Modelado de las señales mioeléctricas producidas por las extremidades superiores, usando machine learning, para personas con niveles de amputación y/o malformación transradial.

Abstract

Currently, people who have some transradial amputation and/or malformation face mobility challenges through the use of mobile prostheses such as myoelectric ones. However, the precise control of these prostheses remains challenging due to the complexity of interpreting the signals generated in the muscles. Therefore, the present project describes the modeling of myolectric signals generated by the upper limbs during the execution of eight predefined activities to improve the results of the classification for prosthetic control for people with this disability: To do this, an ESP32S development board and three MyoWare sensors integrated into a wristband are used, which are placed on three different muscles: tensor, upper and lower biceps. The acquired myoelectric signal provides values from 0V to 5V, which are stored in CSV files for each channel. This allows for obtaining a database to train and validate the machine learning model using Google Colab. The data is normalized and standardized to train a convolutional neural network (CNN) to classify eight categories. The data was split into 80% for training and 20% for testing. The categorical network reached 100% accuracy in five categories: 98.6%, 98.8%, and 98.5% in the three remaining categories. The temporal prediction model achieves a correlation close to 98% in 95% of the signals concerning the original signal. CNNs identify complex patterns that classify different types of movements and predict their future behavior with high accuracy. Finally, the project highlights the potential of Machine Learning and neural networks in controlling functional prostheses, contributing to the advancement of physical rehabilitation for people with this disability.