Abstract: It is difficult for a single signal feature to fully reflect the running status of the equipment. When multi-domain features are used to characterize the equipment operating status, dimension catastrophe will occur to cause classifier performance degradation and degrading the performance of the state monitoring model as the feature dimension increases. To solve this problem, a new fault diagnosis method based on the linear local tangent space alignment (LLTSA) feature dimensionality reduction and extreme learning machine (ELM) model is proposed. Using LLTSA to extract low-dimensional manifolds from the high-dimensional feature space reduces the dimensionality of signal features, and ensures model classification performance. In this method, firstly, the vibration signal is decomposed by complementary ensemble empirical mode decomposition (CEEMD), and the correlation coefficient and kurtosis criterion are used to select the components to reconstruct the vibration signal after the noise reduction. Then, calculate the multi-domain features of the reconstructed signal, and use LLTSA to perform feature dimension reduction. Finally, the ELM fault diagnosis model is established with its low-dimensional essential features to monitor the equipment operating status. The monitoring experiment of the check valve of the high pressure diaphragm pump shows that the characteristic dimension of the vibration signal is reduced, and the redundancy between the features is reduced, which can improve the accuracy of the fault recognition of the ELM model.