Scholarlink

<p>In order to solve the nonlinear and coupling problems of the three-degree-of-freedom (3-DOF) hybrid magnetic bearing (HMB) system with multiple inputs and outputs, an improved online learning Levenberg–Marquardt neural network (LMNN) inverse system with active disturbance rejection control (ADRC) method is proposed. Firstly, the structure and working principle of the six-pole radial-axial HMB are introduced, and a mathematical model of suspension force is established. Secondly, the ADRC is used to solve the problem of online learning parameter acquisition in LMNN, and optimize the error between the composite pseudo-linear system and the ideal system. Online learning LMNN is used to approximate the inverse system of HMB, and a composite pseudo-linear system is formed by serial connection with the original system to achieve linearization and decoupling of the suspension forces. Thirdly, simulations of rotor fluctuation and anti-interference are conducted. The results demonstrate that the proposed method exhibits superior anti-interference capability and 3-DOF decoupling efficiency compared to static BPNN, with about 28.26% reduction in floating axial overshoot. Finally, an experimental platform is constructed to conduct experiments, which validate the performance of the proposed decoupled control system.</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

Decoupling control of neural network inverse system based on improved online learning Levenberg–Marquardt for 3-DOF hybrid magnetic bearing

  • 期刊论文

  • 发表日期：

    2025年11月24日

查看全文
  • Jing Lu,
  • Yutong Yan,
  • Jintao Ju,
  • Huangqiu Zhu

摘要

In order to solve the nonlinear and coupling problems of the three-degree-of-freedom (3-DOF) hybrid magnetic bearing (HMB) system with multiple inputs and outputs, an improved online learning Levenberg–Marquardt neural network (LMNN) inverse system with active disturbance rejection control (ADRC) method is proposed. Firstly, the structure and working principle of the six-pole radial-axial HMB are introduced, and a mathematical model of suspension force is established. Secondly, the ADRC is used to solve the problem of online learning parameter acquisition in LMNN, and optimize the error between the composite pseudo-linear system and the ideal system. Online learning LMNN is used to approximate the inverse system of HMB, and a composite pseudo-linear system is formed by serial connection with the original system to achieve linearization and decoupling of the suspension forces. Thirdly, simulations of rotor fluctuation and anti-interference are conducted. The results demonstrate that the proposed method exhibits superior anti-interference capability and 3-DOF decoupling efficiency compared to static BPNN, with about 28.26% reduction in floating axial overshoot. Finally, an experimental platform is constructed to conduct experiments, which validate the performance of the proposed decoupled control system.