MSc. presentation by D. Mårtensson: Kalman-filter design and evaluation for PMSM rotor-temperature estimation
Place: Seminar Room KC 3N27
Contact: bjorn [dot] olofsson [at] control [dot] lth [dot] se
Save event to your calendar
Daniel mårtensson is defending his Master's thesis at the Dept. of Automatic Control.
Where: Seminar Room of Dept. of Automatic Control KC 3N27
When: June 15, 11:00-11:45
Author: Daniel Mårtensson
Title: Kalman-filter design and evaluation for PMSM rotor-temperature estimation
Advisors: Björn Olofsson, Dept. Automatic Control, LTH; Gabriel Turesson, BorgWarner; Peter Jonsson, BorgWarner
Examiner: Anton Cervin, Dept. Automatic Control, LTH
The permanent magnet synchronous motor, PMSM, is an efficient electrical motor that has seen a greater prevalence in the automotive industry from the increasing demand for electrical vehicles. Managing the temperature of the permanent magnet rotor is important to optimize motor utilization and avoid hardware failures. Direct temperature measurements of the moving rotor with a sensor are, however, both difficult and costly and an observer-based approach to estimate the rotor temperature can instead be attained using measured currents, voltages and a model.
The observers were based on a Kalman Filter, KF, and an extended Kalman filter, EKF. Filter designs were implemented that used low-speed estimators that slowly drive the rotor temperature estimate towards the coolant temperature. For circumstances when the inductance accuracy in the model was limited, EKFs with inductance estimation and gain scheduled noise covariance matrices were also evaluated. There were also potential numerical robustness issues, so normalized and rescaled state variable system descriptions were evaluated.
The observability of the system was found to be poor at low rotational speeds. The simulation analysis of the KFs showed a great reduction in rotor temperature estimation error of roughly 40 ◦C when using a low-speed estimator. In circumstances with limited inductance accuracy, the EKF with inductance estimation had a maximum temperature estimation error magnitude of ≈ 2.5 ◦C compared to ≈ 11 ◦C of the KF design. Using a lower sampling frequency with gain scheduling did, however, come at the cost of robustness. Normalizing or rescaling state variables had a visible effect on the noise covariance settings but did not show noticeable improvements of the computational robustness in a simulation environment with high numerical precision.
The thesis was concluded with a brief analysis using measurement from a real but different motor model. The worst case estimation error magnitudes was approximately 12 ◦C for the rotor temperature. The estimation results are very sensitive to model parameter accuracy and more testing has to be conducted using experimental data, but early results show some promise.