Sensorless induction motor speed control for electric vehicles using enhanced hybrid flux estimator with ann-ifoc controller

• Main Author: Sepeeh, Muhamad Syazmie
• Format: Thesis
• Language: English; English; English
• Published: 2022
• Subjects: TA Engineering (General). Civil engineering (General)
• Online Access: http://eprints.uthm.edu.my/8444/1/24p%20MUHAMAD%20SYAZMIE%20SEPEEH.pdf; http://eprints.uthm.edu.my/8444/2/MUHAMAD%20SYAZMIE%20SEPEEH%20COPYRIGHT%20DECLARATION.pdf; http://eprints.uthm.edu.my/8444/3/MUHAMAD%20SYAZMIE%20SEPEEH%20WATERMARK.pdf; http://eprints.uthm.edu.my/8444/

Basically, a speed sensor is used to sense an electric vehicle’s motor speed at the rated value in order to achieve a high tracking accuracy of the speed, but the use of a sensor is costly and it is sensitive to vibrations. Therefore, this project proposed a new mechanism in order to eliminate the speed sensor by adopting an enhanced hybrid flux estimation. The voltage signal was modified using the rotor-flux-oriented current model’s output for the internal stator flux controller to minimise the back-EMF error to represent a sensorless control. Artificial neural network (ANN)-field-oriented control (FOC) was used in the hybrid flux system. The function of the ANN was to improve speed-tracking performance, and the learning rate of the ANN inside the indirect FOC’s structure trained using the Levenberg-Marquardt (LM) algorithm was varied in order to increase speed-tracking accuracy when combined with the improved ANN speed controller. The hyperparameters of ANNs, such as weights and biases, were randomly initialised and updated using the backpropagation (BP) algorithm in order to increase the convolution of the ANNs. The sensorless ANN-IFOC was modelled, simulated, and tested using MATLAB/Simulink for a 20Hp EV motor based on a small Renault Twizy EV model and triggered by the space-vector pulse-width modulation (SVPWM). The results of the ANN-IFOC hybrid estimator were obtained in four cases, which were 1) constant high and low speeds, 2) constant speed against parameter variation, 3) variable speed, and 4) variable load torque disturbances. All results showed that the proposed method gave excellent agreement, as compared with ANN- and PI-based conventional voltage model estimators, with increased tracking accuracy (1500 rpm: 99.23% and 99.60% to 99.85%; 1000 rpm: 98.90% and 99.45% to 99.85%; and 500 rpm: 97.92% and 99.10% to 99.85%). The proposed model with the sensorless speed controller showed consistent tracking accuracy with faster speed responses and gave the shortest settling time and fewer overshoots compared with the existing PI controller. Furthermore, the drive system was able to control and improve the transient response of the EV motor.