Thermal monitoring of electric machines using fibre-optic sensors with machine learning algorithms

  • Kun SHANG

Student thesis: PhD Thesis


Electric machines (EM) are widely considered as the main energy conversion devices among industry 4.0 era: they are employed in major industry sectors including but not limited to manufacturing, transportation electrification, power generation and industrial machinery. EMs consume ~75% of electricity across all areas of industry. EMs fail mainly because of undetected overheating. Conventional electronic thermal sensors do not operate reliably inside EMs because of strong electromagnetic fields inside that cause interferences with electronic sensors.
In this work, we investigated fibre-optic sensors (FOS) for thermal monitoring of EMs. Fibre Bragg grating (FBG) was evaluated and chosen for the EM thermal monitoring application from various principles of FOS techniques. A novel sensor system combining FBG sensor with machine learning (ML) algorithms was designed and validated progressively to achieve comprehensive thermal monitoring of EMs. The self-constructed lab-version FBG temperature sensor system presented 20-200℃ measurement range with maximum 1kHz sensing frequency, 5cm spatial resolution, and 3.5±1.5℃ error compared with thermocouple measurement. SVR and MLPR machine learning algorithms were proven to be qualified for our EM thermal monitoring task. The FBG-ML combined sensor system presented good regression performance for inner EM temperature prediction with the values of regression evaluation indices as: Mean absolute error (MAE)< 5℃, Root mean squared error (RMSE) 0.9 and Explained variance score (EVS)>0.9. Compared to the conventional thermal sensors and simulation methods for EMs, our FBG-ML sensor prototype can measure the EM thermal distribution in a more accurate, more robust, more convenient manner, and more sensitive to the actual EM working conditions.
The socio-ethically informed standard was developed and published after the construction of sensor prototypes, it made up for the blank of the standard establishment of industrial intelligent monitoring systems (IMS). Additionally, three Chinese invention patents were applied based on our FBG-ML sensor prototype. The utilization of FBG sensor with ML algorithms can improve safety and reliability of electric machines and other power electronic equipment, pre-diagnose equipment failure, reduce maintenance cost, prolong lifetime, and optimize control. The research in this thesis demonstrated both scientific research significance as well as the practical engineering value within Industry 4.0.
Date of AwardMar 2023
Original languageEnglish
Awarding Institution
  • University of Nottingham
SupervisorVladimir Brusic (Supervisor), Yaping Zhang (Supervisor), Michael Galea (Supervisor), Weiduo Zhao (Supervisor) & Shun Bai (Supervisor)

Cite this