Multi-dimensional surface electromyogram signal prosthetic hand control method based on principal component analysis

Patent metadata

Field	Value
Publication number	US-10959863-B2
Application number	US-201816475680-A
Country	US
Kind code	B2
Filing date	May 23, 2018
Priority date	Jun 20, 2017
Publication date	Mar 30, 2021
Grant date	Mar 30, 2021

Abstract

The present invention discloses a multi-dimensional surface electromyogram signal prosthetic hand control method based on principal component analysis. The method comprises the following steps. Wear an armlet provided with a 24-channel array electromyography sensor to a front arm of a subject, and respectively wear five finger joint attitude sensors at a distal phalanx of a thumb and at middle phalanxes of remaining fingers of the subject. Perform independent bending and stretching training on the five fingers of the subject, and meanwhile, collect data of an array electromyography sensor and data of the finger joint attitude sensors. Decouple the data of the array electromyography sensor by principal component analysis to form a finger motion training set. Perform data fitting on the finger motion training set by a neural network method, and construct a finger continuous motion prediction model. Predict a current bending angle of the finger through the finger continuous motion model.

First claim

What is claimed is:

1. A multi-dimensional surface electromyogram signal prosthetic hand control method based on principal component analysis, comprising the following steps:
(1) wearing an armlet provided with an array electromyography sensor to a front arm of a subject, and respectively wearing five finger joint attitude sensors at a distal phalanx of a thumb and at middle phalanxes of remaining fingers of the subject, wherein the array electromyography sensor is a 24-channel array electromyography sensor;
(2) performing independent bending and stretching training on the five fingers of the subject, and meanwhile, collecting data of an array electromyography sensor and data of the finger joint attitude sensors;
(3) decoupling electromyography sensing data by a principal component analysis to form a finger motion training set of the subject, wherein the finger motion training set is represented by a matrix, a number of rows of the matrix is a number of samples, a number of columns of the matrix is a number of channels of the array electromyography sensor, and original 24-dimensional data is reduced to 5-dimensional data by principal component analysis; and removing the sensors worn on the fingers after the training is finished;
(4) performing data fitting on the finger motion training set by a neural network method, and constructing a finger continuous motion prediction model; and
(5) predicting a current bending angle of the finger through the finger continuous motion model in the step (4).

2. The multi-dimensional surface electromyogram signal prosthetic hand control method according to claim 1 , wherein the independent bending and stretching training on the five fingers in the step (2) specifically comprises: repeatedly bending and stretching each finger for ten times, pausing for 30 seconds after completing one round of motions of the five fingers, then performing a second set of motions, two sets of motions being performed in total; collecting and preprocessing the electromyography signals in the training process, and stopping collecting during the pauses; and representing original electromyography data by a muscle activity.

3. The multi-dimensional surface electromyogram signal prosthetic hand control method according to claim 2 , wherein the preprocessing comprises representation and normalization processing on the muscle activity of the electromyography signals, and quaternion solution of attitude data.

4. The multi-dimensional surface electromyogram signal prosthetic hand control method according to claim 1 , wherein in the step (4), a three-layer neural network structure is used, five neurons are arranged in an input layer, 15 neurons are arranged in a hidden layer, and five neurons are arranged in an output layer; transmission functions of the hidden layer and the output layer of the neural network are a Sigmoid function and a linear function respectively; and the finger motion training set collected in the step (3) is used as a sample for error back propagation calculation to solve network parameters thereof.

5. The multi-dimensional surface electromyogram signal prosthetic hand control method according to claim 1 , wherein in the step (5), after the current bending angle of the finger is predicted, a bending angle variation of the finger is converted into an actual control amount of a motor, which specifically comprises the following steps of:
(6.1) designing an underactuated control model for a prosthetic hand finger;
(6.2) calculating a motion equation of an estimated bending angle of the finger and a rotation angle of a stepping motor by analyzing a motion trajectory of the prosthetic hand finger;
(6.3) substituting the estimated bending angle of the finger into the motion equation in the step (6.2) to obtain an output rotation angle of the stepping motor; and
(6.4) controlling the stepping motor to rotate at a corresponding angle through a microcontroller.

Assignees

• Univ Southeast

Inventors

• Song Aiguo
• Hu Xuhui
• Zeng Hong
• Xu Baoguo
• Li Huijun

Classifications

• G06F18/2135

  based on approximation criteria, e.g. principal component analysis · CPC title

• G06F18/24

  Classification techniques · CPC title

• A61F2/72 Primary

  Bioelectric control, e.g. myoelectric · CPC title

• B25J9/161 Primary

  Hardware, e.g. neural networks, fuzzy logic, interfaces, processor · CPC title

• G06N3/084

  Backpropagation, e.g. using gradient descent · CPC title