Scholarlink

<p>Accurate identification of governing equations is both challenging and essential for reliable prediction of rotor system dynamics. This study proposes a Physics-Informed Robust Sparse Identification Framework (PIRSIF), which integrates variance normalization, adaptive equation-specific thresholding, and structural consistency constraints to address the inherent multi-scale characteristics of rotor systems. The performance of PIRSIF is systematically evaluated using a four-degree-of-freedom (4-DOF) Jeffcott rotor benchmark encompassing both linear and nonlinear behaviors. Extensive numerical studies are conducted under varying noise levels and noise types, excitation amplitudes, nonlinear stiffness coefficients, and rotational speeds. Results show that PIRSIF reliably identifies key parameters—such as stiffness, damping, nonlinear stiffness, and unbalance force—with relative errors generally below 3%. Compared with the standard SINDy method, the proposed framework achieves a significant precision improvement for weak-signal dynamics while maintaining superior structural fidelity by effectively suppressing spurious numerical terms. With <i>R</i><sup>2</sup> values consistently exceeding 0.99, PIRSIF provides a robust and physically consistent data-driven tool for the digital-twin modeling and intelligent health monitoring of rotating machinery.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Data-driven modeling of rotor systems using a physics-informed sparse identification method

  • 期刊论文

  • 发表日期：

    2026年4月20日

查看全文
  • Hanghang Chen,
  • Hulun Guo

摘要

Accurate identification of governing equations is both challenging and essential for reliable prediction of rotor system dynamics. This study proposes a Physics-Informed Robust Sparse Identification Framework (PIRSIF), which integrates variance normalization, adaptive equation-specific thresholding, and structural consistency constraints to address the inherent multi-scale characteristics of rotor systems. The performance of PIRSIF is systematically evaluated using a four-degree-of-freedom (4-DOF) Jeffcott rotor benchmark encompassing both linear and nonlinear behaviors. Extensive numerical studies are conducted under varying noise levels and noise types, excitation amplitudes, nonlinear stiffness coefficients, and rotational speeds. Results show that PIRSIF reliably identifies key parameters—such as stiffness, damping, nonlinear stiffness, and unbalance force—with relative errors generally below 3%. Compared with the standard SINDy method, the proposed framework achieves a significant precision improvement for weak-signal dynamics while maintaining superior structural fidelity by effectively suppressing spurious numerical terms. With R2 values consistently exceeding 0.99, PIRSIF provides a robust and physically consistent data-driven tool for the digital-twin modeling and intelligent health monitoring of rotating machinery.