A novel muscle fatigue assessment method integrating the Hammerstein model with surface electromyography (sEMG) signals was proposed, to address the effective identification of dynamic upper-limb muscle fatigue during overhead work. An overhead work experiment was designed to acquire sEMG signals from participants during sustained tasks, followed by signal preprocessing. The sEMG median frequency and root mean square were selected as the model input and output, respectively, to construct a nonlinear system based on the Hammerstein structure. Key model parameters were identified using recursive least squares with a forgetting factor combined with singular value decomposition. The K-means++ clustering algorithm was applied to perform unsupervised classification on the identified parameters, enabling differentiation of distinct fatigue levels. The electromyographic fatigue threshold (EMGFT) was further introduced to determine the exact onset time of muscle fatigue by integrating clustering results. Experimental results demonstrated that the method effectively captured the nonlinear dynamic characteristics of upper-limb muscles during overhead work. The clustering outcomes exhibit strong correlation with Borg CR10 subjective fatigue ratings, indicating good physiological interpretability. Combined with EMGFT analysis, the approach accurately characterized the progression of fatigue, providing a quantifiable technical pathway for upper-limb muscle fatigue monitoring in dynamic overhead work.