Hydraulic pump is under long-term high-pressure and high-speed operating conditions, the parts of the pump body are very easy to failure. Vibration signals measured under actual working conditions often contain many irrelevant signal components such as noise, which makes it difficult for traditional methods to accurately classify and identify fault types. Therefore,a hydraulic pump fault pattern recognition method was proposed based on dynamic mode decomposition (DMD) and t-distributed random nearest neighbor embedding (t-SNE) clustering. The vibration signal was obtained by monitoring the sensor arranged on the pump body. DMD was used to decompose the signal to obtain the mode components characterized the essential characteristics of the signal, then t-SNE was used for dimensionality reduction clustering to achieve accurate identification of different fault types.The feasibility and effectiveness of the method proposed are verified by numerical simulation and test bench experiment.