As a critical component of rotating machinery, the degradation indicator (DI) and remaining useful life (RUL) prediction of bearings are crucial for the safe operation of equipment. This paper proposes a physics-informed method for constructing bearing degradation indicators and predicting RUL. By embedding the physical laws governing the bearing's service process, the method enhances the physical interpretability of the degradation indicator and the robustness of the prediction model. In the construction of the degradation indicator, an exponential physical equation describing the evolution of bearing crack length over time is established based on Paris' law. A loss function incorporating physical constraints is designed to compel the deep learning network to extract features that conform to the material damage accumulation law. In the RUL prediction stage, a time-series constraint loss function is introduced to ensure that the prediction results align with the physical law of "remaining life monotonically decreasing with time," mitigating the black-box characteristics of data-driven models. Experiments utilize the PHM2012 dataset, constructing a physics-informed degradation indicator network and a physics-informed remaining useful life prediction network. The results demonstrate that the degradation indicator constructed by the proposed method outperforms traditional statistical features in terms of trend and monotonicity, and the accuracy of the prediction results is significantly improved.