Is this project an undergraduate, graduate, or faculty project?
Undergraduate
Project Type
group
Campus
Daytona Beach
Authors' Class Standing
Carolyn Ascha Richardson, Senior Katherine Clark Tobiah Rosser Francis Genco Hope Lea
Lead Presenter's Name
Carolyn Ascha Richardson
Lead Presenter's College
DB College of Engineering
Faculty Mentor Name
Fan Yang
Abstract
Transradial (below-the-elbow) amputees account for more than half of all upper limb amputations. Myoelectric prosthetic arms, which use electromyography (EMG) sensors on the surface of the forearm to convert electrical signals from residual limb muscles and mimic hand movement, are popular options for these amputees. However, these prostheses are limited in residual muscle detection and movement accuracy. The objective of this project is to affordably manufacture an externally-powered transradial prosthesis prototype through EMG time-versus-voltage readings collected with six Delsys Trigno Avanti and eight Thalmic MYO EMG sensor channels placed along the extensor digitorum, extensor carpi radialis, extensor carpi ulnaris, flexor carpi radialis, brachioradialis, and palmaris longus muscle bellies of users. Using maximal voluntary contraction (MVC) as a motor-control baseline, users are asked to repeatedly perform three hand gestures across timed movement and rest trials: flexions, extensions, and pinch grips. Raw EMG data is then smoothed, preprocessed, and trained for sequential time-series gesture recognition using a one-dimensional temporal residual network convolutional neural network (1-D ResNet TCNN) model. Using Delsys and MYO EMG sensor data, the 1-D ResNet TCNN model reached 97% classification accuracy for distinguishing hand flexions from rest. The following objectives include transferring TCNN model data to a 32-bit Cortex-M7 microcontroller unit and independently manufacturing EMG electrodes to detect muscle movement while users wear the physical prosthesis. From the existing results of TCNN model classification accuracy, it is apparent that utilizing neural network signal processing can provide amputees with intuitive motor control.
Did this research project receive funding support (Spark, SURF, Research Abroad, Student Internal Grants, Collaborative, Climbing, or Ignite Grants) from the Office of Undergraduate Research?
No
Included in
Orthotics and Prosthetics Commons, Software Engineering Commons, Systems Architecture Commons
Myoelectric Transradial Prosthesis Motor Control With Temporal Convolutional Neural Network Signal Processing
Transradial (below-the-elbow) amputees account for more than half of all upper limb amputations. Myoelectric prosthetic arms, which use electromyography (EMG) sensors on the surface of the forearm to convert electrical signals from residual limb muscles and mimic hand movement, are popular options for these amputees. However, these prostheses are limited in residual muscle detection and movement accuracy. The objective of this project is to affordably manufacture an externally-powered transradial prosthesis prototype through EMG time-versus-voltage readings collected with six Delsys Trigno Avanti and eight Thalmic MYO EMG sensor channels placed along the extensor digitorum, extensor carpi radialis, extensor carpi ulnaris, flexor carpi radialis, brachioradialis, and palmaris longus muscle bellies of users. Using maximal voluntary contraction (MVC) as a motor-control baseline, users are asked to repeatedly perform three hand gestures across timed movement and rest trials: flexions, extensions, and pinch grips. Raw EMG data is then smoothed, preprocessed, and trained for sequential time-series gesture recognition using a one-dimensional temporal residual network convolutional neural network (1-D ResNet TCNN) model. Using Delsys and MYO EMG sensor data, the 1-D ResNet TCNN model reached 97% classification accuracy for distinguishing hand flexions from rest. The following objectives include transferring TCNN model data to a 32-bit Cortex-M7 microcontroller unit and independently manufacturing EMG electrodes to detect muscle movement while users wear the physical prosthesis. From the existing results of TCNN model classification accuracy, it is apparent that utilizing neural network signal processing can provide amputees with intuitive motor control.