sEMG Techniques to Detect and Predict Localised Muscle Fatigue

Recent advances in physiological studies have demonstrated the importance of muscle fatigue detection and prediction in various aspects of our lives, including sports, rehabilitation and ergonomics. Automating muscle fatigue detection/prediction in wearable technology has the potential to aid in many applications. However, current research has made little progress towards automating muscle fatigue detection/prediction in computational models. The work presented in this chapter supports the idea that an automated muscle fatigue detection/prediction system can be used to aid sporting performance and to avoid injury. In support of this view, a wearable system that operates based on the detection and classification of three different stages of muscle fatigue (Non-Fatigue, Transition-to-Fatigue and Fatigue) has been developed. Current research focuses on only two muscle fatigue stages (Non-Fatigue and Fatigue); with this limitation in mind, data was analysed with the aim to develop features that best extract muscle fatigue content, using both statistical models and evolutionary computations tools to help find the number of muscle fatigue stages. This enabled the development of an automatedmuscle fatigue detection system, which provides true prediction capabilities. In doing so, a third stage of fatigue was identified, the so-called Transition-to-Fatigue stage, which occurs before the onset of fatigue. By identifying this transitional fatigue stage, it is possible to predict when fatigue will occur, which provides the foundation of the automated system. To demonstrate the applicability of the Transition-to-Fatigue class, the classification performance of the two class (Non-Fatigue and Fatigue) and three class approaches (Non-Fatigue, Transition-to-Fatigue and Fatigue) were compared. This chapter will include various studies that identify the most suitable methods to apply in the real-time autonomous system. The first section of studies developed various statistical features that best distinguished between the different classes of fatigue, resulting in new combined feature extraction methods called 1D spectro and 1D spectro_std. The second section used evolutionary computation, evolving features and creating pseudo-wavelets improving current state of the art. The various features evolved in this work all produced high classification accuracy from surface electromyography (sEMG) signals emanating from the biceps brachii during both isometric and non-isometric contractions. In the third section, a method to predict the time to fatigue was established using artificial neural network classification based on the three classes of fatigue. This technique was also implemented in 9