Session: GT9.1 - Analytics & Digital Solutions and GT6.2 - Emerging Technologies (includes Wind Energy)

Paper Number: 160465

160465 - Fault Diagnosis of Dual Rotor Multi Bearing System Using Transfer Learning 

Rotating machinery, such as steam turbines and turbo generators, plays a critical role in industrial operations and contributes significantly to national economic productivity. Comprising around 40% of industrial equipment. The rotor-bearing system is the core component of the rotating machines and it is vulnerable to damage due to prolonged operation, environmental influences, fluctuating load demands, and maintenance issues. Such damage can lead to productivity losses or even accidents. Therefore, it is of great practical significance to do condition monitoring of the rotor-bearing system in order to identify the fault and take timely action, which shall ensure the safe and stable operation of rotating machineries.  Machine learning techniques have been increasingly applied to condition monitoring, leveraging vibration signal data to diagnose faults and characterize operational states. However, real-world implementation remains limited due to the unavailability of fault data from real operating machines and the variability of feature sets across different systems. A set of features may work well for one machine and may fail for another. This study addresses these challenges using a transfer learning approach to bridge the gap between simulation models and real-world applications. Vibration data representing three fault classes (balance, imbalance, and rub) were generated using a simulated dual-rotor, four-bearing system. Time-domain and frequency-domain features were extracted, and the ReliefF algorithm was employed to identify the most discriminative features. These optimized features, derived from the simulation model (source domain), were transferred to a real-world test rig (target domain) with matching fault classes. A comparative analysis of fault diagnosis accuracy was performed, evaluating models trained on simulation data with and without transfer learning. The results demonstrate the potential of this methodology to enhance diagnostic accuracy and facilitate the practical implementation of data-driven strategies for fault detection in rotor-bearing systems.

Presenting Author: Waquar Ahmed Khan Indian Institute of technology, Hyderabad

Presenting Author Biography: Academics:
Schooling - Vishwadeep Senior Secondary School, Durg (1995 - 2007)
Graduation - BE(Mechanical Engineering) from Bhilai Institute of Technology, Durg (2007 - 2011)
Post graduation - M.tech(Machine Design Engineering) from Indian Institute of Technology, Roorkee (2011 - 2013)
External PhD (Pursuing) - Indian Institute of Technology, Hyderabad (2021 - present)

Professional Experience:
BHEL, Trichy (2013-2018) in R&D department and was associated with national mission project title "Advanced ultra super critical project".
BHEL, Corporate R&D, Hyderabad (2018 - present) in Machine dynamic and failure analysis lab, work in the domain of vibration & acoustics, rotor dynamics. presently designated as Deputy manager and handling Defence projects.
10 copyrights, 2 design registrations