FCM-RBFN Integration Technique For Improving Isotonic Muscular Endurance Load Prediction
In sports training, muscle endurance training using surface electromyography (sEMG) analysis is manually monitored by human coach. Decisions rely very much on experience. Hence, the endurance training plan for an athlete needs to be individually designed by an experienced coach. The pre-designed tra...
Saved in:
| Main Author: | |
|---|---|
| Format: | Thesis |
| Language: | English English |
| Published: |
2020
|
| Subjects: | |
| Online Access: | http://eprints.utem.edu.my/id/eprint/25381/1/FCM-RBFN%20Integration%20Technique%20For%20Improving%20Isotonic%20Muscular%20Endurance%20Load%20Prediction.pdf http://eprints.utem.edu.my/id/eprint/25381/2/FCM-RBFN%20Integration%20Technique%20For%20Improving%20Isotonic%20Muscular%20Endurance%20Load%20Prediction.pdf |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | In sports training, muscle endurance training using surface electromyography (sEMG) analysis is manually monitored by human coach. Decisions rely very much on experience. Hence, the endurance training plan for an athlete needs to be individually designed by an experienced coach. The pre-designed training plan suits the athlete fitness state in general, but not in real time. Real-time muscle fatigue monitoring and feedback helps in understanding every fitness states throughout the training to optimise muscle performance. This can be realized with muscle fatigue prediction using computational modelling. This research proposed an integrated Fuzzy C-Means and Radial Basis Function Network (FCM-RBFN) technique to model the relationship between muscle loads versus the muscle fatigue using the sEMG signals. The Fuzzy CMeans techniques aims to cluster similar sEMG signal patterns into three separate groups based on muscle strength level, to facilitate the Radial basis function network in future muscle load prediction. The scope of the research limits the non-invasive EMG acquisition to only the isotonic arm lifting task, involving four electrodes on biceps brachii and flexor carpi radialis muscles group. Three sessions of training data, each with a gap of at least three days‟ rest, were acquired from a group of volunteer undergraduate athletes. The research follows the experimental research methodology, including problem investigation, experimental paradigm design, signal pre-processing analysis, feature extraction, model construction, and performance validation. Due to the higher amount of motion artefact, research in isotonic muscle fatigue prediction is very much lesser than the isometric prediction. Hence, the Butterworth high-pass noise filter on isotonic muscle fatigue data were studied using three cut-off thresholds, 5 Hz, 10 Hz, and 20 Hz. The best prediction performance was achieved by the 10 Hz filter with 0.028 average mean square errors. A total of seven popular feature extraction methods, namely, the mean absolute value, the root mean square, the variance of EMG, the standard deviation, the zero crossing, the median frequency, and the mean were explored to construct the predictive feature vectors. The mean square error was used to benchmark the experimental results with the Artificial Neural Network. The experimental result shows that the proposed FCM-RBFN technique is able to predict different load intensity efficiently according to real time muscle condition against fatigue. The experimental findings suggest that a long isotonic training task induces fatigue, hence it contributes to data noise that will affect muscle load prediction in overall. Therefore, training load should be reduced on the first detection of muscle fatigue sEMG signal, in order to prolong the muscle resistance against fatigue. Future research should study on dynamic cluster number instead of the fixed cluster initialization in FCM technique. Also, the proposed model should be validated using multiple sessions in different periods of time length to further support the hypothesis of muscle endurance. |
|---|