Scholarlink

Purpose <p>To address the limitations of traditional physical models (parameter sensitivity; nonlinear disturbances) in accurately predicting aircraft landing gear oscillation dynamics for flight safety, this paper proposes the dual-objective integration framework DFNN-SINDy, which aims to integrate the respective advantages of SINDy (interpretable dynamic equation extraction) and DFNN (high-precision prediction) to provide a comprehensive solution for analyzing landing gear oscillation.</p> Method <p>Based on Takens' embedding theorem, single-dimensional vibration time series are reconstructed into a high-dimensional phase space. A Dynamic Feedforward Neural Network (DFNN) captures temporal dependencies within this space. Simultaneously, Sparse Identification of Nonlinear Dynamics (SINDy) constructs a basis function library (linear/nonlinear terms) and uses L1-regularized sparse regression to extract interpretable dynamic equations.</p> Results <p>For dynamic oscillation prediction, DFNN achieved ultra-low errors (1 × 10⁻<sup>5</sup> to 4 × 10⁻⁶). Under Gaussian noise (σ = 0.1), predicted curves closely tracked real signals. On experimental data (50–150&#xa0;km/h), DFNN error remained stable (2 × 10⁻<sup>3</sup> to 1.5 × 10⁻<sup>2</sup>), showing strong robustness. SINDy successfully identified key parameters (e.g., strut stiffness, error &lt; 0.1%; axle damping) in multi-dimensional simulations, controlling prediction errors within ± 2 × 10⁻<sup>3</sup>. However, SINDy failed on single-dimensional experimental data due to incomplete state space.</p> Conclusion <p>The DFNN-SINDy framework provides a high-precision tool for real-time landing gear oscillation monitoring via DFNN (response time &lt; 0.1&#xa0;s). It also guides structural parameter optimization (e.g., critical damping ratio design) through the interpretable equations output by SINDy.</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

Landing Gear Oscillation Prediction Based on DFNN and SINDY

  • 期刊论文

  • 发表日期：

    2025年10月28日

查看全文
  • Jian Wei,
  • Yangjing Zhang,
  • Jing Tian,
  • Shidong Ma,
  • Zhe Zhao

摘要

Purpose

To address the limitations of traditional physical models (parameter sensitivity; nonlinear disturbances) in accurately predicting aircraft landing gear oscillation dynamics for flight safety, this paper proposes the dual-objective integration framework DFNN-SINDy, which aims to integrate the respective advantages of SINDy (interpretable dynamic equation extraction) and DFNN (high-precision prediction) to provide a comprehensive solution for analyzing landing gear oscillation.

Method

Based on Takens' embedding theorem, single-dimensional vibration time series are reconstructed into a high-dimensional phase space. A Dynamic Feedforward Neural Network (DFNN) captures temporal dependencies within this space. Simultaneously, Sparse Identification of Nonlinear Dynamics (SINDy) constructs a basis function library (linear/nonlinear terms) and uses L1-regularized sparse regression to extract interpretable dynamic equations.

Results

For dynamic oscillation prediction, DFNN achieved ultra-low errors (1 × 10⁻5 to 4 × 10⁻⁶). Under Gaussian noise (σ = 0.1), predicted curves closely tracked real signals. On experimental data (50–150 km/h), DFNN error remained stable (2 × 10⁻3 to 1.5 × 10⁻2), showing strong robustness. SINDy successfully identified key parameters (e.g., strut stiffness, error < 0.1%; axle damping) in multi-dimensional simulations, controlling prediction errors within ± 2 × 10⁻3. However, SINDy failed on single-dimensional experimental data due to incomplete state space.

Conclusion

The DFNN-SINDy framework provides a high-precision tool for real-time landing gear oscillation monitoring via DFNN (response time < 0.1 s). It also guides structural parameter optimization (e.g., critical damping ratio design) through the interpretable equations output by SINDy.