YZYafei Zhang
Yafei Zhang
*This is an algorithmically generated author record based on publicly available information.
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Yafei Zhang has a total of 4 co-authors. Here are the top co-authors by number of shared publications:
- Huaiyuan Guo
School of Automation and Intelligent Sensing, Shanghai Jiao Tong University, Shanghai 201100, China
1 shared publication - Junlong Meng
School of Automation and Intelligent Sensing, Shanghai Jiao Tong University, Shanghai 201100, China
1 shared publication - Ming Yang
Department of Instrument Science and Engineering, Shanghai Jiao Tong University, Shanghai, China
1 shared publication - Runyu Ni
School of Automation and Intelligent Sensing, Shanghai Jiao Tong University, Shanghai 201100, China
1 shared publication
Latest Publications
- Open Access
A Physics-Guided Lightweight Speed Estimator for Embedded Sensorless Control of Ultrasonic Motors
Accurate sensorless speed feedback is essential for compact ultrasonic motor (USM) systems, but physics-based estimators may become inaccurate under nonlinear friction and changing operating conditions, and high-capacity data-driven models are often too computationally demanding for embedded controllers. This paper proposes a Physics-Guided Explicit Knowledge Distillation (Phy-EKD) framework for sensorless speed estimation and closed-loop control. Contact–friction and Butterworth–Van Dyke equivalent-circuit models provide an interpretable speed baseline from measurable electrical signals. An offline teacher learns feature and temporal dependencies, which are aggregated into calibrated fixed priors to guide a lightweight student in producing a relative correction and an additive residual for the baseline. Progressive Annealing Curriculum Learning (PACL) gradually shifts training from relative-correction and additive-residual knowledge transfer to end-to-end speed supervision. The held-out offline evaluation shows that the student achieves a root-mean-square error of and against encoder measurements. Independent encoder-feedback and sensorless-feedback experiments quantify tracking error and transient response. The student uses fewer parameters than the teacher. Its complete acquisition-and-processing time on an STM32F767 is approximately within a control interval, supporting embedded sensorless speed feedback under the tested conditions.Open AccessA Physics-Guided Lightweight Speed Estimator for Embedded Sensorless Control of Ultrasonic Motors
Accurate sensorless speed feedback is essential for compact ultrasonic motor (USM) systems, but physics-based estimators may become inaccurate under nonlinear friction and changing operating conditions, and high-capacity data-driven models are often too computationally demanding for embedded controllers. This paper proposes a Physics-Guided Explicit Knowledge Distillation (Phy-EKD) framework for sensorless speed estimation and closed-loop control. Contact–friction and Butterworth–Van Dyke equivalent-circuit models provide an interpretable speed baseline from measurable electrical signals. An offline teacher learns feature and temporal dependencies, which are aggregated into calibrated fixed priors to guide a lightweight student in producing a relative correction and an additive residual for the baseline. Progressive Annealing Curriculum Learning (PACL) gradually shifts training from relative-correction and additive-residual knowledge transfer to end-to-end speed supervision. The held-out offline evaluation shows that the student achieves a root-mean-square error of and against encoder measurements. Independent encoder-feedback and sensorless-feedback experiments quantify tracking error and transient response. The student uses fewer parameters than the teacher. Its complete acquisition-and-processing time on an STM32F767 is approximately within a control interval, supporting embedded sensorless speed feedback under the tested conditions.