Bachelor/Master Thesis

Rotating machinery such as bearings, gearboxes and roller chains is essential for industrial power transmission. Unexpected component degradation generates abnormal signals, leading to costly downtime and safety risks. Conventional fault diagnosis approaches rely on handcrafted features and shallow machine learning, which face challenges including limited labeled data, cross-condition distribution shifts and weak interpretability under fluctuating speeds and loads. Benefiting from strong multimodal reasoning and knowledge generalization ability, large language model (LLM) enables end-to-end fault identification by converting raw monitoring signals into structured sequential representations. This thesis develops a efficient LLM-based fault diagnosis framework tailored for rotating machinery, comprehensively validated using experimental data under various operating conditions and across different mechanical components.

Bachelor/Master Thesis

LLM-Based Fault Detection and Diagnosis for Rotating Machinery

Tasks

  • Review state-of-the-art LLM applications, signal processing, and cross-domain fault diagnosis literature to identify research gaps
  • Design a signal-to-text conversion pipeline to transform raw (time-series/image) data into sequential embeddings compatible with pre-trained LLMs
  • Implement parameter-efficient fine-tuning (LoRA/QLoRA) strategies to adapt general LLMs for machinery fault classification under limited labeled samples
  • Evaluate accuracy, cross-condition generalization and anti-noise robustness of the proposed LLM framework via comparative tests against conventional  methods

Start Ab sofort

Software Matlab, Python

Requirements

  • Solid foundation in mathematics and data analytics
  • Ability to work independently
  • Enjoyment of programming tasks and strong motivation to learn

Beneficial GPU, HPC, Pytorch, Machine learning

Further Information Dr. Junyu Qi (junyu qi∂kit edu)

Officer Hours Montag/Freitag 10:00–16:00 Uhr, Raum 706, Gebäude 10.23

Literatures:

[1] X. Chen, Y. Lei, Y. Li, S. Parkinson, X. Li, J. Liu, F. Lu, H. Wang, Z. Wang, B. Yang, S. Ye: Large models for machine monitoring and fault diagnostics: Opportunities, challenges, and future direction. Journal of Dynamics, Monitoring and Diagnostics. 2025 Jun 21;4(2):76-90.

[2] D. Li, Z. Pang, K. Yang, Y. Luo and Y. Zeng: FD-MLLM: Fault Diagnosis Framework Based on Multimodal Data and Large Language Model. 2025 IEEE 23rd International Conference on Industrial Informatics (INDIN), Kunming, China, 2025, pp. 1-7

[3] L. Tao, H. Liu, G. Ning, W. Cao, B. Huang, C. Lu: LLM-based framework for bearing fault diagnosis,

Mechanical Systems and Signal Processing, Volume 224, 112127, 2025

[4] Y. Zeng , H. Wang, G. Ran, Z. Dong, X. Li, X. Li, J. He: Signal LLM: Fault perception and maintenance system for dual-drive hydraulic turbine bearings based on cross-attention-LLM. Structural Health Monitoring. 2026.

[5] X. Lee, L. Vidyaratne, A. Farahat, C. Gupta: Exploring LLM-based Frameworks for Fault Diagnosis. Computer Science: Artificial Intelligence, 51, 2025.