利用伺服馬達慣量估測進行控制器性能改善

Abstract

本論文針對應用編碼器角度感測器進行永磁馬達磁場導向速度控制以及擾動解耦控制策略，提出系統轉動慣量的在線估測方法，其應用是在工業控制器或是工廠自動搬運車中，確保能將精確的慣量用作速度控制器的設計以及進行系統負載轉矩的準確估測，以改善速度控制的動態響應。 本論文提出的在線慣量估測架構整合無跡卡爾曼濾波器(Unscented Kalman Filter-UKF)與變遺忘因子最小遞歸二乘法(Variable Forgetting Factor Recursive Least Square-VFFRLS)，先透過UKF對馬達的轉速以及負載轉矩進行估測，估測的轉速以及負載轉矩將用於VFFRLS的算法當中進行慣量的估測，能有效解決RLS算法容易受到輸入訊號雜訊干擾的問題。此外，估測的慣量會代回UKF的馬達模型中，藉此降低UKF模型中慣量的誤差，以得到準確的狀態估測結果。 最後本論文針對一顆470W用於自動搬運車應用的永磁馬達進行實驗測試，根據實驗結果，整體慣量估測的誤差與實際慣量相比能被限縮在2%以內，而在0到100%額定轉速的梯形波速度命令下，自適應控制下的響應上升時間為550ms、安定時間為890ms，與傳統固定增益的速度控制器比較，暫態上升時間減少5.6%、安定時間減少11.4%，能有效改善速度控制的動態響應性能。

This thesis proposes an online inertia estimation method specifically tailored to the application of encoder angle sensors for magnetic field-oriented velocity control and disturbance decoupling strategy in permanent magnet motors. The application of this method is intended for industrial controllers or factory automated guided vehicles(AGV), aiming to ensure precise estimated inertia are used for speed controller design and accurate estimation of load torque. This, in turn, improves the dynamic response of the velocity control. The online inertia estimation framework presented in this thesis integrates the Unscented Kalman Filter (UKF) and the Variable Forgetting Factor Recursive Least Square (VFFRLS) algorithms. The UKF is employed to estimate the motor's speed and load torque, and the estimated speed and load torque are then used in the VFFRLS algorithm to perform inertia estimation, effectively addressing the issue of RLS algorithm susceptibility to input signal noise interference. Furthermore, the estimated inertia is fed back into the UKF motor model to reduce the inertia estimation errors, resulting in more accurate state estimation outcomes. Finally, a 470W motor used in AGV applications is tested. According to the experimental results, the overall inertia estimation error can be constrained within 2%. Under a trapezoidal speed command ranging from 0% to 100% of the rated speed. The rising time under adaptive control is 550ms, and the settling time is 890ms, When compared to the conventional fixed-gain speed controller, the transient rising time is reduced by 5.6% and the settling time is reduced by 11.4%. These findings suggest that our proposed method can effectively optimize the dynamic response performance of speed control.