Low-speed roller bearings are generally in the sheave bearings and turret bearings which are used in the offshore industry under heavy loads. One of the challenges to the integrity of these bearings is the presence of subsurface cracking in the raceways. Downtime cost, which is caused by the failure of these bearings is significant. Based on the potential of the recently-developed monitoring methods for detection of damage in low-speed bearings, this research is focused on the feasibility of the remaining lifetime prediction for roller bearings with existing subsurface cracking in the raceways. The main approach of this thesis comprises a locally-coupled elasto-plastic damage model for low-cycle fatigue. In the numerical simulations, models with and without crack have been implemented and subjected to heavy loading condition. To avoid the crack singularity in the FEM, and reduce the mesh sensitivity, a non-local method has been applied to determine the strain data around the crack tip. In a post-processing part, the nonlinear damage accumulation is implemented with a multi-yield surfaces algorithm. To assess and validate the computational accuracy, the implemented locally-coupled elasto-plastic damage model has been compared with other (published) results. The numerical simulations suggest that the number of cycles to failure is about two times higher when the detectable crack size decreases from 20mm to 5mm. From these results, it is recommended to improve the monitoring system for the detection of smaller cracks. Moreover, further experimental assessment is required to validate and update the estimations in the thesis.