Kalman Filter and model reference adaptive system-based sensorless control of surface-mounted permanent magnet synchronous machine with non-sinusoidal back-electromotive force

• Ano de defesa: 2025
• Autor(a) principal: Santos, Stefan Thiago Cury Alves dos
• Tipo de documento: Tese
• Tipo de acesso: Acesso aberto
• Idioma: eng
• Instituição de defesa: Biblioteca Digitais de Teses e Dissertações da USP
• Palavras-chave em Português: controle sem sensores de posição; Filtro de Kalman; força contraeletromotriz não- senoidal; Kalman Filter; máquina síncrona com ímã permanente; model reference adaptive system; non-sinusoidal back-electromotive force; permanent magnet synchronous machine; sensorless control; sistema adaptativo por modelo de referência
• Link de acesso: https://www.teses.usp.br/teses/disponiveis/18/18153/tde-07052025-155507/

Given the relevance of sensorless control of surface-mounted permanent magnet synchronous machines (SPMSM) and the challenges of estimating their position and speed with non- sinusoidal back-electromotive force (back-EMF), this work aims to study sensorless methods for medium and high speeds and suggests alternatives to the existing techniques in the literature for this type of machine. Two new approaches are proposed: the first is based on the classic Kalman filter and infers the machines rotor position from the back-EMF estimation in the αβ reference frame; and the second, for rotor speed estimation, consists of a new adaptation law, calculated in the dqx reference frame, for reference model adaptive system (MRAS) estimation. Both proposed methods consider the non-sinusoidal characteristic of the back-EMF and are suitable for any SPMSM. The MRAS-based solution is also appropriate for interior permanent magnet synchronous machines. The vector control of the SPMSM is implemented in the dqx reference frame. The experimental results compared the proposed methods, briefly called KF (proposed) and MRAS dqx, with three others already presented in the literature: extended Kalman filter (called EKF) for direct position and speed estimation; classical Kalman filter for back-EMF estimation ignoring back-EMF dynamics (KF (e = 0)); and classical Kalman filter for back-EMF estimation ignoring back-EMF harmonic components other than the fundamental (KF (sinusoidal)). KF (e = 0), KF (sinusoidal), and KF (proposed) concurrently use the proposed MRAS-based method for speed estimation. The five approaches are compared considering transient (step commands for speed variation and reversal) and steady-state responses (speeds from 20 rad/s to 140 rad/s in steps of 20 rad/s, with and without load torque). The minimum operating speed at no load and the execution time for each method are also checked. All the techniques show similar performance about measured and estimated speed dynamics in transient tests, as well as similar minimum operating speeds. However, only EKF and MRAS dqx execute the reversal command and successfully keep the system stable below the minimum operating speed. As for the steady-state tests, KF (proposed) shows excellent position estimation in the speed range between 20 rad/s and 80 rad/s, which results in low torque ripples. MRAS dqx shows excellent results above 100 rad/s, with the lowest position estimation errors and torque ripple. In terms of computational burden, MRAS dqx has the lowest execution time, and KF (proposed) comes in third, with a time just longer than KF (e = 0). Therefore, the experimental results demonstrate the effectiveness and usefulness of the proposed methods in the sensorless control of SPMSM with non-sinusoidal back-EMF, as well as advantages over other existing and consolidated techniques in the literature.