Sensorless Control of SPMSM for Low-Speed Region Based on Physics-Informed Extreme Learning Machine

This paper proposes a sensorless control method for surface-mounted permanent magnet synchronous motors (SPMSMs) operating in the low-speed region, based on a physics-informed extreme learning machine (PI-ELM) framework. Consistent estimation of rotor position and speed is achieved by incorporating intrinsic motor constraints, including the electrical relationship and the Pythagorean identity, into the ELM training process. Accurate estimation with a reduced input set is obtained as these constraints are embedded during training. This greatly reduces the computational burden and data requirements, allowing for real-time implementation on low-cost digital signal processors (DSPs). Rotor position and speed are extracted from the estimated sine and cosine components through a lightweight phase-locked loop (PLL), and low-speed offset drift is periodically suppressed by a correction mechanism applied to the PLL output. Experimental results on an SPMSM test platform have validated the effectiveness of the proposed approach, demonstrating accurate and stable sensorless operation down to 0.01 p.u. and a memory requirement of less than 1.1 kB.