Towards a Hierarchical Approach for Outlier Detection in Industrial Production Settings

In the context of Industry 4.0, the degree of cross-linking between machines, sensors, and production lines increases rapidly.However, this trend also offers the potential for the improve-ment of outlier scores, especially by combining outlier detectioninformation between different production levels. The latter, in turn, offer various other useful aspects like different time series resolutions or context variables. When utilizing these aspects, valuable outlier information can be extracted, which can be then used for condition-based monitoring, alert management, or predictive maintenance. In this work, we compare different types of outlier detection methods and scores in the light of the aforementioned production levels with the goal to develop a modelfor outlier detection that incorporates these production levels.The proposed model, in turn, is basically inspired by a use casefrom the field of additive manufacturing, which is also known asindustrial 3D-printing. Altogether, our model shall improve the detection of outliers by the use of a hierarchical structure that utilizes production levels in industrial scenarios.