Nonlinear hysteretic parameter identification using an attention-based long short-term memory network and principal component analysis

Zhenghao Ding, Yang Yu, Yong Xia

Research output: Journal article publicationJournal articleAcademic researchpeer-review

Abstract

Hysteretic models are used to describe the nonlinear memory-based relationship between the input and output of some physical systems. A long short-term memory neural network-based method is proposed to identify nonlinear hysteretic parameters. Either force or vibration response data are used as the input of the network and the nonlinear hysteresis parameters as the output. The principal component analysis technique is applied to eliminate the redundant dimensionality of the input data. The attention mechanism is utilized to enhance the generalization ability of the standard network. Three representative hysteretic models are employed to verify the effectiveness of the present method. Both numerical and experimental results demonstrate that the proposed method could yield accurate identification results in all cases, even when uncertain and limited input data are used. Compared with the sensitivity methods and heuristic algorithms, the proposed method is more computationally efficient and can obtain more accurate identification results.

Original languageEnglish
JournalNonlinear Dynamics
DOIs
Publication statusAccepted/In press - 2022

Keywords

  • Attention mechanism
  • LSTM
  • Nonlinear hysteresis parameter identification
  • Principal component analysis
  • Vibration data

ASJC Scopus subject areas

  • Control and Systems Engineering
  • Aerospace Engineering
  • Ocean Engineering
  • Mechanical Engineering
  • Applied Mathematics
  • Electrical and Electronic Engineering

Fingerprint

Dive into the research topics of 'Nonlinear hysteretic parameter identification using an attention-based long short-term memory network and principal component analysis'. Together they form a unique fingerprint.

Cite this