Unsupervised Muscle Fatigue Clustering Based on sEMG Signals using DeepCNN Autoencoders

Muscle fatigue assessment is key for optimizing physiotherapy, rehabilitation, and exercise. Muscle fatigue detection not only helps in reducing rehabilitation discomfort but also assists athletes in working out efficiently. The main motivation behind this research is to address the pressing need for a precise and reliable tool to assess muscle fatigue during physical activities. In addition to enhancing therapeutic interventions, this approach improves microtears-induced muscle strengthening by advancing the understanding of muscle fatigue levels. Previous studies have mainly employed supervised approaches to detect fatigue by processing surface electromyography (sEMG) signals through traditional methods in the Time, Frequency, and Time-Frequency domains. These approaches are vulnerable to challenges, such as signal sensitivity, noise interference, and inter-individual variability. Thus, the proposed method uses the spring energy model of the sEMG signal because of its excellent approximation with muscle characteristics. In order to handle noisy and ambiguous data, this paper employs a self-supervised Deep Convolutional Neural Network (DeepCNN) Autoencoder for feature extraction and K-means clustering due to their strong performance. Five healthy participants carried out two experiments using a knee rehabilitation robot (FUM-Physio). The results showcase a well-ordered clustering that closely mirrors the progression of muscle fatigue during these experiments. This indicates the robustness and generality of the proposed method.