To investigate the changes in upper limb muscle synergy characteristics caused by excavator driving fatigue, a method combining discrete wavelet transform and non-negative matrix factorization (NMF) is proposed for analyzing the upper limb muscle synergy characteristics of drivers. Based on the excavator simulation driving platform, experiments are implemented to collect surface electromyographic (sEMG) signals of 10 upper limb muscles from adult male subjects. The discrete wavelet transform is used to perform multi-scale decomposition of the signals, and soft thresholding is utilized to remove noise and baseline drift. NMF is used to extract muscle synergy elements and weights. Combined with the degree of synergy activation and obvious activation time indicators, the collaborative structure and temporal changes of upper limb muscles before and after fatigue are analyzed. The results show that, under the condition of variance accounted for (VAF) greater than 0.95, the number of synergistic elements was determined to be 2. During the post-fatigue mining stage, synergistic element 1 in the left and right upper limbs changed from brachioradialis and extensor digitorum, and biceps to brachioradialis and extensor digitorum, while synergistic element 2 expanded from triceps and trapezius to biceps, triceps, and trapezius. During the rotation and loading stages, the left upper limb shows the same reorganization pattern, while the right upper limb cooperates to maintain stability. The degree of collaborative activation during the mining phase decreased from 84% and 72% to 81% and 69%, respectively, with an overall decrease of about 3%. The activation levels of collaborative element 1 and collaborative element 2 during the rotation and loading stages decrease from 81% and 76% to 77% and 74%, respectively. The overall collaborative decrease is about 3%, which is smaller than that during the excavation stage. But the significant activation time change rate reaches 17%. The fatigue of excavator operation can cause the reorganization and temporal changes of upper limb muscle synergy, and its adaptation mechanism is stage-dependent. The central nervous system achieves stable control by changing the synergy activation timing and prolonging the activation duration. This study can provide analysis methods and data references for human-machine collaborative control and assisted driving of excavators.