Author Identifier (ORCID)
Seyyed Morteza Ghamari: https://orcid.org/0000-0001-5082-820X
Asma Aziz: https://orcid.org/0000-0003-3538-0536
Abstract
Brushless DC (BLDC) are common in electric cars, industrial automation, and robotics because of their high efficiency, high torque control, and compact size. Nevertheless, strong speed and current regulation is not easily attained because of system variation, load variations and the shortcomings of traditional fixed-gain proportional-integral (PI) controllers. In this paper, a new snake optimization-assisted deep transfer learning-based reinforcement learning (SOA-DTL-RL)-based adaptive cascade PI controller is proposed that combines transfer learning with fast adaptation, Reinforcement learning with real-time optimization, and snake optimization with optimal initial gain selection to guarantee the robust speed and current regulation in BLDC motors. The proposed approach combines transfer learning (TL) to use already trained control knowledge, but the reinforcement learning (RL) is used to dynamically optimize PI parameters to real-time system changes. The use of TL with RL allows the controller to have both the advantage of adapting quickly to new information by using previous knowledge and the advantage of learning in real-time to ensure there is no need to retrain a lot of the controller and makes it more robust in changing environments. One of the most important drawbacks of TL-based controllers is that they rely on clear initial gains and therefore may converge slowly or be unstable. To resolve this, snake optimization algorithm (SOA) is used to set the PI gains optimally in advance, so as to have a well-optimized initial point of real-time adaptation. Moreover, use of a deep neural network (DNN) in the TL-RL model improves generalization, enabling the controller to effectively learn complicated state-action relationships. The proposed cascade SOA-DTL-RL controller will guarantee the fast transient response, enhanced disturbance rejection, and high tracking accuracy under different operating conditions. The efficacy of the framework is proven by hardware-in-the-loop (HIL) real-time testing on the Typhoon HIL 606 platform, which showed great improvements in performance with respect to response time, robustness, and energy efficiency when compared to traditional PI controllers. An HIL experimental setup is used to validate that the proposed controller can be applied in real-time with minimal computational overhead, with robust speed and current regulation of BLDC motors in industrial and automotive applications.
Document Type
Journal Article
Date of Publication
1-1-2026
Volume
19
Issue
1
Publication Title
IET Power Electronics
Publisher
Wiley
School
School of Engineering
RAS ID
88867
Funders
Edith Cowan University
Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 License.
Comments
Ghamari, S., & Aziz, A. (2026). A novel deep transfer learning-based adaptive cascade PI controller enhanced by reinforcement learning algorithm and snake optimization for robust speed regulation of brushless DC motors. IET Power Electronics, 19(1). https://doi.org/10.1049/pel2.70157