For the study of speed control precision of permanent magnet synchronous motor (PMSM) on electric locomotives during traction, a mathematical model of PMSM was established in the d-qcoordinate system after considering unknown radially time-varying load torque of wheelset generated by wheel rail contact irregularity and static weight of locomotive body. Nonlinear torque observer (NTO) was designed to estimate the actual value of the time-varying torque with the observation error approached by adaptive fuzzy logic system. According to the Lyapunov stability theory, adaptive fuzzy sliding mode controller with NTO was constructed after the introduction of Nussbaum function in voltage controller of daxis for the investigation of influence of zero crossing of daxis voltage on speed control and torque observation. Theoretical analyses and simulation results demonstrate that the PMSM speed control system on electric locomotives has closed loop bounded error output against zero crossing of daxis voltage and unknown time-varying torque on the railway whose observation error is constricted to 0.