Predictive Powertrain Health Care: Adaptive methodology for monitoring drive components with the help of data science approaches

The development of fuel cell powered propulsion systems is gaining momentum as the world transitions to more sustainable energy sources. However, the complexity of these systems and the lack of knowledge about their behavior under different conditions poses a challenge for maintenance and control strategies. In this research project, an intelligent monitoring and condition estimator for hybrid powertrain systems based on machine learning (ML) approaches will be developed.

In particular, the fuel cell system and the electric motor component are studied in detail to test the capabilities of ML for both component and system level monitoring. As fuel cells have only recently been considered as an alternative to internal combustion engines and their high complexity, the presence and causes of faults can be difficult to detect.

Therefore, unsupervised learning based anomaly detectors, which detect the deviations from the expected behavior, will be investigated for this use case. The project will investigate electric motors with a focus on bearing failures, which are the most common failure mode. Here, mainly supervised approaches will be used to identify critical load profiles and possibly adapt control strategies to specifically prevent damage from bearing currents that could occur due to the trend towards ever higher grid voltages and faster switching inverters. Although the methods used differ, they are used for both supervised and unsupervised learning processes.

The industry has a great interest in bringing drivetrain components such as the electric motor or fuel cell to market quickly. Manufacturers are interested in short development and implementation times. This project offers a solution to a time-critical issue for original equipment manufacturers (OEMs).  The data science methods developed by the research group are scalable and broadly applicable to the powertrain.