Thesis: LLM-based Automated Root-Cause Analysis for Unexpected System Behavior

Aufgaben:

We are seeking a motivated student for a master thesis in LLM-based Automated Root-Cause Analysis for the automotive industry. You will develop methods to analyze extensive textual data from multiple sources. You will work on prompt engineering and fine-tuning of large language models, in collaboration with industry experts. The objective of the thesis is to learn the dependency between observations in the time-series data and the correlations with the physical system, to finally rank the potential failure modes, and to cluster similar failure conditions. The thesis will offer valuable insights into addressing root causes of issues of powertrain systems. YOUR RESPONSIBILITIES: Gather from various sources (e.g., FP-sheets, technical reports, presentations, word documents) to create a knowledge database. Define the database structure for efficient processing of textual data. Apply simple analytics to detect deviations from the expected behavior or use existing analysis results. Assign appropriate textual descriptions to the observations (e.g., very high battery cell temperature, low pack voltage, normal contactor temperature, etc.). Apply a wide range of techniques (e.g., few-shot classification, RAGs, Chain-Of-Thoughts) from the field of LLMs to learn the dependency between observations and the knowledge database content. Provide a ranking of the top potential failure modes along with their probabilities for a specific combination of observed conditions. Cluster similar observed conditions related to different failure modes in the knowledge database. Define a validation approach with the appropriate metrics from the literature and validate the outcome of the analysis.

Qualifikation:

BSc in domains similar to: Applied Statistics/Mathematics, Computer Science, Data Science, Automotive or Electrical Engineering Hands-on experience with data analysis, machine learning, natural language processing and database management. Knowledge or experience with open source LLMs, as well as with public APIs and open-source libraries (e.g., Hugging Face) for working with LLMs. Strong proficiency in Python and application of data analysis methods and algorithms. Ability to work independently, conduct experiments, and analyze complex data sets. Strong problem-solving and critical-thinking skills. Strong communication skills to present findings and recommendations effectively.

Wir bieten:

You can write your thesis independently and receive professional guidance and support from our experienced employees. You will have the opportunity to exchange ideas with experts in the company and benefit from their expertise. Take the opportunity to immerse yourself in the world of AVL and embed your theoretical knowledge in a practical environment. The successful completion of the thesis is remunerated with a one-time fee of EUR EUR3,500.00 before tax.

Über Uns:

AVL is one of the world's leading mobility technology companies for development, simulation and testing in the automotive industry, and beyond. We provide concepts, solutions and methodologies in fields like vehicle development and integration, e-mobility, automated and connected mobility (ADAS/AD), and software for a greener, safer, better world of mobility.