In this work we have studied different indicators of muscle fatigue from the electrical signal produced by the muscles when contract (sEMG or EMG: surface electromyography): Mean Frequency of the power spectrum (MNF), Median Frequency (Fmed), Dimitrov Spectral Index (FInsm5), Root Mean Square (RMS), and Zerocrossing (ZC). The most reliable features are selected to develop a detection algorithm that estimates muscle fatigue. The approach used in the algorithm is probabilistic and is based on the technique of Gaussian Mixture Model (GMM). The system is divided into two stages: training and validation. During training, the algorithm learns the distribution of data regarding fatigue evolution; after that, the algorithm is validated with data that have not been used to train. Therefore, two experimental sessions have been performed with 6 healthy subjects for biceps.

Rehabilitation exercises cause fatigue because tasks are repetitive. Therefore, inevitable human motion performance changes occur during the therapy. Although traditionally fatigue is considered an event that occurs in the musculoskeletal level, this paper studies whether fatigue can be regarded as context that influences lower-dimensional motor control organization and coordination at neural level. Non Negative Factorization Matrix (NNFM) and Detrended Fluctuations Analysis (DFA) are the tools used to analyze the changes in the coordination of motor function when someone is affected by fatigue. The study establishes that synergies remain fairly stable with the onset of fatigue, but the fatigue affects the dynamical coordination understood as a cognitive process. These results have been validated with 9 healthy subjects for three representative exercises for upper limb: biceps, triceps and deltoid.