A neural network-based model for lower limb continuous estimation against the disturbance of uncertainty. (January 2022)

• Record Type: Journal Article
• Authors: Li, Wanting; Liu, Keping; Sun, Zhongbo; Li, Chunxu; Chai, Yuanyuan; Gu, Jian
• Is Part Of: Biomedical signal processing and control. Volume 71(2022)Part A
• Publication Date: 2022-01
• Subjects: Neural network -- Surface electromyography -- Continuous estimation -- Correlation analysis -- Zeroing neural network; Signal processing -- Periodicals; Biomedical engineering -- Periodicals; Signal Processing, Computer-Assisted -- Periodicals; Image Processing, Computer-Assisted -- Periodicals; Biomedical Engineering -- Periodicals; 610.28
• DOI: 10.1016/j.bspc.2021.103115
• Languages: English
• ISSNs: 1746-8094

Abstract

Highlights: An FWNN-ZNN model is developed to estimate lower limb motor intention. The muscles associated with hip and knee joints are selected via physiology and correlation analysis. The availability and superiority is verified by comparative experiments. Abstract: In this paper, a novel prediction model is proposed to estimate human continuous motion intention using a fuzzy wavelet neural network (FWNN) and a zeroing neural network (ZNN). During walking, seven channel surface electromyography (sEMG) signals and motion data of hip and knee are collected, and two signals are selected and processed from the seven muscles based on physiological and correlation analysis. Then, FWNN is built as an intention recognition model, with sEMG signals as input and physical hip and knee information as output. Meanwhile, ZNN is exploited into the FWNN model, forming a hybrid model to eliminate the prediction errors of the FWNN model. Finally, comparative numerical simulations are established to indicate the validity of the FWNN–ZNN model with root mean square error (RMSE), mean absolute error (MAE) and coefficient of determination ( R 2 ) as evaluation indexes. Results show that the proposed FWNN–ZNN model can more accurately estimate human motion intention, which lays the theoretical foundation for the human–robot interaction of rehabilitation robots.