Description
As part of your master's thesis, you will further develop and evaluate approaches for automated error detection in 1D simulation results (time series). The question is whether detection is also possible with dynamic load cases and vehicles of various drive topologies.
In a previous work, a method was developed that uses an autoencoder to reliably detect anomalies in electric vehicles during acceleration maneuvers. The methods developed should be extended to more dynamic load cases, such as the WLTP or circuit route maneuvers, as well as other vehicle topologies. In addition, the methods should be integrated into a data pipeline to enable an ongoing process for updating the baseline behavior.
The work includes literature research on the state of the art and the extension of existing algorithms. Finally, the feasibility will be checked in a proof of concept.
Responsibilities:
- Conduct literature research on the state of the art (unsupervised machine learning, clustering, anomaly detection, autoencoder)
- Review existing methods
- Extend existing methods
- Develop a comprehensive validation concept
- Confirm feasibility within the scope of a proof of concept
- Document results
Requirements:
- Studies in computer science or similar, technical studies with a focus on software development and machine learning
- Study focuses: ML, AI, data analytics
- Technical skills: TensorFlow, PyTorch, RayTune or similar
- IT skills: Python, MATLAB (desirable)
- Language skills: German (B2) and English (B2)
- Soft skills: Independent and structured working style, team and communication ability