sEMG based Trajectory Control of Artificial Hand Mechanism using ANN and SMC
No Thumbnail Available
Date
2023-02
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Addis Ababa University
Abstract
In this thesis, an approach for controlling an artificial hand mechanism using surface
Electromyography (sEMG) signal, artificial neural network(ANN) and sliding
mode control (SMC) is presented. An artificial hand mechanism also known as
a prosthetic hand or bionic hand, is device designed to replace the function of a
missing or non-functioning hand. The 3D model of five finger robotic hand was
designed using SolidWorks software and then exported to MATLAB/Simulink.
This model was then used to extract the kinematic and dynamic properties of
the robotic hand for which controller design can be performed and tested. The
kinematic model was used to calculate the position and orientation of the hand,
while the dynamic model was used to determine the forces and torques required
to move the hand fingers to achieve the desired gesture.
To use this robotic hand as a replacement for an actual hand lost due to a
variety of reasons, integration of this electromechanical system and human biological
system is required. To this end, surface Electromyography (sEMG) was
used to detect hand muscle activities which were interpreted into six specific hand
movements (gestures). By sensing the activities of two muscle groups of the user’s
arm, the fingers in the robotic hand was controlled to follow specific trajectories.
The classification of the sEMG signals into one of the six gestures was done using
Long Short-Term Memory (LSTM) neural network that was trained from a dataset
collected from five people. To improve the performance of the LSTM during classification,
feature extraction operation of the sEMG signal was performed during
training and classification. Feature extraction methods are applied to identify
relevant patterns and characteristics from the raw sEMG signals. These characteristics
can subsequently be used to categorize and distinguish between various
hand movements, enabling the control of the hand using sEMG signals. As a result,
the neural network’s accuracy rose from 58.64% when utilizing raw data to
99.3% when using extracted features.
Once the required hand gesture was identified from the sEMG signal, interpretation
of the gesture into individual fingers joint angles was done using cubicpolynomial
path-planning algorithm. These joint angle trajectories were used to
command the robotic hand using Sliding-Mode Control (SMC) controller. The
reason behind using SMC was its robustness in handling external disturbance,
parameter uncertainty and unmodelled dynamics, which are inherent in robotic
systems. To test the performance of the controller, parametric uncertainty was
added to the system dynamics during simulation. The results show that the SMC
controller effectively tracked the desired trajectory in the presence of ±30% link
mass variation from their nominal values.