The early weak fault diagnosis of check valve can prevent economic loss and safety accident caused by the failure of the high pressure diaphragm pump due to the wear and breakdown of check valve. Aiming at the problem that early weak fault feature of the check valve of the reciprocating high pressure diaphragm pump are not obvious and are disturbed by a lot of noise, Method: a method for early weak fault diagnosis of check valve based on composite multiscale fluctuation dispersion entropy (CMFDE) is proposed. First of all, replace the normal cumulative distribution function mapping in the CMFDE method with tangent sigmoid to improve the noise resistance of CMFDE. Then, calculate the composite multiscale fluctuation dispersion entropy of the vibration signal, and construct the feature matrix and input it into the support vector machine (SVM) classifier for fault diagnosis. Finally, the actual engineering data of the check valve is used to verify the effectiveness of the method, and a comparative experiment is carried out. The experimental results and comparative analysis show that it is not necessary to reduce the noise of the original signal of check valve, which simplifies the diagnosis process. The composite multiscale fluctuation dispersion entropy feature can accurately reflect the different signal characteristics of check valve, which improves the identification rate of the early weak fault diagnosis of check valve, and the fault diagnosis result is less affected by classifiers, with the identification accuracy reaching 96. 667%.