Machine learning Electromyography signals to control Prosthetic Hand

Project Title

Machine learning Electromyography signals to control Prosthetic Hand

Project Area of Specialization Biomedical EngineeringProject Summary

A prosthesis or prosthetic is an artificial device that is implanted on an amputee person, after any congenital disorder or any injury. The Prosthetic works like the missing part of the body. In our project, we are designing a machine learning-based Prosthetic Hand model that uses classified EMG signals. In this era, the demand for prosthetics has increased from disabled patients, soldiers, or disabled children by birth. The available robotic hands/prosthesis only provide up to three degrees of freedom. And only a limited set of simplified movements for e.g. opening and closing can be achieved. Our aim is to design a Robotic Hand that provides a wide variety of finger, hand, and wrist (up to 52) movements. The data of movements will be taken from the NINAPRO database and will be implemented in MATLAB for execution in the Robotic hand. This project is based on advanced Robotic hand and their control via a Machine Learning Model. The data for the Machine learning model is extracted from NINAPRO(Non-Invasive Adaptive Prosthesis) database. This database constitutes of data from 27 subjects performing upper limb movements. There are a total of 52 movements for each subject. Each movement consists of about 300ms of rest and about 500ms in state of motion (performing the specified movement). The data sets constitute data at intervals of 1milliseconds. Whereas each movement is repeated 10 times for each subject. This data is pre-processed in MATLAB to extract various Time domain and frequency domain features, after processing the data via mathematical techniques it is then fed into a Machine learning model, SVM (Support vector machine). SVM model then generalizes data for all the movements. This model is connected to a Prosthetic Hand via a controller. Since the EMG has a variation among different humans, so when the model receives a different EMG signal it generalizes that data compares it with the pre-processed data and produces the required output on the prosthetic hand.v

Project Objectives

The main objectives of our project are:

1) Classification of EMG signals via Supervised Machine Learning

2) Design Prosthetic hand

3) Control Prosthetic hand via Machine Learning Model

Project Implementation Method

First, the data extracted from the NINAPRO database is cleaned. Then different features are extracted from the data for each row in MATLAB. The data constitutes of EMG signals and glove data for all the subjects. The features are then manipulated using Probability and statistics. The methods used are the Least square method, Linear Regression model, ReLU model. The processed data is divided into training data and test data. Then the data is implemented into a Machine Learning Model known as the SVM (Support Vector Machine). First, the model is trained on the training data. 80% of the data available is used as training data. After the model is trained it is tested for accuracy on the test data. Test data is obtained from the remaining 20% of our processed data. Our aim is to achieve at least 80% accuracy on the test data. The prosthetic hand is designed in Blender software and is printed via a 3D printer. Then the motion control is implemented using servomotors. Finally, the designed prosthetic hand is connected via a controller to the PC which constitutes the Machine Learning model. To verify the model, a subject wears a glove(cyber globe) which is connected to the machine learning model and the prosthetic hand replicates the movement performed by the subject.

Benefits of the Project

There are many advantages but most important and highlighted one Is that person will be able to do their own work by prosthetic which gives them a more realistic approach. The advantage of this technique is that the signal will be acquired from the patient's body and after suitable processing, it is used as a control input to drive motors that are coupled to the prosthetic hand. So, the hand can be worn by amputees and the control mechanism will be initiated by their own EMG signals. The present prosthetics provide a large amount of learning for the amputee in or order to be used and only provide a few basic movements for example opening and closing of a hand. Our model is based on Machine learning and hence can be easily worn and used by the amputee providing a wide variety of movements. To use our model the user doesn’t need a large amount of training and the model can work deliberately on any input since it is designed to work on unseen data via Machine learning.

Technical Details of Final Deliverable

The final project is a prosthetic hand coupled with a machine learning model, that makes it applicable for all the users. Our device is designed to help those with full or partial hand loss retain the function and appearance of a regular hand. The available prosthetic hands are cosmetic replicas with only basic functions. Whereas our prosthetic hand is innovative and can perform 52 movements with a greater degree of movement providing flexibility and dexterity to the user.

Final Deliverable of the Project HW/SW integrated systemCore Industry HealthOther Industries Others Core Technology RoboticsOther Technologies Artificial Intelligence(AI), 3D/4D PrintingSustainable Development Goals Good Health and Well-Being for PeopleRequired Resources

Item Name	Type	No. of Units	Per Unit Cost (in Rs)	Total (in Rs)
			Total in (Rs)	80000
3D- printed hand / prosthetic hand	Equipment	1	50000	50000
Servo motors , actuator and stepper motors	Equipment	10	1000	10000
Embedded system	Equipment	1	10000	10000
Thesis/transport/stating	Miscellaneous	5	2000	10000