The turbopump of liquid rocket engines is prone to failure under non-stationary conditions characterized by high speed, high temperature gradients, and high pressure. The Vold-Kalman filtering (VKF) method can detect turbopump rotor faults from complex time-varying vibration signals. However, due to the complex vibration transmission path of the turbopump, VKF relies on high sampling rate and high-precision phase information of the carrier of the collected vibration signals, resulting in low fault detection accuracy in practical scenarios where key-phase signal loss and low sampling frequency (one pulse per revolution) occur. Moreover, VKF uses batch optimization methods, which are slow to solve and cannot perform online fault detection on the rocket-borne computer. To address these two issues and achieve real-time diagnosis of turbopump faults in milliseconds, a filtering diagnostic method, the Double Point-wise Vold-Kalman Filter ( DPVKF ), was proposed. DPVKF firstly establishes time-varying linear Gaussian models for the state transition and state observation of each order component. Then, it accurately reconstructs the high-precision transient phase of the corresponding carrier from low-precision speed pulses and vibration signals. Subsequently, under the guidance of the reconstructed phase, DPVKF obtains the optimal linear unbiased estimates of the complex envelopes for each order. Finally, DPVKF extracts the fault characteristics of the turbopump rotor under complex excitation interference. Fault simulation experiments and operational tests of the cryogenic bearings on a certain type of turbopump demonstrate that the proposed method can achieve high real-time performance and high reliability in turbopump rotor fault diagnosis.