In train traction converters, the current sensor and the power device IGBT module, as weak links, are subject to frequent faults due to overheating and overvoltage, etc. Therefore, this paper proposes a method for online diagnosis of composite faults by combining data-driven regression and classification algorithms with a single-phase pulsed rectifier as the research object. The method establishes a fault detection predictor model, and sets a threshold for the residual difference between the predicted value and the measured value of the sensor to determine the system operation status in real-time. After the detection of a fault, uses FFT and Relieff algorithms to extract and select the frequency domain features of the fault data, and identifies the fault location and cause through the fault diagnosis classifier model. Based on the characteristics of current sensor faults and IGBT open-circuit faults, a fault diagnosis decision method based on feature superposition is proposed, which can diagnose the current sensor fault mode with 98.5% accuracy within 10 ms; overcome the IGBT similar fault interference, and diagnose single and multiple IGBT open-circuit fault modes with 99.5% accuracy within 15 ms. Online simulation results show that the fault diagnosis method can detect the occurrence of faults at any time and distinguish current sensor faults and IGBT open-circuit faults at different times, making it an accurate and practical method for online rectifier diagnosis.