物理データ駆動型機械学習モデル:IPMCソフト人工筋肉の非線形・動的・開ループ同定へのハイブリッド手法
Physics-Data Driven Machine Learning Based Model: A Hybrid Way for Nonlinear, Dynamic, and Open-loop Identification of IPMC Soft Artificial Muscles
IPMCアクチュエータの非線形・ヒステリシス挙動を、解析モデルと深層ニューラルネットを組み合わせた非自己回帰型ハイブリッドモデルで同定し、リアルタイム応用可能な高精度モデルを実現した。
著者: Mohsen Annabestani, Mohammad Hossein Sayyad, Zahra Meskar, Mehdi Fardmanesh, Barbara Mazzolai
分類: cs.LG, cs.RO
原文アブストラクト
Ionic Polymer Metal Composites (IPMCs) are one of the most preferred choices among biocompatible materials for industrial and biomedical applications. Despite their advantages, some of their drawbacks include non-linear and hysteretic behavior, which complicates the modeling process. In previous works, usually autoregressive models were used to predict the behavior of an IPMC actuator. The main drawback of using an autoregressive model is that it cannot be used in mobile and real-time applications. In this study, we proposed a hybrid analytical intelligent model for an IPMC actuator. The most outstanding feature of this model is its non-autoregressive structure. The hybrid concept proposed in this study can be generalized to various problems other than IPMCs. The structure used in this work comprises an analytical model and a deep neural network, providing a non-linear, dynamic, and non-autoregressive model for the IPMC actuator. Lastly, the average NMSE achieved using the proposed hybrid model is 9.5781e-04 showing a significant drop in the error rate compared to other non-autoregressive structures.
関連論文
- 拮抗型空気圧人工筋のダイナミクスを学習する物理埋め込みNeural ODEソフトアクチュエータ/モデリング
