High-Bandwidth Iterative Learning-Based Torque-Sensorless Ripple Decomposition and Feedforward Compensation for PMSMs

Torque ripple in permanent magnet synchronous motors is a critical limitation in high-precision servo applications. This study develops a torque-sensorless ripple minimization framework for arbitrary constant-velocity operations. First, by using high-bandwidth iterative learning control, reliable torque ripple minimization signals are collected over a wide velocity range without external torque sensors or accurate motor parameters. Then, from the collected data, an identification framework for two rotor-angle-dependent maps is developed: a current-dependent torque ripple map and a current-independent cogging torque map. Finally, these maps are combined to generate a feedforward torque compensation signal during actual operation. By explicitly decomposing current-dependent torque ripple components from current-independent cogging components, the torque ripple characteristics are addressed comprehensively compared to conventional torque-sensorless approaches. Experimental results demonstrate effective torque ripple minimization across a wide range of operating speeds, confirming the practicality of the developed framework.