A self-sensing data augmentation framework for a PMSM-driven ball screw system

The permanent magnet synchronous motor (PMSM)-driven ball screw system plays an important role in modern manufacturing, directly affecting positioning performance and manufacturing efficiency. Thus, extensive research has been conducted to monitor the health condition of ball screw systems using data-driven approaches. However, in industrial applications, it is challenging to measure sufficient fault signals. Meanwhile, the linear motion of ball screw systems introduces distinct fault characteristics and uncertainties compared to rotating machinery. To address these issues, a self-sensing data augmentation framework is proposed for PMSM-driven ball screw systems. Firstly, the self-sensing motor driver is employed as a measurement unit to collect q -axis current signals of the ball screw system, and the spatial distribution characteristics of this system in the spatial domain are introduced. Secondly, based on dynamics and fault mechanisms, a novel phenomenological decomposition method is proposed to observe the intrinsic physical sub-signals within the q -axis current. Then, considering the uncertainty in fault feature distribution along the ball screw travel, a spatial-probabilistic reconstruction method is proposed to generate fault sub-signals across different speeds. Comparative experiments validate the proposed framework under speed interpolation and one-sided extrapolation conditions in the one-shot scenario, as well as its zero-shot capability achieved with prior knowledge of the amplitude-frequency response. These results validate the proposed framework as an effective solution for sensor-free monitoring of ball screw systems under varying operational speeds.