Guardians of the Fuel Cells: An Automated Monitoring Framework with Digital Twins

The challenging durability targets for fuel cells, reaching up to 25,000 h, require precise monitoring to ensure slow aging and to prevent damage. As open systems, fuel cells are susceptible not only to internal faults but also to harmful external influences, such as contamination. Since cell voltage is a sensitive indicator for the fuel cell's state of health, we have presented a digital twin-based monitoring approach to detect anomalous events based on cell voltage [1].To automate this approach, we present a novel web-based framework for automated fuel cell monitoring. This framework builds on the developed fuel cell digital twin models and enables the automated application of these models. A web-based graphical user interface allows the user to create and train a digital twin for a new fuel cell. After training the model, the framework uses this digital twin to monitor new measurement data in real time.As shown in Figure 1, the lifecycle of a digital twin in the framework consists of three steps. First, the framework uses a pre-trained digital twin and trains it on the specifics of each new fuel cell using transfer learning. Since a pre-trained model is used, only a few data points (< 100) are required. Second, the transferred model is used to check new measurement data in real time. If anomalies are detected, a notification is sent to the user via an email notification service. After a defined period of time (e.g. every second week), the digital twin is retrained using past data. This allows the model's predictive accuracy to improve over time. The retrained model is then used for data monitoring.