Abstract: To address potential anomalies in the fluid end of three-cylinder reciprocating drilling pumps caused by faults in valve or piston components, this paper proposes an anomaly detection method based on valve vibration phase information. The method leverages changes in valve timing induced by fluid-end abnormalities, using the phase characteristics of valve impacts to identify the internal state of each cylinder. This approach offers strong physical interpretability and high engineering applicability. Specifically, the envelope spectrum of the vibration signal is first analyzed using autocorrelation to extract the valve impact frequency as a basis for feature separation. A homologous multi-source mixture decoupling algorithm is then employed to isolate the impact sequences of each cylinder’s valves. Finally, the time delay between inter-cylinder impact signals is determined via pairwise cross-correlation to construct relative phase features. These features are compared with theoretical phase differences to enable anomaly detection and faulty cylinder localization. The proposed method is experimentally validated on drilling pump platforms, and the results demonstrate its high accuracy and robustness in fluid-end anomaly detection and fault localization. This study provides a simple and reliable technical approach for monitoring the operational status of key components in drilling pumps.