EMG-based neural network estimator for knee joint movement /
In advanced countries, aging of the society with low birth rate is a serious challenge. It is of paramount importance to assist the daily living of the aging persons in order to make them live independently. Osteoarthritis (OA) is one of the common diseases that affect the joint motion of the aging...
Saved in:
| Main Author: | |
|---|---|
| Format: | Thesis |
| Language: | English |
| Published: |
Gombak, Selangor :
Kulliyyah of Engineering, International Islamic University Malaysia,
2016
|
| Subjects: | |
| Online Access: | Click here to view 1st 24 pages of the thesis. Members can view fulltext at the specified PCs in the library. |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | In advanced countries, aging of the society with low birth rate is a serious challenge. It is of paramount importance to assist the daily living of the aging persons in order to make them live independently. Osteoarthritis (OA) is one of the common diseases that affect the joint motion of the aging people. This disease changes the load distribution pattern at the joint thereby alters the joint biomechanics. The surface electromyogram (sEMG) had been widely used as input signal to the wearable robot for assisting the patient's joint motion because it shows an almost one-to-one relationship with corresponding muscle and the dynamic and kinematic aspect of the limbs. However, the interaction between the robot and the user remain a challenge. This can be eliminated by quantifying the amount of assistance needed by the patient. In this study, an estimator was designed to quantify the amount of assistance needed by the patient during normal gait. The EMG signals and the angular displacement of 14 healthy subjects were used for the design of an estimator the knee joint displacement for the patient. A goniometer sensor was designed to sense the angular displacement of the joint during normal gait. Due to fuzziness nature of the EMG signals of the muscles, machine learning algorithm was used as an estimator to estimate the assistance needed based on the EMG and angle signal data. Artificial Neural-Network (ANN) is one of the machines learning (ML) algorithms deployed in the past and depicted good performance when dealing with EMG-based control and estimation. As such ANN was deployed as the estimator for the study. The EMG signals and the angular displacement data were acquired by placing two electrodes each on two muscles (protagonist and antagonist) of the lower limb and the goniometer aligned with the knee to measure the displacement of the knee joint. The subject completes the single giant cycle. The EMG and the angular displacement were recorded online in Simulink pack of MATLAB program. The data were processed offline and the important features were extracted from the raw data. Four times domain (TD) features were deployed namely: root mean square (RMS), mean absolute value (MAV), variance (VAR) and standard deviation (STD) of the raw EMG signal. The four features were used as input and goniometer data as the output to the ANN estimator. The input data were divided into two datasets. In the first dataset, the four features were used separately to train the ANN and data of sub 4 attained the best training performance of 0.817 regressions and a mean square error (MSE) of 0.026. In the dataset 2, the whole features were used as input to the ANN. The ANN was designed with 5 neurons in the hidden layer to avoid overfitting of the data after trying other numbers of neurons. The four features were used together as inputs to the ANN training in the second dataset and attained a performance of 0.822 regressions and MSE value of 0.030. The ANN was designed with 10 neurons in the hidden layer. The network was validated and the average root means square error between the actual and the estimated angle was computed with a lower error of 0.04986 when using dataset 2. The study is limited to offline validation. |
|---|---|
| Physical Description: | xviii, 120 leaves : ill. ; 30cm. |
| Bibliography: | Includes bibliographical references (leaves 97-99). |