Abstract: Many transportation equipment such as rail transit vehicles use three-phase asynchronous motors as driving motors, which often work in a variable frequency speed regulation state. In order to detect and locate inter-turn short circuit faults in the speed regulation process, firstly, the zero sequence voltage and three-phase current of the asynchronous motor are measured. Discrete wavelet transform(DWT) are used to decompose and reconstruct the zero sequence voltage signal and three-phase current into different frequency bands, and the root mean square(RMS) of each frequency band is calculated. Secondly, the RMS of each frequency band is taken as the fault feature, a multi classification model is established using error correction output code(ECOC) combined with support vector machine(SVM), and Bayesian optimization method(BO) is used to optimize the hyperparameters of the multi classification model to obtain the optimal model. An experimental platform is designed to validate the proposed method, and the samples obtained are labeled as four categories: no fault, a-phase inter-turn short circuit, b-phase inter-turn short circuit, and c-phase inter-turn short circuit. The classification accuracy of the model established through experimental data on the training and testing sets is 99.83% and 95.31%, respectively. This result proves the effectiveness of the inter-turn short fault detection and localization method for asynchronous motor speed regulation process based on DWT and BO-ECOC-SVMs.