This study investigates the broken rotor bar issue in variable-frequency traction motors of urban rail transit trains and proposes an online diagnostic method. Based on rotating coordinate transformation and complex frequency-domain signal analysis theory, the fault characteristics of broken rotor bars expressed through current vector modulus spectrum were derived. Considering train control characteristics, it was identified that the transition phase from constant torque to constant power control is optimal for extracting broken bar defect features. Accounting for elevation differences between adjacent stations and gradient variations along operating sections, the concept of "homogeneous operating conditions" for traction motor faults was introduced, emphasizing the necessity of conducting fault diagnosis under such conditions. Building upon these theoretical derivations and practical operational analyses, an online diagnostic method was developed. Simulations demonstrated that the amplitude of broken bar characteristic harmonics in the current vector modulus spectrum increases proportionally with fault progression. Field data analysis confirmed that the characteristic frequencies of broken bars in each motor can be consistently identified during control mode transitions within sections of similar gradients. The research validates that the proposed method effectively diagnoses broken rotor bar faults, offering a viable technical solution for proactive maintenance strategies and operational safety assurance of traction motors in rail transit systems.