日本フィジカルAI新聞

世界のフィジカルAIを、日本語で。

形状記憶合金/モデリングarXiv:2103.03586

形状記憶合金ワイヤ駆動構造のためのハイブリッド動的モデリング枠組み

A Hybrid Dynamical Modeling Framework for Shape Memory Alloy Wire Actuated Structures

シェア:XThreadsFacebookLINEはてブBluesky

形状記憶合金ワイヤアクチュエータのMASモデルを5つの動作モードに分割してハイブリッド力学系として再定式化し、精度を保ちつつシミュレーション時間を最大80%削減した。

著者: Michele A. Mandolino, Francesco Ferrante, Gianluca Rizzello

分類: cs.RO

原文アブストラクト

In this paper, a hybrid model for single-crystal Shape Memory Alloy (SMA) wire actuators is presented. The result is based on a mathematical reformulation of the M\"uller-Achenbach-Seelecke (MAS) model, which provides an accurate and interconnection-oriented description of the SMA hysteretic response. The strong nonlinearity and high numerical stiffness of the MAS model, however, hinder its practical use for simulation and control of complex SMA-driven systems. The main idea behind the hybrid reformulation is based on dividing the mechanical hysteresis of the SMA into five operating modes, each one representing a different physical state of the material. By properly deriving the switching conditions among those modes in a physically-consistent way, the MAS model is effectively reformulated within a hybrid dynamical setting. The main advantage of the hybrid reformulation is the possibility of describing the material dynamics with a simplified set of state equations while maintaining all benefits of the physics-based description offered by the MAS model After describing the novel approach, simulation studies are conducted on a flexible robotic module actuated by protagonist-antagonist SMA wires. Through comparative numerical analysis, it is shown how the hybrid model provides the same accuracy as the MAS model while saving up to 80% of the simulation time. Moreover, the new modeling framework opens up the possibility of addressing SMA control from a hybrid systems perspective.

  1. 1Boston Dynamics、車輪付きAtlasの研究試作機を紹介研究・技術
  2. 2Hugging Face、ロボット学習基盤「LeRobot v0.7.0」を公開 対話操作と大規模データ学習に対応新製品・発表
  3. 3AGIBOTのG2 6台、上海のコンビニで24時間試験 1,730タスクを実施量産・展開
  4. 4UC BerkeleyとNVIDIAの研究者、AIエージェントでロボット制御コードを仮想環境で改善研究・技術
  5. 5NVIDIA、CoRLでGR00T 2とIsaac Lab 3.0を発表新製品・発表
PR本紙発行元 EmplifAI