense of distinguishing marks) in a generic way. To this end, a broad variety of mathematical transformations followed by statistical value calculations are applied to all sensor data histories yielding a large amount of potential features. Using Principal Component Analysis or Linear Discriminant Analysis, we are able to extract the relevant features which reliably build clusters of measurement points in the feature space. These clusters are finally used to distinguish between different operational conditions and wear-out conditions. This way, starting with a finger print of the machine free from defects, we are able to detect variations of the monitored machine concerning e.g. slightly increasing wear-out phenomena. Our approach was already successfully applied to hydraulic pumps, electric drive trains e.g. for conveyor belts, and also packaging machines.en621004poster