In this paper, we proposed a new method for fault diagnosis of mechanical bearings. The method is based on topological data analysis (TDA) technology, such as persistent homology to analyze time series. We used Case Western Reserve University bearing dataset to conduct experiments for different fault diameters and different load conditions. Through comparison with previous work, the main contribution of this paper is to use topological data analysis to enhance the representativeness and expressiveness of the extracted features. The experimental results under different working conditions to verify the effectiveness and development of the proposed model. We provided a new idea for the field of fault diagnosis.