The primary function of the ammunition loading manipulator is to extract the projectiles from the magazine and transport them accurately and timely to the handover position of the ammunition coordinating arm.The positioning accuracy of the manipulator significantly impacts the entire ammunition supply process，and improving its reliability is of great importance to the overall automatic loading system.A parameterized co-simulation model that integrates electromechanical coupling is established a specific ammunition loading manipulator.The accuracy of the simulation model is validated using bench test data.A neural network surrogate model is developed to replace the complex dynamic simulation model，enabling the acquisition of response values after the manipulator reaches its target position.Finally，the reliability and sensitivity of the manipulator’s positioning accuracy are analyzed using a hybrid importance sampling method combined with the surrogate model.The results show that the positioning reliability of the manipulator is 0.9779 with high accuracy under the influence of the selected 12 parameters.The sensitivity analysis shows that the friction coefficient between gears has the greatest impact on positioning reliability，which points out the direction for subsequent structural optimization.