Abstract: Based on the time domain vibration signals of CNC machine tool bearings, a feature enhancement method based on manifold learning is proposed. The time series of collected signals are reconstructed in phase space, and the information entropy of different sub-manifolds is calculated to construct the representation of the original signal in the feature space. The manifold distance in the feature space is used as a measure of different types of faults in the original signal set. By using the Isometric Feature Mapping (ISOMAP) algorithm, while retaining the information of fault types, the isomorphic low-dimensional manifolds of signals in the feature space are obtained for fault type classification. Through the verification analysis of example data sets, it is shown that the information entropy-ISOMAP transformation can express and enhance the type features of bearing faults in the low-dimensional feature space, and can be effectively applied to diagnose single and compound fault scenarios of CNC machine tool bearings.