Scholarlink

<p>Shape memory alloys (SMAs) show exceptional potential in actuator design due to their shape memory effect and superelasticity, yet their thermoelectric hysteresis challenges accurate modeling. This study proposes a hybrid framework integrating long short-term memory (LSTM) networks with physical kinematics to predict SMA actuator responses. Unlike conventional approaches, our method decouples material behavior prediction from actuator geometry: A single-layer LSTM network processes voltage-time sequences to predict SMA wire’s temperature and resistance dynamics, while a physics-based model computes angular displacement through phase transformation and constitutive equations. Trained on 10 experimental conditions (1–2&#xa0;°C/s heating rates) and tested on 3 unseen cases (0.8&#xa0;°C/s), the model achieves a mean absolute error of &lt; 5% in angular displacement prediction, with root mean square errors of 2.5 × 10<sup>−5</sup> for temperature/resistance outputs. The modular architecture eliminates neural network retraining during structural iterations—only ordinary differential equation parameters require adjustment. This approach advances rapid actuator optimization, demonstrating faster computational efficiency compared to full-physics models. Our findings establish a computationally sustainable paradigm for smart material-based actuation systems, extensible to piezoelectric and magnetostrictive materials through constitutive parameter substitution.</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

An LSTM-driven thermoelectric coupling response prediction method for shape memory alloy actuators

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年5月21日

查看全文
  • Ding Shaozhe,
  • Liu Longbin,
  • Zhang Shifeng,
  • Li Mingkun

摘要

Shape memory alloys (SMAs) show exceptional potential in actuator design due to their shape memory effect and superelasticity, yet their thermoelectric hysteresis challenges accurate modeling. This study proposes a hybrid framework integrating long short-term memory (LSTM) networks with physical kinematics to predict SMA actuator responses. Unlike conventional approaches, our method decouples material behavior prediction from actuator geometry: A single-layer LSTM network processes voltage-time sequences to predict SMA wire’s temperature and resistance dynamics, while a physics-based model computes angular displacement through phase transformation and constitutive equations. Trained on 10 experimental conditions (1–2 °C/s heating rates) and tested on 3 unseen cases (0.8 °C/s), the model achieves a mean absolute error of < 5% in angular displacement prediction, with root mean square errors of 2.5 × 10−5 for temperature/resistance outputs. The modular architecture eliminates neural network retraining during structural iterations—only ordinary differential equation parameters require adjustment. This approach advances rapid actuator optimization, demonstrating faster computational efficiency compared to full-physics models. Our findings establish a computationally sustainable paradigm for smart material-based actuation systems, extensible to piezoelectric and magnetostrictive materials through constitutive parameter substitution.