With the development of modern industrial systems, the role of bearings has become increasingly important, and their detection and diagnosis have become more meaningful. The traditional fault diagnosis methods for bearings often use vibration analysis and acoustic analysis, which highly rely on additional sensors such as accelerometers, microphones, etc., leading to a series of problems, including but not limited to increasing sensor costs and system complexity. Therefore, the fault diagnosis of bearings without sensors such as based on motor servo signals has become a topic of great concern for scholars. However, due to the weak and difficult to detect bearing faults, it is currently unclear how to characterize the mapping relationship between the electric transmission system and bearing faults. Given this limitation, this paper proposes a model-based servo signal bearing fault diagnosis method. Firstly, a vector control model for permanent magnet synchronous motor (PMSM) and a torsional dynamics model for damaged bearings are constructed. On this basis, the electromechanical coupling mechanism between the two is derived and explained through mathematical formulas, and the results shows that bearing faults would trigger characteristic frequencies in the servo signal spectrum. Then, an experimental platform is built, which can detect the servo signals of servo motors synchronously. After this step, bearings with typical type of faults are tested on this platform. The experimental results further verify the conclusions in mathematical formula derivation. This paper provides a convenient and cost-friendly bearing faults diagnosis method.