Numerical modelling in assisting machine learning for the enhanced condition monitoring of rotors

Anil Kumar, Jianlong Wang,
Numerical modelling in assisting machine learning for the enhanced condition monitoring of rotors,
Results in Engineering,
Volume 28,
2025,
107250,
ISSN 2590-1230,
https://doi.org/10.1016/j.rineng.2025.107250.
(https://www.sciencedirect.com/science/article/pii/S2590123025033055)
Abstract: The rotor is a vital component in many mechanical systems, and its defects can cause catastrophic failures if not detected early. This study introduces a novel machine learning (ML) approach assisted by numerical modelling for rotor defect detection, addressing the challenges faced by experimental investigations due to operational restrictions. System identification is used to estimate key modal parameters such as mass, stiffness, damping, and eccentricity. To assess robustness, Gaussian noise was added to displacement signals at 10 noise levels (1 %–50 %), with 10 trials each, showing that the identified parameters remain stable despite noise. Synthetic fault data are created by introducing defects into the model, allowing the simulation of various fault scenarios. The model’s accuracy is validated by comparing the mean squared error (MSE) between simulation and experimental results. A Support vector machine (SVM), optimized using the Optuna framework, was trained on simulated data. Once established, the model is tested on unseen real-world data to evaluate its fault detection performance, which proves to be accurate. The main novelty of this study is the development of a hybrid diagnostic framework that combines numerical modeling with machine learning for rotor fault detection. The key achievement of this work is that the diagnostic model, trained solely on simulation data, which can accurately identify defects under real-world condition. A comparative analysis demonstrates that the proposed ML models outperform others, with SVM achieving the highest mean accuracy of 99.85 % for Set 1 and 100 % for Set 2.
Keywords: Machine learning; Numerical model; Rotor defects; System identification