Efficient fault diagnosis of traction control system (TCS) has always been a vital technology for high speed train. Transient fault (TF) derives in harsh operation conditions and manufacturing defect, and its timely diagnosis can prevent catastrophic consequences of TCS or provide degradation information of components. However, TF is easily swallowed by background noise, especially at early stage of failure. This problem brings hardship into feature extraction. Aiming at overcoming the noise problem confronted in weak fault feature extraction, the typical TFs of traction control system are described, and feasibility of speed-based fault diagnosis is demonstrated. Then, an optimal fractional-order method is proposed for TF diagnosis, which suppresses background noise and amplifies the faulty part of the signal. Afterwards, kurtosis and fault duration time are applied to locate the fault component. Three fault diagnosis cases are investigated to validate the application of this methodology.