Desenvolvimento de uma prótese mioelétrica utilizando controle inteligente

Abstract

In recent years, the number of EMG-based care technologies for people with amputated limbs (upper and lower) has developed widely, but such technologies, such as prostheses, have several limitations, either by even at high cost. This work presents the development of an electromyographic (EMG) acquisition, processing and classification methodology for two channels using the AD620 instrumentation amplifier. Obtaining the EMG signal was performed for a healthy volunteer using non-invasive Ag-AgCl surface electrodes positioned in the muscles of the left forearm to capture the signals of four distinct movements: open hand, contraction hand, contraction claw and thumb contraction. These signs have as final objective the activation of a myoelectric prosthesis manufactured with the aid of a 3D printer. After obtaining these signals, a processing methodology was developed that allowed the extraction of seven statistical variables to each window, where it was composed of 100 samples of the EMG signal and obtained every 0.1 second, with an overlap of 50%. The processed signals were classified with the aid of Artificial Neural Networks, specifically the multilayer type. The means of adjustment in relation to the movements were: 100% for open hand, 98.99% for hand contraction, 99.99% for thumb contraction and 98.74% for claw contraction, resulting in an average accuracy of 99.43 %. Therefore, this approach represents a significant step in the development of more intuitive and low cost 3D printed myoelectric prostheses, with the possible extension to the control of several auxiliary devices.