Redes neurais artificiais para o controle da mancha produzida por estimilação eletrica neuromuscular

AUTOR(ES)
DATA DE PUBLICAÇÃO

1996

RESUMO

This work consisted in the application of artificial neural network algorithms for control of gait generated by NeuroMuscular Electrical Stimulation (EENM) in spinal cord injured subjects. Two independent approaches were attempted -namely, a theoretical, and an experimental approach. ln the theoretical study, a three-Iayered neural network was set up to modei the segmental sensorimotor transformations involved in leveI walking. Network inputs were proportional to hip, knee, and ankle flexion/extension angles, and to vertical ground reaction forces. The output signal was proportional to the activity leveI of five muscles essential to gait production as seen from the sagittal plane. Results fram simulations for various network structures indicate that motion control improves when the latest output is used as part of the input vector. AIso, it appears that motion control is better when done with two independent networks (one for the stance phase of gait and another for the swing phase) than with a single network for the entire gait cycle. Further, inclusion of explicit recursion in the learning algorithm, recall of previous inputs, hidden layer intraconnectivity, selective hidden layer to output layer connectivity, and the use of signal derivatives all resulted in r;o improvement in the simulated gait controI. ln the experimental approach, a three-Iayered artificial neural network was used for adaptive control of the swing phase only. Network inputs consisted of knee and ankle goniometer signals for System 1, and knee and hip angular data for System 2. Controller output was proportional to changes in applied NMES pulse idth (PW). Stimulation was applied to the left femoral and common peroneal nerves. The neural networks were trained off-line and on-line. Network performance was assessed by applying a number of different stimulation PWs and later comparing the resulting motion to a sample good step observed during the same test session. On-line training consisted of negative and positive re-enforcement applied at chosen times. Both on-line and off-line training algorithms included an enhanced back-propagation scheme. Performance evaluation results favour the use of System I over System 2. AIso, nerwork performance during on-line learning is better than that of a network submitted to off-line learning only. Then, according to some of the results, a prototype based on System I was set up and used for generation of the entire gait cycle in one spinal subject

ASSUNTO(S)

redes neurais (computação) membros inferiores controle automatico estimulação neural

ACESSO AO ARTIGO

Documentos Relacionados