Dynamic unbalance identification in steady-state rotating machinery: A hybrid methodology integrating physical and data-driven techniques

• Published in: Journal of sound and vibration Vol. 597; p. 118817
• Main Authors: Michalski, Miguel Angelo de Carvalho, de Melo, Italo Skovroski, de Souza, Gilberto Francisco Martha
• Format: Journal Article
• Language: English
• Published: Elsevier Ltd 17.02.2025
• Subjects: Fault identification; FEM; Finite elements modeling; Kalman filter; PCA; Principal component analysis; Rotordynamics; Fault identification; Kalman filter; Rotordynamics; Finite elements modeling; Principal component analysis; PCA; FEM
• ISSN: 0022-460X
• DOI: 10.1016/j.jsv.2024.118817

•Real-time monitoring of unbalance progression in rotating machinery.•Eliminates the need for phase angle measurements during operation.•Enables continuous, accurate monitoring during steady-state conditions.•Combines FEM modeling, Kalman Filter, and advanced PCA for fault detection.•Case studies validate accuracy using synthetic and real industrial data. Rotating machinery plays a strategic role in key industrial sectors, making its analysis a subject of great interest for both academia and industry. Effective maintenance planning for this equipment is essential for asset management and for meeting current industrial requirements. To address this demand, this work presents a novel unbalance identification approach based on a digital representation of a rotating machine's dynamic behavior in relation to the development of specific faults. A hybrid methodology is proposed, integrating Finite Element Modeling, a Kalman Filter for parameter estimation, and Referenced Moving Window Principal Component Analysis. This Principal Component Analysis extension enhances vibration pattern recognition, enabling accurate fault quantification directly from slight changes in the system's steady-state behavior. The methodology uniquely eliminates the need for phase angle measurements, facilitating continuous monitoring of unbalance progression in steady-state conditions. Two case studies demonstrate the methodology's potential: one utilizing synthetic data from a Floating Production Storage and Offloading centrifugal compressor unit, and the other based on real data from a hydroelectric turbine-generator. These studies illustrate the integration of computational modeling, data-driven analysis, and monitored vibration data, achieving robust and accurate unbalance identification. This approach provides valuable insights into current capabilities and opens promising pathways for future applications, particularly in the digital twin domain for rotating machinery.