Scholarlink

<p>Magnetorheological (MR) dampers adjust damping in real time using MR fluid and a magnetic field, thereby enhancing suspension and vibration control. A Long Short-Term Memory (LSTM) neural network captures the nonlinear behaviour of MR dampers. The LSTM model trains and tests on a Lord RD 8041-1 MR damper, predicting the damping force across varying current inputs from 0 to 2 A&#xa0;in 0.5&#xa0;A increments and different excitation frequencies. The force–displacement loops confirm the nonlinear hysteretic behaviour of MR dampers, where higher currents (0–2&#xa0;A) widen the loops, enhance the damping force, and increase energy dissipation for improved vibration control. The model achieves high prediction accuracy, with R<sup>2</sup> values exceeding 0.996, evaluated on the test set following an 80:20 train-test split. K-fold cross-validation further confirmed consistent generalisation performance across data partitions. Although prediction errors increase at higher currents, with MSE rising from 100.47 at 0&#xa0;A to 3017.99 at 2&#xa0;A and RMSE from 10.80 to 65.49, this trend corresponds with the increasing nonlinearity in damper behaviour under elevated magnetic fields. K-fold cross-validation confirms low variance, and stable loss curves validate model convergence. Minimal systematic deviations in residuals further verify the LSTM’s real-time MR damper control reliability. This work demonstrates the effectiveness of LSTM networks in capturing complex damper dynamics, offering a reliable alternative to traditional models. The approach holds strong potential for real-time implementation in intelligent suspension and vibration control systems.</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

Prediction of Damping Force in Magnetorheological Dampers Using Long Short-Term Memory (LSTM) Neural Networks

  • 期刊论文

  • 发表日期：

    2025年6月13日

查看全文
  • Aadil Arshad Ferhath,
  • Kamalakkannan Kasi

摘要

Magnetorheological (MR) dampers adjust damping in real time using MR fluid and a magnetic field, thereby enhancing suspension and vibration control. A Long Short-Term Memory (LSTM) neural network captures the nonlinear behaviour of MR dampers. The LSTM model trains and tests on a Lord RD 8041-1 MR damper, predicting the damping force across varying current inputs from 0 to 2 A in 0.5 A increments and different excitation frequencies. The force–displacement loops confirm the nonlinear hysteretic behaviour of MR dampers, where higher currents (0–2 A) widen the loops, enhance the damping force, and increase energy dissipation for improved vibration control. The model achieves high prediction accuracy, with R2 values exceeding 0.996, evaluated on the test set following an 80:20 train-test split. K-fold cross-validation further confirmed consistent generalisation performance across data partitions. Although prediction errors increase at higher currents, with MSE rising from 100.47 at 0 A to 3017.99 at 2 A and RMSE from 10.80 to 65.49, this trend corresponds with the increasing nonlinearity in damper behaviour under elevated magnetic fields. K-fold cross-validation confirms low variance, and stable loss curves validate model convergence. Minimal systematic deviations in residuals further verify the LSTM’s real-time MR damper control reliability. This work demonstrates the effectiveness of LSTM networks in capturing complex damper dynamics, offering a reliable alternative to traditional models. The approach holds strong potential for real-time implementation in intelligent suspension and vibration control systems.