Purpose: The sliding mode observer (SMO) exhibits excellent dynamic performance and robustness, making it widely used in motor control applications. However, traditional SMOs suffer from issues such as chattering and higher-order harmonics, which significantly reduce their observation accuracy. This paper aims to tackle the above drawbacks by proposing a two-stage filtered SMO based on a fourth-order generalized integrator (FOGI). Design/methodology/approach: By replacing the conventional low-pass filter (LPF) in the SMO with a FOGI, higher-order harmonics can be effectively eliminated, thereby considerably improving observation accuracy. Moreover, the FOGI has no phase shift at the central frequency, which eliminates the need for additional phase compensation. Using a two-stage filtering approach, information from the first-stage filter is fed back into the observer, while the second stage filtered information is used as input to a phase-locked loop. Findings: Simulation and experimental results demonstrate that the proposed two-stage FOGI-filtered SMO effectively suppresses chattering and significantly improves position observation accuracy. Originality/value: A FOGI-based two-stage filtered SMO has been developed to improve rotor position and speed estimation in sensorless PMSM control. By replacing the conventional LPF with a FOGI and introducing adaptive center frequency tuning and two-stage filtering, the observer achieves enhanced harmonic suppression, improved dynamic performance and reduced chattering. Simulation and experimental results confirm significant reductions in back-EMF harmonics, electrical angle error and speed estimation error, with faster transient response and halved speed fluctuation.

This study presents an exponential sliding-mode observer with a continuous boundary layer for position-sensorless interior permanent magnet synchronous motor drives. The observer combines linear and fractional-power error feedback with a smooth switching function. The exponential reaching law provides finite-time convergence, while the continuous observer analysis establishes ultimate boundedness and finite-time entry into the prescribed boundary layer under bounded forcing. Compared with a super-twisting sliding-mode observer, the proposed method achieves smaller speed-estimation error, lower current distortion, and reduced calculated torque fluctuations in the reported tests. The speed-reversal experiments further demonstrate accurate rotor-position tracking at commanded speeds of ±500 and ±1000 r/min. Component-ablation results confirm the combined contribution of the linear reaching term and continuous smoothing mechanism to the estimation performance.