Fleet-based health monitoring for end-of-production-line and operational testing

Applications of machine learning for end-of-production-line and operational health monitoring avoid time-consuming manual data analysis and design of handcrafted features. However, many of these approaches require a large historical data set and analyze only a single unit which is a machine or a component. In practice, this data set can be difficult and expensive to collect. However, some industrial applications involve a fleet of similar operating units. It is often safe to assume that the condition of the majority of the units in a fleet is healthy. In this work, we demonstrate a fleet-based, unsupervised, and generic anomaly detection framework that detects unit faults by identifying deviating behavior. It allows incorporating domain expertise by user-defined comparison measures. Moreover, contrary to most black-box artificial intelligence techniques, easy interpretability allows a domain expert to validate the predictions made by the framework. Finally, we demonstrate its applicability in both end-of-production-line and operational testing.