From Anomaly Detection to Predictive Fault Forecasting for Charging Infrastructure

Institut

Lehrstuhl für Fahrzeugtechnik (TUM-ED)

Typ

Semesterarbeit / Masterarbeit /

Inhalt

 

Beschreibung

Description

The growth of electric mobility increases the demands on the availability and reliability of charging infrastructure. In practice, charging sessions often fail not because of the power electronics themselves, but due to disruptions in the communication between charge point, backend, and vehicle. Although the OCPP protocol standardizes the communication between charging station and backend, the way states and faults are reported differs considerably between protocol versions. For diagnostics and operation, this results in a heterogeneous, manufacturer-dependent data space that can hardly be handled manually. In the KI-LOAD research project at the Chair of Automotive Technology, real operational and charging data are used to systematically capture fault patterns and analyze them using machine learning methods.

Objective of the thesis:

The aim of this thesis is to develop an end-to-end data pipeline that ranges from the automatic detection of anomalous charging sessions (anomaly detection) to the predictive forecasting of faults before they lead to failures. The goal is to classify fault patterns in a data-driven way, identify their causes, and develop and evaluate predictive models for the predictive maintenance of the charging infrastructure.

The thesis (seminar or master's thesis) comprises the following work packages:

• Literature research and state of the art on fault analysis, anomaly detection, and predictive maintenance for charging infrastructure
• Preparation and structuring of real charging data as well as OCPP event and fault data from field operation
• Development of a standardized fault taxonomy and classification of typical fault patterns based on charging profile, error codes, and history
• Anomaly detection of conspicuous charging sessions using statistical and learning-based methods, such as clustering and unsupervised approaches
• Development and evaluation of predictive models for fault and failure forecasting in the sense of predictive maintenance
• Definition of suitable evaluation metrics and evaluation of the models against the project targets

The thesis can be written in either German or English.

Voraussetzungen

• Programming skills: Advanced proficiency in Python and time series analysis
• Machine learning foundations: Solid theoretical and practical understanding of neural networks and clustering algorithms
• Independent and structured way of working
• Very good German or English language skills

Tags

FTM Studienarbeit, FTM EV, FTM PBergmann

Möglicher Beginn

sofort

Kontakt

Philipp Bergmann, M.Sc.
Raum: MW 3511
Tel.: +49.89.289.10342
p.bergmanntum.de

Ausschreibung