Objective To investigate the relationship between upper limb joint angles and muscle fatigue in overhead tasks and develop a muscle fatigue prediction model based on residual neural networks (ResNet). Methods Through the simulation of drilling experiments performed with different working postures and on different operating surfaces, the maximum voluntary contraction, maximum endurance time, maximum residual muscle force, and subjective fatigue ratings were measured. The collected data were processed and used as input for the ResNet prediction model, which was constructed to predict muscle fatigue levels. Results The ResNet model exhibited outstanding predictive accuracy, with a root mean square error (RMSE) of 0.028. Compared with traditional backpropagation neural networks (RMSE=0.053) and multilayer perceptron neural networks (RMSE=0.059), they displayed smaller errors and better fitting. Conclusions The proposed residual neural network muscle fatigue prediction model can effectively and accurately predict muscle fatigue, providing strong support for improving work efficiency and reducing the risk of work-related musculoskeletal disorders.